Pose estimation for health data analysis: advancing AI in neuroscience and psychology

Introduction: The integration of artificial intelligence (AI) with health data analysis offers unprecedented opportunities to advance research in neuroscience and psychology, particularly in extracting meaningful patterns from complex, heterogeneous, and high-dimensional datasets. Traditional methods often struggle with the dynamic and multi-modal nature of health data, which includes electronic health records, wearable sensor data, and imaging modalities. These methods face challenges in scalability, interpretability, and their ability to incorporate domain-specific knowledge into analytical pipelines, limiting their utility in practical applications.

Methods: To address these gaps, we propose a novel approach combining the Dynamic Medical Graph Framework (DMGF) and the Attention-Guided Optimization Strategy (AGOS). DMGF leverages graph-based representations to capture the temporal and structural relationships within health datasets, enabling robust modeling of disease progression and patient interactions. The framework integrates multi-modal data sources and applies temporal graph convolutional networks, ensuring both scalability and adaptability to diverse tasks. AGOS complements this by embedding domain-specific constraints and employing attention mechanisms to prioritize critical features, ensuring clinically interpretable and ethically aligned decisions.

Results and discussion: Together, these innovations provide a unified, scalable, and interpretable pipeline for tasks such as disease prediction, treatment optimization, and public health monitoring. Empirical evaluations demonstrate superior performance over existing methods, with enhanced interpretability and alignment with clinical principles. This work represents a step forward in leveraging AI to address the complexities of health data in neuroscience and psychology, advancing both research and clinical applications.

1 Introduction

The intersection of pose estimation and health data analysis represents a burgeoning area of research in artificial intelligence, with profound implications for neuroscience and psychology (1). This task involves analyzing human body movements and postures to infer health-related insights, including motor disorders, cognitive impairments, and emotional states (2). The significance of this research lies not only in its ability to enhance diagnostics but also in enabling remote and continuous monitoring of patients. Moreover, pose estimation contributes to personalized treatments by providing objective (3), fine-grained measurements of movement and posture dynamics. Given the rapid developments in AI and its applications in neuroscience and psychology, there is a growing need for accurate and efficient pose estimation techniques tailored to health data analysis (4). As such, the study of pose estimation transcends traditional computational tasks, offering novel opportunities to deepen our understanding of the human mind and body (5).

To address the limitations of early methods in pose estimation for health data analysis, researchers initially focused on symbolic AI and knowledge representation approaches (6). These traditional methods relied heavily on handcrafted rules and logic-based systems to represent human poses and movement patterns. Knowledge-based systems were integrated with biomechanical models to infer health-related insights, particularly for tasks such as gait analysis and posture classification (7). For example, systems were designed to use predefined skeletal models to analyze deviations in walking patterns, aiding in the diagnosis of neurological conditions such as Parkinson's disease. However, these methods faced significant challenges, such as the inability to generalize across diverse populations and the high dependency on expert knowledge to define rules (8). Symbolic approaches struggled to handle noisy and incomplete data, which is often inherent in health-related applications. While foundational in establishing the field, these methods ultimately lacked the adaptability and scalability necessary for broader health data applications (9).

With the rise of machine learning, data-driven approaches emerged to overcome the limitations of symbolic methods. By leveraging annotated datasets of human movements, machine learning models were trained to detect and classify poses (10), enabling more flexible and scalable solutions. Algorithms such as Support Vector Machines (SVMs) and Random Forests were applied to tasks like fall detection in elderly individuals or recognizing emotional states through postural cues (11). These methods significantly improved pose estimation accuracy by learning patterns directly from data rather than relying on predefined rules (12). However, their performance was heavily reliant on the quality and quantity of labeled training data, which is often scarce in health-related domains due to privacy concerns and data collection constraints (13). Traditional machine learning methods struggled to capture the temporal dynamics of human movements, limiting their applicability for analyzing complex neurological and psychological phenomena that require an understanding of motion over time (14).

The advent of deep learning and pre-trained models marked a significant leap in pose estimation, particularly in health data analysis. Convolutional Neural Networks (CNNs) and, later, Vision Transformers (ViTs) have been employed to extract fine-grained features from visual data, enabling high-precision pose estimation (15). Pre-trained models, such as OpenPose and MediaPipe, have further revolutionized the field by providing ready-to-use frameworks that can be fine-tuned for specific health applications. For instance, deep learning has been applied to detect early signs of Alzheimer's disease through gait analysis or to monitor stress levels via micro-expressions and postural shifts (16). These methods offer unparalleled accuracy and generalizability, even in unstructured environments such as clinics or homes. However, challenges remain, including the high computational cost of deep learning models and their dependence on large-scale labeled datasets (17). Moreover, ethical concerns related to data privacy and the potential for biased algorithms underscore the need for careful consideration in deploying such technologies for health applications (18).

Based on the limitations of the aforementioned approaches, we propose a novel pose estimation framework designed for health data analysis in neuroscience and psychology. Our method integrates domain-specific knowledge with the latest advancements in deep learning, creating a hybrid system that overcomes the scalability and generalizability issues of earlier approaches. By leveraging pre-trained models fine-tuned with small, high-quality datasets annotated by domain experts, our framework balances accuracy and efficiency. It incorporates temporal modeling techniques, such as Recurrent Neural Networks (RNNs) and Graph Neural Networks (GNNs), to capture the dynamic nature of human movements, providing deeper insights into neurological and psychological conditions. This hybrid approach addresses the challenges of data scarcity and privacy while maintaining robust performance across diverse applications and populations. Our framework not only advances the state-of-the-art in pose estimation but also paves the way for more effective and accessible health monitoring solutions.

The proposed method has several key advantages:

• Our framework introduces a hybrid system that combines domain knowledge with advanced deep learning techniques, offering a unique solution tailored to health data analysis.

• By fine-tuning pre-trained models with small, expert-annotated datasets and incorporating temporal dynamics, the method achieves scalability, adaptability, and high efficiency in diverse scenarios.

• Experiments demonstrate significant improvements in pose estimation accuracy and reliability, enabling precise diagnostics and monitoring of neurological and psychological conditions.

The primary task addressed in this study is health-related pose estimation and analysis, aimed at supporting research and clinical applications in neuroscience and psychology. We define the task as the automatic extraction and interpretation of human posture and movement patterns from heterogeneous data sources, including video recordings, sensor data, and electronic health records, to achieve three core objectives: monitoring disease progression in neurological conditions, such as tracking changes in gait stability in Parkinson's disease patients; evaluating the effectiveness of treatment and rehabilitation interventions by quantifying improvements in movement and posture; and detecting subtle postural cues associated with psychological states, including markers of anxiety, depression, and social interactions. Our Dynamic Medical Graph Framework (DMGF) and Attention-Guided Optimization Strategy (AGOS) are designed to address these objectives by modeling temporal and structural dependencies in multi-modal health data and ensuring clinical interpretability of predictions. This task definition clarifies the relevance of our proposed approach and frames the subsequent discussion on datasets, methodologies, and experimental evaluations in the context of real-world clinical and behavioral applications.

To ensure clarity and maintain a coherent narrative aligned with the objectives of neuroscience and psychology, we emphasize that the central focus of this study is on leveraging pose estimation for health data analysis, particularly to infer motor and cognitive functions relevant to these fields. While methods like BERT embeddings and named entity recognition are traditionally associated with natural language processing, their mention in the manuscript was intended to illustrate the potential for integrating diverse data modalities—including clinical notes and textual records—alongside visual data such as video recordings of movement. However, we recognize that this linkage was not clearly articulated and could have caused confusion. Therefore, in this revision, we explicitly clarify that our primary aim is to develop robust pose estimation techniques, supported by the Dynamic Medical Graph Framework (DMGF) and the Attention-Guided Optimization Strategy (AGOS), to analyze human movement and postures in the context of neurological and psychological conditions. These methods are tailored to model temporal dynamics, structural dependencies, and multimodal features inherent in health data, with a clear focus on clinical applications such as disease progression tracking, patient rehabilitation, and mental health assessments. Any reference to NLP components is now clarified as illustrative rather than central to our current work. We ensure that our experimental choices, including datasets and evaluation metrics, are explicitly tied to the goals of neuroscience and psychology, with consistent justification throughout the manuscript. This alignment reinforces the relevance and clinical significance of our contributions in the targeted health domain.

2 Related work

2.1 Pose estimation in neuroscience applications

Pose estimation has become a pivotal tool in neuroscience research due to its capacity to analyze and interpret human movement patterns with high precision (19). One prominent area where pose estimation is transforming neuroscience is in the study of motor disorders, such as Parkinson's disease, Huntington's disease, and stroke-induced impairments (20). By capturing fine-grained movement dynamics, pose estimation algorithms provide researchers with non-invasive methods to quantify kinematic abnormalities. For example, the use of deep learning-based frameworks, such as OpenPose or MediaPipe, has enabled the accurate tracking of limb movements in real-world settings, bypassing the constraints of traditional motion capture systems that require reflective markers or specialized equipment (21). These advancements have allowed for the precise characterization of tremor frequency, gait disturbances, and limb coordination deficits. To movement disorder diagnostics, pose estimation is being leveraged in neurorehabilitation. Automated feedback systems, powered by pose estimation, are utilized in rehabilitation exercises to provide real-time corrective guidance to patients (22). This is particularly valuable in remote therapy scenarios where physical therapists cannot be present. The integration of pose estimation with wearable sensors has further enhanced the fidelity of these systems, offering multimodal data streams that combine joint angles, muscle activity, and force generation metrics (23). Pose estimation is instrumental in understanding brain-body interaction mechanisms. Recent studies have utilized pose data to map the neural correlates of voluntary and involuntary movements, enabling insights into motor cortex plasticity and its role in recovery from injury. Another emerging application of pose estimation in neuroscience is its use in animal studies. Rodent and primate movement analysis, powered by pose-tracking algorithms, has contributed to understanding neural circuits involved in locomotion and complex motor tasks (24). Algorithms such as DeepLabCut have made it feasible to study fine motor behaviors in animals, advancing research in behavioral neuroscience and neurodegenerative disease models. Pose estimation continues to expand the methodological toolkit of neuroscience, enabling a deeper understanding of movement-related brain function and dysfunction.

2.2 Pose analysis in psychological studies

The application of pose estimation in psychology has opened novel avenues for studying nonverbal communication, emotional expressions, and behavioral patterns (25). Body language and posture are integral to human communication and are often more informative than verbal cues in specific psychological contexts (26). Modern pose estimation frameworks allow for the objective and quantifiable analysis of these cues, providing significant insights into areas such as social interaction, emotional regulation, and mental health. One impactful area of research is the analysis of body posture and movements in mental health diagnostics. Disorders such as depression, anxiety, and autism spectrum disorder (ASD) exhibit unique movement and posture patterns, which can be identified through pose-tracking systems (27). For instance, individuals with depression may exhibit reduced movement dynamics, such as slower gait speeds and limited arm swings, while individuals with anxiety disorders may display jittery, fragmented movements. Pose estimation tools are now being used in clinical assessments to provide objective markers for these conditions, aiding in early diagnosis and personalized treatment planning. Pose estimation is facilitating groundbreaking research in social psychology (28). By quantifying interpersonal dynamics during interactions, such as synchrony in body movements or the mirroring of gestures, researchers can gain insights into phenomena like trust, empathy, and group cohesion. These studies rely on pose estimation algorithms that can track multiple individuals simultaneously, enabling detailed analyses of group behaviors. Moreover, pose data is being integrated with machine learning models to classify specific psychological states, such as attentiveness, stress, or fatigue, based on subtle changes in posture and motion (29). Another promising direction is the use of pose estimation in psychotherapeutic interventions. Virtual reality (VR) environments equipped with pose estimation algorithms allow for the immersive analysis of patient behaviors during therapeutic exercises. For example, individuals undergoing exposure therapy for phobias or PTSD can be monitored for body tension and stress-induced movements. This information can help therapists adapt the intervention in real time, tailoring the experience to individual needs (30). Through these applications, pose estimation is establishing itself as a transformative tool in psychological research and practice.

2.3 Advances in pose-based health monitoring

Pose estimation has found extensive applications in health data analysis, contributing to the development of advanced health monitoring systems that enhance patient care and wellbeing. One prominent area of focus is in elderly care and fall prevention (31). By continuously tracking the posture and movements of elderly individuals, pose estimation systems can identify postural instability or gait abnormalities that may indicate an increased risk of falls. These systems often employ real-time processing to generate alerts, enabling timely interventions. Moreover, pose data has been used to study the biomechanics of aging, providing insights into changes in joint flexibility, muscle strength, and coordination that occur with age (32). Beyond elderly care, pose estimation is revolutionizing sports medicine and injury prevention. Real-time motion analysis, powered by pose estimation, is now a standard tool for monitoring athletic performance and identifying movement patterns that increase the risk of injury. For instance, improper knee alignment during running or landing can be detected using pose tracking algorithms, enabling athletes to correct their form and prevent conditions like ACL injuries (33). Pose-based systems are used in physical therapy settings to assess progress during rehabilitation. By comparing pre- and post-treatment movement data, clinicians can objectively evaluate recovery and adjust therapy protocols as needed (34). Another significant application is in the domain of chronic disease management. Patients with conditions such as arthritis, obesity, or cardiovascular diseases benefit from pose estimation systems that monitor physical activity levels and adherence to prescribed exercise regimens. These systems often integrate with wearable devices to provide comprehensive feedback on joint movements, energy expenditure, and overall physical activity (35). Pose data is increasingly being used in predictive modeling for health outcomes. By analyzing long-term movement patterns, researchers can identify early markers of conditions like frailty, osteoporosis, or metabolic disorders, enabling proactive intervention strategies. Pose estimation is enhancing remote health monitoring solutions, which have become especially critical in the wake of global health crises. Telemedicine platforms now incorporate pose-tracking capabilities, allowing healthcare providers to remotely assess patient mobility, posture, and rehabilitation exercises (36). These advancements have democratized access to healthcare, particularly for individuals in rural or underserved areas. By bridging the gap between technology and healthcare, pose estimation continues to contribute to the evolution of personalized, efficient, and accessible health monitoring systems.

3 Method

3.1 Overview

The rapid advancement in artificial intelligence (AI) has brought transformative changes to the field of health data analysis, where the focus is to derive actionable insights from vast and heterogeneous datasets. The integration of AI in health data analysis encompasses a wide spectrum of tasks, including disease prediction, personalized treatment recommendations, and public health monitoring. We introduce novel computational frameworks that address the challenges of scalability, heterogeneity, and temporal dynamics inherent in health data.

The following subsections are structured to articulate the key contributions and foundations of this work. In Section 3.2, we define the fundamental preliminaries, including the mathematical representation of health data and the challenges posed by its multi-dimensional nature. This lays the groundwork for the subsequent innovations. In Section 3.3, we present our proposed Dynamic Medical Graph Framework (DMGF), which is a novel model designed to capture the intricate interdependencies within health data across spatial and temporal dimensions. Section 3.4 introduces a new Attention-Guided Optimization Strategy (AGOS), which addresses the critical challenge of integrating domain-specific knowledge into AI pipelines, enabling robust and interpretable decision-making processes. This section provides a holistic overview of our approach to leveraging AI for health data analysis. We delineate the systematic steps through which our methodology addresses the critical challenges in this domain. The subsequent sections will elaborate on these components in detail, showcasing their theoretical underpinnings, computational designs, and empirical validations.

Figure 1 presents the overarching architecture of the Dynamic Medical Graph Framework (DMGF), which captures temporal and structural dependencies in multimodal health data using dynamic graphs. This figure provides a conceptual blueprint, showing how data from multiple modalities—such as visual frames, sensor signals, and textual records-are processed through graph-based temporal learning and fused to generate comprehensive health predictions. Figures 2–4 collectively illustrate the hierarchical and interconnected structure of our proposed framework. Figure 2 delves deeper into the attention-guided message passing mechanism within the DMGF. It details how local and global similarities are dynamically computed to refine node embeddings, enhancing the interpretability and adaptability of the graph-based representations in health data contexts. Figure 3 introduces the Attention-Guided Optimization Strategy (AGOS), a complementary module that operates in tandem with DMGF to refine feature selection and ensure domain-constrained optimization. Figure 4 presents the integration of these components for clinical interpretability and decision support. These three components form a cohesive and sequential pipeline: DMGF serves as the core backbone for multimodal graph modeling, the attention-guided mechanism refines message passing within the graphs, and AGOS ensures robustness and clinical relevance in the final predictions. This modular yet integrated design allows our framework to achieve high accuracy while maintaining clinical interpretability. To clarify the relationship among the architectures shown in Figures 1–3, we emphasize that these components are designed to work in a sequential and complementary manner as part of an integrated system. Figure 1 illustrates the overall pipeline, where multimodal data—comprising visual, sensor, and textual inputs—are first preprocessed and structured into a dynamic graph format. This overarching view introduces the conceptual role of graph-based learning in health data modeling. Building on this, Figure 2 elaborates on the internal structure of the Dynamic Medical Graph Framework (DMGF), which serves as the backbone of the model. DMGF performs dynamic graph construction, temporal learning using GRUs, and feature embedding across time and modality dimensions. Figure 3 further decomposes the message passing mechanism used within DMGF by introducing attention-guided computation, where both local and global node similarities are exploited to refine graph features. This mechanism enhances interpretability by emphasizing clinically significant patterns. The three architectures are neither redundant nor alternative; instead, they represent progressively deeper levels of abstraction and operation within a unified and modular system. Each builds upon the last to enable effective modeling of spatiotemporal and semantic dependencies in multimodal health data.

Figure 1

Figure 1. Illustration of real-world clinical applications of our proposed framework. (a) Data Acquisition – Multimodal data acquisition including wearable sensors, video, and clinical notes. (b) Multimodal Fusion & Dynamic Medical Graph Construction – Integration of multimodal inputs into a graph structure. (c) Clinical Interpretability and Decision Support - Providing interpretable clinical metrics and decision guidance.

Figure 2

Figure 2. Overview of the Dynamic Medical Graph Framework (DMGF). This framework integrates attention-guided message passing, graph-based temporal learning, and multi-modal data fusion to model complex interactions in healthcare data. Key components include a graph-based temporal learning module for capturing structural and temporal dependencies, an attention-guided mechanism to enhance interpretability, and a multi-modal fusion layer for integrating heterogeneous medical data. The framework is optimized with temporal consistency and contrastive learning to ensure robust and clinically meaningful predictions.

Figure 3

Figure 3. Attention-guided message passing for dynamic medical graphs. This figure illustrates the framework of DMGF, which employs an attention-guided message passing mechanism to enhance interpretability in medical interactions. The model utilizes local and global similarity computations to refine textual and visual representations, pooling these features through an adaptive focal loss function. A temporal graph convolutional module with attention-based weight assignments dynamically refines node interactions over time, incorporating gated recurrent units (GRUs) for modeling temporal dependencies. The process culminates in a readout function, aggregating node embeddings with learned importance weights to generate final predictions. Regularization techniques enforce temporal smoothness, ensuring stability and clinical relevance in predictions.

Figure 4

Figure 4. The architecture of the Attention-Guided Optimization Strategy (AGOS), integrating Attention-Driven Feature Selection, Domain-Constrained Optimization, and a Temporal-Aware Attention Mechanism. AGOS employs convolutional layers, LSTM networks, and attention mechanisms to enhance feature selection, align predictions with clinical knowledge, and capture temporal dependencies in medical time-series data. The Attention-Driven Feature Selection module extracts meaningful information using Gabor layers, convolutional processing, and fully connected layers, ensuring that clinically relevant features are emphasized while suppressing noise. The Domain-Constrained Optimization framework enforces fairness, clinical alignment, and temporal consistency by integrating domain-specific constraints into the loss function, improving both interpretability and robustness. The Temporal-Aware Attention Mechanism dynamically adjusts attention weights over time, capturing critical phases in disease progression and allowing the model to focus on pivotal moments in patient trajectories.

To further improve conceptual clarity and address the concern of overloading technical buzzwords, Figure 1 provides a visual illustration of how our framework is applied in real-world clinical scenarios. The diagram demonstrates how multimodal data—including wearable sensor measurements, video-based posture sequences, and textual clinical notes—are collected and integrated into a dynamic medical graph representation. This unified representation enables comprehensive spatio-temporal modeling of patient-specific movement patterns and psychological states. Through attention-guided feature selection and domain-constrained optimization, the framework highlights clinically relevant features while suppressing noise, facilitating both interpretability and predictive accuracy. This diagram thus bridges the gap between technical methodology and practical deployment, underscoring the potential for real-world applications in neuro-rehabilitation and psychological behavior monitoring.

3.2 Preliminaries

In this subsection, we organize the problem of health data analysis in the context of artificial intelligence (AI) and introduce the mathematical notations and structures utilized throughout this work. Health data is inherently high-dimensional, heterogeneous, and temporally dynamic, presenting unique challenges for effective computational modeling. We aim to establish a formal representation of the problem to enable the development of robust models capable of addressing these challenges.

Let $D$ denote a health dataset comprising N records. Each record is represented as a tuple r_i = (x_i, y_i), where ${\text{x}}_i\in {ℝ}^d$ corresponds to the d-dimensional input features, and y_i represents the associated outcomes or labels, such as disease diagnosis, treatment effectiveness, or risk scores. For temporal data, each x_i is further decomposed as ${\text{x}}_i={\left\{{\text{x}}_i^{(t)}\right\}}_{t=1}^{T_i}$, where T_i denotes the number of time points for record i and ${\text{x}}_i^{(t)}\in {ℝ}^{d_t}$ represents the feature vector at time t.

The health data is typically collected from multiple sources, such as electronic health records (EHRs), wearable sensors, genomic sequences, and medical imaging. We define $S=\left\{S_1,S_2,\dots ,S_K\right\}$ as the set of K heterogeneous data sources, where each $S_k$ contains domain-specific features and exhibits varying data distributions. A critical challenge lies in integrating these heterogeneous sources into a unified representation.

The goal of health data analysis is to learn a mapping $f:X\to Y$, where $X$ represents the input space and $Y$ denotes the output space. The function f is parameterized by a model $M_{\theta }$ with parameters θ, which are optimized to minimize a task-specific loss function $L(\text{y},\widehat{\text{y}})$, where $\widehat{\text{y}}=f(\text{x})$ is the predicted outcome. This work focuses on designing f such that it accounts for the temporal dynamics, spatial relationships, and domain-specific constraints inherent in health data.

Temporal dependencies in health data, such as disease progression or physiological changes, play a critical role in predictive modeling. Let ${\text{X}}_i=\left[{\text{x}}_i^{(1)},{\text{x}}_i^{(2)},\dots ,{\text{x}}_i^{(T_i)}\right]\in {ℝ}^{T_i\times d_t}$ denote the temporal feature matrix for record i. A temporal model must learn to capture these dependencies by leveraging sequential structures, often using recurrent neural networks (RNNs), temporal convolutional networks (TCNs), or attention mechanisms. To quantify temporal relationships, we define a temporal kernel function $K_t(t,t^{\prime })$, which measures the similarity between features at time points t and t′, ensuring that temporally close events are weighted more heavily in the modeling process.

Heterogeneous data sources introduce challenges due to varying feature spaces and missing data. To address this, we define a feature alignment function ${\varphi }_k:S_k\to Z$, which maps features from each source $S_k$ to a shared latent space $Z$. The unified representation is then constructed as z_i = [ϕ₁(x_i), ϕ₂(x_i), …, ϕ_K(x_i)], where ${\text{z}}_i\in {ℝ}^{d_z}$ is the latent feature vector. Missing data is handled via imputation techniques $I(·)$, which estimate missing values based on observed data and learned patterns.

Many health-related tasks involve understanding relationships between entities, such as interactions among patients, diseases, or treatments. We define a dynamic graph $G=(V,E,\text{A})$, where $V$ is the set of nodes, $E$ is the set of edges representing relationships, and $\text{A}\in {ℝ}^{\left|V\right|\times \left|V\right|}$ is the adjacency matrix. Temporal graphs are represented as $G_t$ for t = 1, …, T, capturing evolving relationships over time. A graph convolutional operator $G(\text{H},\text{A})=\sigma (\text{A}\text{H}\text{W})$, where H is the node feature matrix, W is the learnable weight matrix, and σ(·) is an activation function, is used to model these relationships.

To provide a clearer understanding of the computational characteristics of the proposed model, we outline its complexity in terms of time, memory, and scalability. The primary computational load arises from three sources: the graph convolutional layers in the Dynamic Medical Graph Framework (DMGF), which have a per-layer complexity of $O(\left|E\right|·d)$, where |E| is the number of edges and d is the feature dimension; the temporal modeling using gated recurrent units (GRUs), which scales linearly with the number of time steps T and features as $O(T·d)$; and the attention-guided feature selection and domain-constrained optimization modules, which introduce matrix multiplications and regularization terms that remain tractable due to parallelizability. Empirically, our full model contains ~17.2 million parameters, and the average inference time is around 82 ms per sample on an NVIDIA A100 GPU. While the model is more complex than classical CNN or SVM-based systems, its modular design allows for parallel execution across patient samples and batch processing. For clinical deployment, we acknowledge that large-scale temporal graphs may present memory constraints. We plan to explore model pruning, quantization, and distilled variants to reduce computational overhead without compromising accuracy. These steps will support adaptation to real-time settings and deployment on edge devices or embedded systems in future versions.

Regarding computational complexity, our proposed framework is designed to balance accuracy and efficiency for health data analysis tasks. The Dynamic Medical Graph Framework (DMGF) primarily relies on graph convolutional operations, where the complexity of each layer scales with the number of graph edges |E| and the feature dimensionality d, resulting in an overall complexity of $O(\left|E\right|d)$ per layer. The temporal modeling, implemented through gated recurrent units (GRUs), introduces an additional linear dependence on the number of time steps T. Thus, for dynamic graphs spanning T time points, the overall complexity for the graph-based temporal module is ~$O(T\left|E\right|d)$. The attention-guided feature selection and message passing mechanisms further scale with the number of features and hidden dimensions, but these modules leverage parallelizable matrix operations, ensuring tractability for typical health data scales. The domain-constrained optimization strategy (AGOS) adds a small overhead through additional regularization terms in the loss function, which are linear in the number of features. While these components collectively introduce higher computational demands compared to standard pose estimation models, they remain practical for moderate-sized clinical datasets (hundreds to thousands of patients) when deployed on modern GPUs or specialized accelerators. We also recognize the potential for future optimization through lightweight model variants or pruning strategies, ensuring scalability for real-time clinical applications without compromising interpretability.

3.3 Dynamic Medical Graph Framework (DMGF)

In this section, we present the key innovations of the Dynamic Medical Graph Framework (DMGF), a novel approach designed to model complex interactions in healthcare data by integrating graph-based representations and temporal dynamics (as shown in Figure 2).

To address the evolving nature of health data and multimodal signals, our dynamic medical graph (DMGF) is constructed at each time step t with nodes representing medical entities such as body joints, wearable sensor features, and semantic clinical notes. The edges encode relationships that capture both spatial proximity and domain-specific interactions. To account for temporal evolution, the adjacency matrix A_t and feature matrix H_t are dynamically updated at each step based on temporal similarities and multimodal data fusion. Temporal dependencies are measured using kernel functions on sequential sensor signals and video-based joint trajectories, while cross-modal alignment is performed by mapping different data sources into a shared latent space. The attention mechanism operates within this evolving graph structure by learning attention coefficients that weigh edges and node features according to clinical importance. For each node, a query-key-value scheme computes local attention weights across connected nodes and global attention weights across time. This dual attention design ensures that salient features from different data modalities are emphasized while maintaining temporal consistency. This approach enables our model to dynamically adapt to the changing clinical states and multimodal contexts of each patient, improving both predictive accuracy and clinical interpretability.

Our framework's multimodal fusion module is designed to integrate heterogeneous data sources that provide a comprehensive view of patient movement and health context in neuroscience and psychology applications. The primary input modalities include: visual data, captured as sequences of video frames or static images, which serve as the basis for extracting human body joint positions and movement patterns; wearable sensor data, such as accelerometers, gyroscopes, or electromyography signals, which offer rich temporal information about joint angles, muscle activity, and postural stability; and textual data from electronic health records, clinical notes, or patient-reported outcomes, which capture semantic descriptions of patient status or therapeutic interventions. The multimodal fusion process combines these distinct data types to create a unified representation that leverages the strengths of each modality. This unified representation allows our Dynamic Medical Graph Framework (DMGF) to model both structural and temporal dependencies, while the Attention-Guided Optimization Strategy (AGOS) ensures that clinically relevant features are emphasized. By fusing these modalities, the proposed approach aligns with real-world health data scenarios in neuroscience and psychology, where clinicians and researchers often rely on complementary data streams to understand and monitor patient conditions.

In this study, the multimodal fusion process incorporates three primary data modalities: visual data, such as video recordings or frame-based posture images, which are processed to extract human joint positions, pose trajectories, and spatial movement patterns; wearable sensor data, including accelerometer, gyroscope, and electromyography signals, which provide fine-grained temporal measurements of body dynamics, such as limb acceleration, orientation, and muscle activation; and textual clinical data, such as diagnostic reports, progress notes, and patient self-reported outcomes, which embed contextual and semantic information regarding the patient's health status or treatment protocol. These data sources are first processed into modality-specific feature vectors, then mapped into a shared latent space through dedicated encoders. The fusion strategy—based on attention-weighted aggregation—enables the model to prioritize clinically relevant information across modalities and capture cross-modal interactions. This integrated approach ensures a more holistic representation of patient condition, enhancing the model's ability to track disease progression, assess mental health status, and support personalized treatment planning.

3.3.1 Graph-based temporal learning

To effectively capture the evolution of medical relationships over time, DMGF represents patient data as a sequence of dynamic graphs $G_t=(V,E_t,A_t)$, where $V$ denotes the set of medical entities, $E_t$ represents time-dependent edges, and $A_t$ is the adjacency matrix encoding medical interactions at time t. The temporal evolution of node embeddings is modeled using a graph-based recurrent unit, allowing DMGF to capture both structural dependencies and sequential dynamics. At each time step, the node features are updated through a graph convolution operation before being fed into a recurrent unit:

$\begin{array}{ll}{\text{H}}_t^{(l+1)}=\sigma \left(A_t{\text{H}}_t^{(l)}{\text{W}}^{(l)}\right),& (1)\end{array}$

where ${\text{H}}_t^{(l)}$ represents the node features at layer l, W^(l) is a learnable weight matrix, and σ(·) is an activation function such as ReLU. The final node representations from the graph convolution layers are then processed through a gated recurrent unit (GRU) to model the temporal dependencies:

$\begin{array}{ll}{\text{H}}_t=\text{GRU}\left({\text{H}}_{t-1},{\text{H}}_t^{(L)}\right),& (2)\end{array}$

where ${\text{H}}_t^{(L)}$ is the output of the final graph convolution layer. To enhance stability and prevent vanishing gradients in long-term medical sequences, we incorporate residual connections:

$\begin{array}{ll}{\text{H}}_t={\text{H}}_t+\text{GRU}\left({\text{H}}_{t-1},{\text{H}}_t^{(L)}\right).& (3)\end{array}$

To ensure effective propagation of temporal dependencies, a learnable time encoding T_t is introduced and added to the node representations:

$\begin{array}{ll}{\text{H}}_t={\text{H}}_t+{\text{T}}_t,{\text{T}}_t=\text{MLP}({\text{t}}_t),& (4)\end{array}$

where t_t represents the time information and is processed by a multi-layer perceptron (MLP). To refine the learned temporal representations, an attention mechanism is applied to dynamically weigh the importance of historical states:

$\begin{array}{ll}{\alpha }_t=\frac{exp({\text{H}}_t{\text{W}}_{\alpha }{\text{H}}_{t-1}^{\top })}{\sum _{t^{\prime }}exp({\text{H}}_{t^{\prime }}{\text{W}}_{\alpha }{\text{H}}_{t^{\prime }-1}^{\top })},& (5)\end{array}$

where W_α is a learnable parameter matrix. The final temporal representation is obtained by applying these attention weights:

$\begin{array}{ll}{\text{H}}_t={\displaystyle \sum _{t^{\prime }}}{\alpha }_{t^{\prime }}{\text{H}}_{t^{\prime }}.& (6)\end{array}$

To optimize the framework, we minimize a temporal consistency loss that ensures smooth transitions between consecutive time steps:

$\begin{array}{ll}{L}_{\text{temporal}}={\displaystyle \sum _{t=1}^T}\left|\right|{\text{H}}_t-{\text{H}}_{t-1}|{|}^2.& (7)\end{array}$

This complete framework enables DMGF to effectively model long-term dependencies in patient health trajectories, capturing disease progression patterns and facilitating predictive analytics in dynamic medical environments.

3.3.2 Attention-guided message passing

To enhance interpretability and focus on crucial medical interactions, DMGF employs an attention mechanism within graph convolutional layers, allowing the model to selectively emphasize critical relationships among medical entities over time. The attention weight between nodes i and j at time t is computed as:

$\begin{array}{ll}{\alpha }_{ij}^{(t)}=\frac{exp\left({\text{q}}_i^{(t)}·{\text{k}}_j^{(t)}\right)}{\sum _{j^{\prime }}exp\left({\text{q}}_i^{(t)}·{\text{k}}_{j^{\prime }}^{(t)}\right)},& (8)\end{array}$

where ${\text{q}}_i^{(t)}={\text{W}}_q{\text{h}}_i^{(t)}$ and ${\text{k}}_j^{(t)}={\text{W}}_k{\text{h}}_j^{(t)}$ are query and key vectors, respectively, transformed by learnable weight matrices W_q and W_k. The attention mechanism enables DMGF to learn adaptive node interactions, refining hidden states in the graph convolutional framework. The feature aggregation for node i is then performed as:

$\begin{array}{ll}{\text{h}}_i^{(t+1)}=\sigma \left({\displaystyle \sum _{j\in N(i)}}{\alpha }_{ij}^{(t)}{\text{W}}_v{\text{h}}_j^{(t)}\right),& (9)\end{array}$

where W_v is a learnable transformation matrix, $N(i)$ represents the neighbors of node i, and σ(·) is a non-linear activation function. To incorporate temporal dependencies, DMGF models node evolution using gated recurrent units (GRUs):

$\begin{array}{llll}{\text{z}}_i^{(t)}& =& \sigma \left({\text{W}}_z{\text{h}}_i^{(t)}+{\text{U}}_z{\text{h}}_i^{(t-1)}+{\text{b}}_z\right),& (10)\end{array}$

$\begin{array}{llll}{\text{r}}_i^{(t)}& =& \sigma \left({\text{W}}_r{\text{h}}_i^{(t)}+{\text{U}}_r{\text{h}}_i^{(t-1)}+{\text{b}}_r\right),& (11)\end{array}$

$\begin{array}{llll}{\tilde{\text{h}}}_i^{(t)}& =& tanh\left({\text{W}}_h{\text{h}}_i^{(t)}+{\text{U}}_h({\text{r}}_i^{(t)}\odot {\text{h}}_i^{(t-1)})+{\text{b}}_h\right),& (12)\end{array}$

$\begin{array}{llll}{\text{h}}_i^{(t)}& =& (1-{\text{z}}_i^{(t)})\odot {\text{h}}_i^{(t-1)}+{\text{z}}_i^{(t)}\odot {\tilde{\text{h}}}_i^{(t)},& (13)\end{array}$

where ${\text{z}}_i^{(t)}$ and ${\text{r}}_i^{(t)}$ are update and reset gates, respectively, while ${\tilde{\text{h}}}_i^{(t)}$ represents the candidate hidden state. The learned node representations are further refined through a graph-based readout function to produce the final predictive output:

$\begin{array}{ll}\widehat{\text{y}}=f\left({\displaystyle \sum _i}{\beta }_i{\text{h}}_i^{(T)}\right),& (14)\end{array}$

where β_i denotes importance weights computed via a global attention mechanism, ensuring that medically relevant nodes contribute more significantly to the final prediction. To improve generalization and prevent overfitting, DMGF employs an auxiliary loss term to enforce smoothness in attention distributions:

$\begin{array}{ll}{L}_{\text{smooth}}={\displaystyle \sum _{t=1}^T}{\displaystyle \sum _{i,j}}{\left({\alpha }_{ij}^{(t)}-{\alpha }_{ij}^{(t-1)}\right)}^2.& (15)\end{array}$

This regularization encourages temporal consistency in learned relationships, making predictions more stable and clinically meaningful. Through this integration of attention-guided graph convolutions, temporal memory mechanisms, and adaptive regularization, DMGF effectively models complex, dynamic medical interactions to support accurate and interpretable healthcare predictions (as shown in Figure 3).

3.3.3 Multi-modal data fusion

Deep medical graph fusion (DMGF) integrates heterogeneous medical data sources by constructing a unified graph representation, effectively enhancing predictive accuracy in healthcare analytics. Given K feature modalities, each node v_i at time step t has multi-modal feature representations ${\text{h}}_i^{(t,k)}$. These features are fused using a learnable fusion function ϕ(·) to obtain a comprehensive representation:

$\begin{array}{ll}{\text{h}}_i^{(t)}=\varphi \left({\text{h}}_i^{(t,1)},{\text{h}}_i^{(t,2)},\dots ,{\text{h}}_i^{(t,K)}\right).& (16)\end{array}$

The fusion function ϕ(·) can be attention-based weighted aggregation, concatenation, or other deep learning methods. For instance, using an attention mechanism, the weighted feature aggregation is computed as:

$\begin{array}{ll}{\text{h}}_i^{(t)}={\displaystyle \sum _{k=1}^K}{\alpha }_i^{(k)}{\text{h}}_i^{(t,k)},& (17)\end{array}$

where the attention weight ${\alpha }_i^{(k)}$ is obtained through a softmax function:

$\begin{array}{ll}{\alpha }_i^{(k)}=\frac{exp\left({\text{w}}^{\top }{\text{h}}_i^{(t,k)}\right)}{\sum _{j=1}^{K}exp\left({\text{w}}^{\top }{\text{h}}_i^{(t,j)}\right)},& (18)\end{array}$

where w is a learnable parameter vector. DMGF employs Graph Neural Networks (GNNs) to learn structured relationships through a message-passing mechanism:

$\begin{array}{ll}{\text{h}}_i^{(t+1)}=\sigma \left({\displaystyle \sum _{j\in N(i)}}W{\text{h}}_j^{(t)}+B\right),& (19)\end{array}$

where $N(i)$ represents the neighboring nodes of v_i, W and B are learnable parameters, and σ(·) is a non-linear activation function such as ReLU. To further enhance feature integration, DMGF considers inter-modal relationships and defines a cross-modal similarity matrix:

$\begin{array}{ll}{S}_{k_1,k_2}=\frac{{\text{H}}^{(k_1)}{\text{H}}^{{(k_2)}^{\top }}}{\left|\right|{\text{H}}^{(k_1)}\left|\right|\left|\right|{\text{H}}^{(k_2)}\left|\right|}.& (20)\end{array}$

This matrix guides cross-modal interactions, ensuring efficient information sharing between different data sources. Moreover, DMGF incorporates contrastive learning to optimize node representations, where the loss function is defined as:

$\begin{array}{ll}L=-{\displaystyle \sum _{(i,j)\in P}}log\frac{exp(\text{sim}({\text{h}}_i,{\text{h}}_j))}{\sum _{(i,k)\in N}exp(\text{sim}({\text{h}}_i,{\text{h}}_k))}.& (21)\end{array}$

Here, $P$ denotes the set of positive sample pairs, $N$ represents the negative samples, and sim(·) measures similarity. This loss encourages similar nodes to be closer while ensuring distinction among different categories. DMGF effectively integrates multi-modal features, employs GNN-based graph learning, models cross-modal relationships, and optimizes representations through contrastive learning. These innovations enable DMGF to achieve high predictive accuracy in healthcare applications, including disease prediction, personalized treatment recommendations, and risk assessment.

3.4 Attention-Guided Optimization Strategy (AGOS)

In this section, we introduce the Attention-Guided Optimization Strategy (AGOS), a novel methodology designed to enhance AI-driven health data analysis by integrating domain knowledge, improving model robustness, and ensuring interpretability. Below, we present three key innovations of AGOS (as shown in Figure 4).

3.4.1 Attention-driven feature selection

AGOS utilizes an attention mechanism to prioritize clinically relevant features while suppressing irrelevant dimensions, ensuring that the model focuses on high-impact medical markers. Given an input feature matrix X∈ℝ^N×d, where N represents the number of samples and d the feature dimensions, an attention weight vector a∈ℝ^d is computed to assign importance scores to different features. The attention mechanism is formulated as:

$\begin{array}{ll}\text{a}=\text{softmax}({\text{W}}_a{\text{X}}^{\top }+{\text{b}}_a),& (22)\end{array}$

where ${\text{W}}_a\in {ℝ}^{1\times d}$ and b_a ∈ ℝ are learnable parameters that control feature selection. The refined feature representation is then obtained through element-wise multiplication:

$\begin{array}{ll}{\tilde{\text{x}}}_i=\text{a}\odot {\text{x}}_i.& (23)\end{array}$

To improve the robustness of feature selection, we introduce a gating mechanism that reweighs feature importance dynamically based on medical context:

$\begin{array}{ll}\text{g}=\sigma ({\text{W}}_g{\text{X}}^{\top }+{\text{b}}_g),& (24)\end{array}$

where ${\text{W}}_g\in {ℝ}^{1\times d}$ and b_g ∈ ℝ are additional learnable parameters, and σ(·) is the sigmoid activation function. The final weighted representation integrates the attention and gating outputs:

$\begin{array}{ll}{\widehat{\text{x}}}_i=\text{g}\odot {\tilde{\text{x}}}_i.& (25)\end{array}$

To enhance feature interactions, we introduce a second-order transformation by computing pairwise feature correlations:

$\begin{array}{ll}\text{Z}={\text{X}}^{\top }\text{X},& (26)\end{array}$

where Z∈ℝ^d×d captures feature dependencies. The attention-modulated feature interaction matrix is then computed as:

$\begin{array}{ll}{\text{Z}}_a=\text{a}\odot \text{Z}.& (27)\end{array}$

To regularize attention weights and encourage sparsity, we impose an ℓ₁ constraint on a:

$\begin{array}{ll}{L}_{\text{attn}}=\lambda \left|\right|\text{a}|{|}_1,& (28)\end{array}$

where λ is a regularization coefficient. The optimized feature representation is obtained by applying a non-linear transformation:

$\begin{array}{ll}\text{H}=tanh({\text{W}}_h{\widehat{\text{X}}}^{\top }+{\text{b}}_h),& (29)\end{array}$

where ${\text{W}}_h\in {ℝ}^{d\times d}$ and ${\text{b}}_h\in {ℝ}^d$ refine the feature representation for downstream tasks. This structured approach ensures that AGOS effectively captures clinically relevant features while enhancing model interpretability and decision-making accuracy.

3.4.2 Domain-constrained optimization

To align AI predictions with clinical guidelines and ethical constraints, AGOS integrates domain-specific regularization terms into the loss function, ensuring fairness, interpretability, and adherence to medical protocols. Let $L_{\mathrm{\text{base}}}$ be the base loss function, AGOS extends it as follows:

$\begin{array}{ll}L=L_{\text{base}}+{\lambda }_{\text{domain}}{L}_{\text{domain}},& (30)\end{array}$

where λ_domain is a tunable hyperparameter that balances the trade-off between predictive performance and domain constraints. The domain-specific loss $L_{\mathrm{\text{domain}}}$ consists of multiple penalty terms that enforce fairness, clinical validity, and uncertainty handling. One key component is fairness regularization, ensuring equitable performance across demographic groups. Given a set of demographic subgroups $G$, AGOS minimizes the disparity in predictive performance:

$\begin{array}{ll}{ℒ}_{\text{fair}}={\displaystyle \sum _{g\in G}{\left(E_{x~P_g}[\ell (\widehat{y},y)]-E_{x~P}[\ell (\widehat{y},y)]\right)}^2},& (31)\end{array}$

where P_g denotes the distribution of data for subgroup g, and ℓ(ŷ, y) represents the prediction loss for a given sample. To align predictions with clinical knowledge, AGOS incorporates constraints based on medical guidelines, expressed as logical or mathematical rules. Let $C$ represent a set of clinical rules, AGOS introduces a penalty term:

$\begin{array}{ll}{ℒ}_{\text{clinical}}={\displaystyle \sum _{c\in C}\max }(0,f_c(\widehat{y},x)-{\tau }_c),& (32)\end{array}$

where f_c(ŷ, x) measures the degree of guideline violation, and τ_c is a tolerance threshold. AGOS accounts for uncertainty in medical predictions by penalizing overconfident estimations. The uncertainty-aware regularization is defined as:

$\begin{array}{ll}{ℒ}_{\text{uncertainty}}=\frac{1}{N}{\displaystyle \sum _{i=1}^N\text{Var}}({\widehat{y}}_i|D),& (33)\end{array}$

where $\mathrm{\text{Var}}({ŷ}_i|D)$ represents the model's predictive variance, computed via Monte Carlo dropout or ensemble methods. AGOS also integrates sparsity constraints to ensure interpretability by reducing reliance on redundant features. Given feature importance weights w, a sparsity-inducing penalty is added:

$\begin{array}{ll}{L}_{\text{sparsity}}=\left|\right|\text{w}|{|}_1.& (34)\end{array}$

For time-series healthcare applications, AGOS incorporates a temporal consistency constraint to smooth predictions over consecutive time points, reducing abrupt fluctuations:

$\begin{array}{ll}{L}_{\text{temporal}}={\displaystyle \sum _{t=1}^{T-1}}\left|\right|{\widehat{\text{y}}}_t-{\widehat{\text{y}}}_{t+1}|{|}_2^2.& (35)\end{array}$

The final objective function combines all these components:

$\begin{array}{l}L=L_{\text{base}}+{\lambda }_{\text{fair}}{L}_{\text{fair}}+{\lambda }_{\text{clinical}}{L}_{\text{clinical}}+{\lambda }_{\text{uncertainty}}{L}_{\text{uncertainty}}\\ +{\lambda }_{\text{sparsity}}{L}_{\text{sparsity}}+{\lambda }_{\text{temporal}}{L}_{\text{temporal}}.& (36)\end{array}$

Through this multi-objective loss framework, AGOS ensures that AI-driven healthcare models remain robust, interpretable, fair, and clinically aligned, enabling trustworthy medical decision-making while maintaining predictive accuracy.

3.4.3 Temporal-aware attention mechanism

Adaptive graph-based online selection (AGOS) dynamically adapts its attention distribution over time, capturing critical phases in disease progression and ensuring precise modeling of time-series medical data. Given a time-series input for patient i, denoted as ${\text{X}}_i=\left[{\text{x}}_i^{(1)},\dots ,{\text{x}}_i^{(T)}\right]$, AGOS employs a temporal attention mechanism to assign dynamic importance weights to different time steps. The attention weights α_t at time t are computed as:

$\begin{array}{ll}{\alpha }_t=\frac{exp({\text{q}}^{\top }{\text{k}}_t)}{\sum _{t^{\prime }=1}^{T}exp({\text{q}}^{\top }{\text{k}}_{t^{\prime }})},& (37)\end{array}$

where q is the global query vector, and k_t represents the key vector at time t. This attention formulation ensures that important temporal states receive higher attention, allowing AGOS to identify pivotal moments in patient trajectories. The attended representation of the time-series is then computed as:

$\begin{array}{ll}{\text{h}}_i={\displaystyle \sum _{t=1}^T}{\alpha }_t{\text{x}}_i^{(t)},& (38)\end{array}$

where h_i serves as a compact representation of the patient's history, highlighting the most informative time steps. To enhance interpretability, AGOS incorporates domain-constrained optimization, ensuring the attention mechanism aligns with clinical knowledge. The attention distribution is regularized with entropy minimization to enforce sparsity:

$\begin{array}{ll}{L}_{\text{entropy}}=-{\displaystyle \sum _{t=1}^T}{\alpha }_{t}log{\alpha }_t.& (39)\end{array}$

This loss term ensures that AGOS focuses on a subset of critical time points rather than distributing attention uniformly across all time steps. AGOS integrates domain priors through a weighted auxiliary loss:

$\begin{array}{ll}{L}_{\text{prior}}={\displaystyle \sum _{t=1}^T}{w}_t·{({\alpha }_t-p_t)}^2,& (40)\end{array}$

where w_t represents a clinical prior weight and p_t is the expected importance of time step t, derived from expert annotations. The total loss function for AGOS thus combines predictive loss, attention regularization, and domain priors:

$\begin{array}{ll}L=L_{\text{pred}}+{\lambda }_1{L}_{\text{entropy}}+{\lambda }_2{L}_{\text{prior}},& (41)\end{array}$

where λ₁ and λ₂ are hyperparameters controlling the influence of regularization terms. Moreover, AGOS incorporates temporal-aware self-attention by refining the key vector as:

$\begin{array}{ll}{\text{k}}_t=W_k{\text{x}}_i^{(t)}+{\displaystyle \sum _{j\in N(t)}}{\beta }_{tj}{W}_a{\text{x}}_i^{(j)},& (42)\end{array}$

where $N(t)$ denotes neighboring time steps, and β_tj represents a temporal attention coefficient computed as:

$\begin{array}{ll}{\beta }_{tj}=\frac{exp({\text{x}}_i^{{(t)}^{\top }}{W}_r{\text{x}}_i^{(j)})}{\sum _{j^{\prime }\in N(t)}exp({\text{x}}_i^{{(t)}^{\top }}{W}_r{\text{x}}_i^{(j^{\prime })})}.& (43)\end{array}$

This self-attention mechanism enables AGOS to refine representations based on temporal dependencies, capturing long-range interactions between distant time steps. AGOS combines attention-driven feature selection, domain-constrained optimization, and temporal-aware attention to create an interpretable, robust, and clinically aligned AI framework for health data analysis. By dynamically focusing on critical periods in patient histories, AGOS significantly improves predictive accuracy in time-sensitive healthcare applications, including early disease detection, treatment outcome forecasting, and real-time risk assessment (as shown in Figure 5).

Figure 5

Figure 5. It shows the Temporal-Aware Attention Mechanism (TAM) within the AGOS framework. AGOS uses dynamic attention to capture important phases in disease progression across time-series medical data. Its temporal attention mechanism assigns varying importance weights to time steps, ensuring precise representation of patient histories. The framework integrates domain-constrained optimization with entropy regularization and clinical priors for better interpretability and alignment with clinical knowledge. Temporal-aware self-attention refines temporal dependencies, enabling robust modeling of long-range interactions, significantly enhancing predictive performance in healthcare applications.

4 Experimental setup

4.1 Dataset

The AFLW Dataset (37) is a large-scale collection of annotated facial images designed for facial landmark detection and pose estimation. It contains around 25,000 images with diverse head poses, expressions, and occlusions, ensuring robustness in real-world scenarios. Each face is annotated with up to 21 keypoints, making it valuable for training deep learning models. The dataset is widely used for benchmarking facial alignment and gaze estimation methods, contributing significantly to advancements in facial analysis research. The PoseTrack Dataset (38) is a benchmark for multi-person pose estimation and human tracking in videos. It consists of thousands of annotated video frames with rich pose annotations, enabling models to learn human motion dynamics. Each person is labeled with body keypoints across consecutive frames, making it useful for action recognition and motion prediction. The dataset is challenging due to varying viewpoints, occlusions, and crowded scenes, making it a crucial resource for advancing video-based human pose estimation. The JHMDB Dataset (39) is a human action recognition dataset that integrates pose estimation with video understanding. It includes 928 video clips spanning 21 action categories, each annotated with human body keypoints and segmentation masks. The dataset provides a well-balanced selection of indoor and outdoor activities, making it suitable for studying human motion in realistic settings. By offering both spatial and temporal annotations, JHMDB facilitates the development of models for action recognition, pose-based activity analysis, and motion forecasting. The DeepLesion Dataset (40) is a large-scale medical imaging dataset focused on lesion detection and segmentation in CT scans. Collected from the National Institutes of Health Clinical Center, it contains over 32,000 lesions annotated on a diverse set of patient scans. The dataset includes 3D volumetric information, making it ideal for training deep learning models in automated radiology applications. DeepLesion plays a crucial role in advancing computer-aided diagnosis, enabling researchers to develop more accurate and generalizable lesion detection systems for clinical practice.

4.2 Experimental details

In our experiments, we evaluated the proposed model on several standard benchmarks, including AFLW, PoseTrack, JHMDB, and DeepLesion datasets. The implementation was carried out using PyTorch, with training conducted on an NVIDIA Tesla A100 GPU. The model parameters were initialized using the Xavier initialization method, and optimization was performed using the AdamW optimizer with a learning rate of 1e⁻⁵. A linear learning rate warm-up schedule was applied over the first 10% of training steps, followed by a cosine decay schedule. The batch size was set to 32, and dropout regularization with a rate of 0.3 was applied to prevent overfitting. For the transformer-based architecture, we used a pre-trained BERT-base model as the encoder, with 12 layers, 768 hidden dimensions, and 12 attention heads. The sequence length was capped at 128 tokens for JHMDB and 256 tokens for the other datasets. Fine-tuning was performed for 10 epochs, with early stopping based on validation performance using the F1 score. Gradient clipping with a maximum norm of 1.0 was employed to stabilize training. Evaluation metrics included precision, recall, and F1 score, computed over named entities across all datasets. The evaluation was conducted in two modes: strict entity matching and partial entity matching, to assess both exact and approximate predictions. For datasets like JHMDB, which involve noisy and emerging entities, partial matching was particularly relevant for understanding model robustness. Hyperparameter tuning was conducted using grid search. Learning rates in the range of {1e⁻⁶, 1e⁻⁵, 5e⁻⁵} and dropout rates of {0.1, 0.3, 0.5} were explored. The best-performing configuration was selected based on validation F1 score. Label smoothing with a coefficient of 0.1 was applied to mitigate overconfidence in predictions, particularly for datasets with imbalanced class distributions such as JHMDB. Data preprocessing included tokenization using the WordPiece tokenizer, followed by lowercasing and removal of special characters for JHMDB and DeepLesion datasets. For PoseTrack and AFLW datasets, we retained casing and punctuation to preserve linguistic features. For the multilingual components of PoseTrack, translations were handled using the pre-trained XLM-R model, enabling a unified evaluation across languages. To assess model generalization, we performed a cross-domain evaluation where models trained on AFLW were tested on JHMDB and DeepLesion datasets. This evaluation demonstrated the adaptability of our model in transferring knowledge across datasets with differing characteristics. Statistical significance testing was conducted using the paired bootstrap resampling method, with significance thresholds set at p < 0.05. The mean and standard deviation of the F1 scores were reported for each dataset. The code and datasets will be made publicly available to ensure transparency and reproducibility.

4.3 Comparison with SOTA methods

In this section, we provide a comprehensive comparison of the proposed PoseNet model with several state-of-the-art (SOTA) methods, including HRNet, SimpleBaseline, DarkPose, OpenPose, DEKR, and PRTR. Tables 1, 2 summarize the performance results for all models.

Table 1

Table 1. Evaluation of pose estimation methods on AFLW and PoseTrack datasets.

Table 2

Table 2. Evaluation of pose estimation techniques on the JHMDB and DeepLesion Datasets.

Our PoseNet model consistently outperforms existing methods across all datasets and metrics. For the AFLW dataset, PoseNet achieves the highest F1 score of 91.89 ± 0.03, significantly exceeding the closest competitor, DEKR, which achieves an F1 score of 90.89 ± 0.02. PoseNet also attains the highest AUC of 92.14 ± 0.03, indicating superior classification performance. Similarly, on the PoseTrack dataset, PoseNet demonstrates substantial improvements, with an F1 score of 91.45 ± 0.03, outperforming DarkPose (90.12 ± 0.03) and DEKR (89.50 ± 0.02). The results validate the robustness of PoseNet in handling multilingual and domain-diverse datasets, a key feature of PoseTrack. For the JHMDB dataset, which emphasizes noisy and emerging entity recognition, PoseNet achieves a remarkable F1 score of 90.89 ± 0.03 and an AUC of 91.47 ± 0.03. This surpasses DarkPose, the second-best model, which achieves an F1 score of 88.78 ± 0.02. The performance improvement is attributed to PoseNet's ability to effectively capture contextual dependencies in noisy environments, as well as its fine-tuned handling of rare and emerging entities. On the DeepLesion dataset, PoseNet continues to lead with an F1 score of 91.33 ± 0.03, significantly outperforming DEKR (90.34 ± 0.02) and PRTR (87.92 ± 0.03). The consistent improvement across datasets highlights PoseNet's generalization capabilities.

Figures 6, 7 visually depict the comparative performance across models. Analyzing the results further, we observe that PoseNet's architecture, leveraging advanced attention mechanisms and a robust fine-tuning strategy, contributes significantly to its performance gains. The combination of pre-trained BERT embeddings and domain-specific fine-tuning enables the model to capture both general and domain-specific linguistic features effectively. The introduction of label smoothing and gradient clipping ensures stability during training, particularly for imbalanced datasets like JHMDB. The figures highlight the consistent dominance of PoseNet across all metrics and datasets. These results demonstrate that PoseNet not only outperforms existing SOTA models but also establishes new benchmarks for accuracy, recall, and F1 score in named entity recognition tasks. Moreover, the cross-dataset consistency of PoseNet underscores its robustness and adaptability to various linguistic domains and dataset characteristics. The proposed PoseNet model achieves superior performance on all four datasets. The improvements are particularly pronounced in challenging datasets like JHMDB and multilingual datasets like PoseTrack. These results validate the effectiveness of PoseNet as a robust and generalizable model for named entity recognition tasks.

Figure 6

Figure 6. Analysis of the performance of SOTA techniques on the AFLW and PoseTrack datasets.

Figure 7

Figure 7. Performance analysis of SOTA techniques on the JHMDB and DeepLesion datasets.

4.4 Ablation study

To evaluate the contributions of individual components of PoseNet, we conducted an extensive ablation study. The study examines the effects of removing key modules, including Graph-Based Temporal Learning, Multi-Modal Data Fusion and Domain-Constrained Optimization, on the overall performance of PoseNet. The evaluation was performed on all four datasets: AFLW, PoseTrack, JHMDB, and DeepLesion. Tables 3, 4 summarize the results.

Table 3

Table 3. Results of the ablation study for PoseNet on the AFLW and PoseTrack datasets.

Table 4

Table 4. Results of the ablation study for PoseNet on the JHMDB and DeepLesion datasets.

Removing Graph-Based Temporal Learning, which is responsible for fine-grained feature extraction, leads to a noticeable performance drop across all datasets. On the AFLW dataset, the F1 score decreases from 91.89 ± 0.03 to 88.30 ± 0.02, and the AUC drops from 92.14 ± 0.03 to 90.01 ± 0.03. Similarly, on the JHMDB dataset, removing Graph-Based Temporal Learning results in an F1 score reduction from 90.89 ± 0.03 to 86.12 ± 0.02. This indicates that Graph-Based Temporal Learning plays a critical role in capturing fine-grained linguistic features necessary for accurate named entity recognition, particularly in datasets with high variability like JHMDB. Excluding Multi-Modal Data Fusion, which implements contextual attention mechanisms, causes a significant decrease in Recall and F1 scores, demonstrating its importance for modeling long-range dependencies. For example, on the PoseTrack dataset, the F1 score decreases from 91.45 ± 0.03 to 90.34 ± 0.03, while the Recall drops from 92.87 ± 0.02 to 91.02 ± 0.03. On the DeepLesion dataset, the F1 score drops from 91.33 ± 0.03 to 88.92 ± 0.03. These results underscore the importance of Multi-Modal Data Fusion in capturing contextual relationships that are essential for improving recall, particularly in datasets with diverse linguistic structures. Removing Domain-Constrained Optimization, which introduces domain-specific embeddings, leads to a moderate performance degradation. For instance, on the DeepLesion dataset, the F1 score drops from 91.33 ± 0.03 to 89.67 ± 0.03, while the AUC decreases from 92.12 ± 0.02 to 90.21 ± 0.02. On the AFLW dataset, the F1 score decreases from 91.89 ± 0.03 to 90.15 ± 0.02. The results suggest that Domain-Constrained Optimization contributes to improving domain adaptation, particularly for datasets like DeepLesion that include a wide range of topics and writing styles.

Figures 8, 9 visually illustrate the performance trends, showing significant gains in Accuracy, Recall, F1 Score, and AUC with the inclusion of all modules. The ablation results validate the necessity of each module in enhancing PoseNet's robustness and adaptability to diverse datasets. The ablation study demonstrates that all three modules—Graph-Based Temporal Learning, Multi-Modal Data Fusion, and Domain-Constrained Optimization—are essential for achieving the best performance with PoseNet. Each module contributes uniquely to the model's ability to generalize across datasets, handle noisy and emerging entities, and adapt to domain-specific features.

Figure 8

Figure 8. Analysis of the ablation study for our approach on the AFLW and PoseTrack Datasets. Graph-based temporal learning(GBTL), multi-modal data fusion (MMDF), domain-constrained optimization (DCO).

Figure 9

Figure 9. Ablation analysis of our approach on the JHMDB and DeepLesion Datasets. Graph-based temporal learning (GBTL), multi-modal data fusion (MMDF), domain-constrained optimization (DCO).

While our current experimental evaluation utilizes widely accepted computer vision and medical imaging datasets (AFLW, PoseTrack, JHMDB, DeepLesion), we recognize that these datasets do not directly represent neuroscientific or psychological use cases. These datasets primarily validate the technical effectiveness of our pose estimation framework and associated optimization strategies, particularly in diverse visual conditions and large-scale annotation scenarios. However, they do not adequately reflect the domain-specific challenges of patient-centric analysis, such as those encountered in studies of motor disorders or psychological assessments. This limitation highlights an important gap in our current study's alignment with its stated objectives in neuroscience and psychology. To address this, we propose as part of our future work to incorporate experiments using health-specific datasets that feature actual patient movements and clinical scenarios. For instance, datasets capturing gait dynamics in Parkinson's disease or posture-based behavioral cues in psychological studies would offer a more robust validation of our framework's relevance and practical utility. By emphasizing this alignment in future research, we aim to ensure that our models not only perform well on standard benchmarks but also contribute meaningfully to domain-relevant clinical applications in neuroscience and psychology.

The quantitative results presented in Table 5 demonstrate that our proposed framework consistently surpasses recent state-of-the-art methods across all evaluated health-related datasets. For the Parkinson Gait dataset, our method achieves an accuracy of 92.56%, which is 3.44% higher than the GraphStacked-Hourglass model. Similarly, recall and F1 score improvements reach 3.8 and 3.62%, respectively. In the RehabMov dataset, our method improves upon GCN-LSTM by 2.96% in accuracy, 3.09% in recall, and 3.03% in F1 score, highlighting the effectiveness of our approach in handling rehabilitation-related health data. For the MentalHealth3D dataset, which is more challenging due to its reliance on subtle posture-based cues linked to psychological states, our framework demonstrates a 2.55% increase in accuracy, 2.22% in recall, and 2.38% in F1 score compared to the TransformerPose model. These consistent improvements across diverse datasets and spatio-temporal health applications confirm the superiority of our multimodal, domain-constrained attention mechanisms and their ability to generalize to real-world clinical and behavioral data.

Table 5

Table 5. Performance comparison with recent SOTA methods in health-related pose estimation tasks.

As shown in Table 6, our framework's performance demonstrates sensitivity to key hyperparameters. For the graph convolution layer depth, performance peaks at three layers, suggesting an optimal balance between local feature extraction and over-smoothing effects. Similarly, attention dimensionality shows the highest F1 score at 128 dimensions, beyond which additional complexity leads to diminishing returns. For domain-constrained regularization, moderate weights (around 0.5) provide the best trade-off between prediction accuracy and fairness constraints. These results validate the robustness of our framework across reasonable parameter settings, while highlighting the importance of careful hyperparameter tuning to maximize performance in clinical pose estimation tasks.

Table 6

Table 6. Parameter sensitivity analysis: model performance across different hyperparameter settings.

To further validate the clinical utility of our framework, we introduce a case study using real patient data from the Parkinson Gait dataset. The patient, a 67-year-old male diagnosed with early-stage Parkinson's disease, was monitored over six gait assessment sessions. Multimodal data were collected, including video-recorded walking tasks, wearable IMU sensor data, and periodic clinician reports. Our model processed these inputs through the DMGF and AGOS modules to construct temporal graphs and identify salient movement changes. Over time, the system detected a progressive decrease in stride length and increase in lateral sway. These outputs were corroborated by clinical notes, which described early bradykinesia symptoms. Table 7 summarizes the model's predictive alerts, gait metrics, and clinician annotations across sessions. This case demonstrates the potential of our framework to support early diagnosis, automate progress tracking, and inform intervention adjustments in real clinical workflows.

Table 7

Table 7. Case study: gait feature tracking for Parkinson's disease patient.

To further assess the robustness and generalizability of our model, we conducted a comprehensive parameter sensitivity analysis focusing on three critical hyperparameters: the number of graph convolutional layers in the DMGF module, the dimensionality of attention vectors used in AGOS, and the regularization weight assigned to domain-constrained loss components. This analysis evaluates how changes in these hyperparameters affect the overall performance metrics—Accuracy, Recall, and F1 Score—on the Parkinson Gait dataset, which is representative of real-world clinical data. As shown in Table 8, the model achieves optimal performance when using three GCN layers. With only two layers, the model underfits complex node relationships; whereas with four layers, performance declines due to over-smoothing effects common in deeper GCNs. For the attention dimension, 128 yields the best result by balancing expressiveness and training stability. A lower dimension of 64 restricts feature representation, while a higher value of 256 adds unnecessary complexity and marginal benefit. In terms of domain-constrained regularization, we observe that a moderate weight (0.5) leads to the best performance. Lower weights (0.1) fail to enforce fairness and clinical alignment, while larger weights (1.0) over-penalize model flexibility. These findings not only validate the default configuration of our framework but also provide practical guidance for tuning in diverse clinical deployment scenarios, such as mobile health monitoring or hospital-based gait assessments. The relatively smooth variation across parameter settings indicates that our architecture maintains strong stability and resilience, even in the face of hyperparameter perturbations. In addition to the tabular analysis, we visualize the impact of hyperparameter tuning in Figure 10. The plots illustrate the variation in Accuracy, Recall, and F1 Score across different values of three key hyperparameters: GCN layer depth, attention dimension, and domain regularization weight. The results indicate that a GCN depth of three layers provides the best balance between expressiveness and over-smoothing. Attention dimension of 128 yields superior performance, likely due to its capacity to model feature interactions without unnecessary overhead. The optimal value for domain regularization weight is 0.5, which ensures fairness and clinical alignment while maintaining predictive accuracy. The consistent trend across metrics further confirms the stability and generalizability of our framework under varying configurations.

Table 8

Table 8. Parameter sensitivity analysis: model performance across different hyperparameter settings.

Figure 10

Figure 10. Parameter sensitivity analysis curves. Each plot shows how the model's accuracy, recall, and F1 score vary with respect to one hyperparameter: (Left) GCN layer depth, (Middle) attention dimension, and (Right) domain regularization weight. Optimal values were found at three GCN layers, 128 attention dimensions, and 0.5 regularization weight.

5 Conclusions and future work

To enhance transparency and provide a balanced perspective on the strengths and limitations of our framework, we include an error analysis based on our experiments. We observed that the primary sources of prediction errors are related to motion variability among individuals and occlusions in real-world settings. For example, patients with neurological disorders often exhibit irregular or fragmented movements that challenge the stability of temporal attention mechanisms. The presence of sensor noise and limited viewpoints in clinical videos can affect the precision of multi-modal data fusion. These factors sometimes lead to slight underestimation or overestimation of joint angles, particularly during rapid movements. Nevertheless, our framework demonstrates robust performance in typical scenarios and provides interpretable predictions that align with clinical observations. To further mitigate these issues, future work will explore the incorporation of domain-adaptive learning and cross-patient generalization techniques, enabling more consistent predictions across diverse cohorts. We have also revised our terminology throughout the manuscript, ensuring that each technical term is explained with clear context and relevance, and we have added visual examples that depict real-world clinical scenarios. These improvements aim to make our contributions more transparent and easier to interpret for both technical and clinical audiences.

This study explores the integration of artificial intelligence (AI) into health data analysis, aiming to address challenges in neuroscience and psychology. Traditional methods have struggled with the dynamic, multi-modal, and high-dimensional nature of health datasets, which encompass electronic health records, wearable sensors, and imaging data. To overcome these limitations, the authors propose a novel methodology combining the Dynamic Medical Graph Framework (DMGF) and the Attention-Guided Optimization Strategy (AGOS). DMGF utilizes graph-based representations to model temporal and structural relationships in health data, enabling effective tracking of disease progression and patient interactions. This framework also incorporates temporal graph convolutional networks, which allow for scalability and adaptation across various tasks. AGOS complements DMGF by embedding domain-specific constraints and leveraging attention mechanisms to prioritize key features, ensuring interpretability and alignment with clinical needs. The approach was validated through empirical evaluations, demonstrating improved performance over existing techniques, with notable gains in interpretability and adherence to clinical principles. The study highlights this framework's potential for tasks such as disease prediction, treatment optimization, and public health monitoring, representing a significant advancement in AI-driven health data analysis.

Despite its promising contributions, the study has two notable limitations. While the framework addresses scalability and interpretability, its reliance on advanced graph-based and attention mechanisms may impose computational constraints, especially in resource-limited settings. Future research should explore lightweight implementations or hardware optimization techniques to ensure broader accessibility. The integration of domain-specific knowledge, though a strength, may introduce biases if not carefully validated. Ensuring that the framework remains robust and generalizable across diverse populations and clinical conditions requires extensive validation on larger, more heterogeneous datasets. Looking forward, further development could focus on real-time applications in clinical environments, where immediate decision-making is crucial. Expanding the framework to integrate new data modalities, such as genomic or microbiome data, could unlock novel insights into the interplay between biology, behavior, and mental health. These advancements would strengthen the role of AI in neuroscience and psychology, bridging the gap between theoretical research and practical clinical solutions.

Data availability statement

The original contributions presented in the study are included in the article/supplementary material, further inquiries can be directed to the corresponding author.

Author contributions

JY: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Data curation, Funding acquisition, Resources, Project administration, Supervision, Visualization, Writing – original draft, Writing – review & editing. DZ: Data curation, Writing – original draft, Writing – review & editing, Visualization, Supervision, Funding acquisition.

Funding

The author(s) declare that financial support was received for the research and/or publication of this article. This paper is the research result of the open project of Hubei Teacher Education Research Center in 2025, titled “System Construction and Practical Direction of Generative Artificial Intelligence Supporting Applied Talents Training in Universities” (Project No. jsjy202507).

Conflict of interest

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Generative AI statement

The author(s) declare that no Gen AI was used in the creation of this manuscript.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

References

1. Xu Y, Zhang J, Zhang Q, Tao D. ViTPose: simple vision transformer baselines for human pose estimation. In: Neural Information Processing Systems. (2022). Available online at: https://proceedings.neurips.cc/paper_files/paper/2022/hash/fbb10d319d44f8c3b4720873e4177c65-Abstract-Conference.html

PubMed Abstract | Google Scholar

2. Zheng C, Zhu S, Mendieta M, Yang T, Chen C, Ding Z. 3D human pose estimation with spatial and temporal transformers. In: IEEE International Conference on Computer Vision. Montreal, QC: IEEE (2021). doi: 10.1109/ICCV48922.2021.01145

Crossref Full Text | Google Scholar

3. Wang G, Manhardt F, Tombari F, Ji X. GDR-net: geometry-guided direct regression network for monocular 6D object pose estimation. Computer Vision and Pattern Recognition. Nashville, TN: IEEE (2021). doi: 10.1109/CVPR46437.2021.01634

Crossref Full Text | Google Scholar

4. Yang Z, Zeng A, Yuan C, Li Y. Effective whole-body pose estimation with two-stages distillation. In: 2023 IEEE/CVF International Conference on Computer Vision Workshops (ICCVW). Paris: IEEE (2023). doi: 10.1109/ICCVW60793.2023.00455

Crossref Full Text | Google Scholar

5. Rempe D, Birdal T, Hertzmann A, Yang J, Sridhar S, Guibas L. HuMoR: 3D human motion model for robust pose estimation. In: IEEE International Conference on Computer Vision. Montreal, QC: IEEE (2021). doi: 10.1109/ICCV48922.2021.01129

Crossref Full Text | Google Scholar

6. Wen B, Yang W, Kautz J, Birchfield ST. FoundationPose: unified 6D pose estimation and tracking of novel objects. In: Computer Vision and Pattern Recognition. Seattle, WA: IEEE (2023). doi: 10.1109/CVPR52733.2024.01692

Crossref Full Text | Google Scholar

7. Shan W, Liu Z, Zhang X, Wang Z, Han K, Wang S, et al. Diffusion-based 3D human pose estimation with multi-hypothesis aggregation. In: IEEE International Conference on Computer Vision. Paris: IEEE (2023). doi: 10.1109/ICCV51070.2023.01356

Crossref Full Text | Google Scholar

8. Sundermeyer M, Hodan T, Labbé Y, Wang G, Brachmann E, Drost B, et al. BOP challenge 2022 on detection, segmentation and pose estimation of specific rigid objects. In: 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). Vancouver, BC: IEEE (2023). doi: 10.1109/CVPRW59228.2023.00279

Crossref Full Text | Google Scholar

9. Kim JW, Choi J, Ha E, ho Choi J. Human pose estimation using MediaPipe pose and optimization method based on a humanoid model. Appl Sci. (2023) 13:2700. doi: 10.3390/app13042700

Crossref Full Text | Google Scholar

10. He Y, Huang H, Fan H, Chen Q, Sun J. FFB6D: A full flow bidirectional fusion network for 6D pose estimation. In: Computer Vision and Pattern Recognition. Nashville, TN: IEEE (2021). doi: 10.1109/CVPR46437.2021.00302

PubMed Abstract | Crossref Full Text | Google Scholar

11. Labb'e Y, Carpentier J, Aubry M, Sivic J. CosyPose: consistent multi-view multi-object 6D pose estimation. In: European Conference on Computer Vision. Cham: Springer (2020). doi: 10.1007/978-3-030-58520-4_34

Crossref Full Text | Google Scholar

12. Fang H, Li J, Tang H, Xu C, Zhu H, Xiu Y, et al. AlphaPose: whole-body regional multi-person pose estimation and tracking in real-time. IEEE Trans Pattern Anal Mach Intell. (2022) 45:7157–73. doi: 10.1109/TPAMI.2022.3222784

PubMed Abstract | Crossref Full Text | Google Scholar

13. Lauer J, Zhou M, Ye S, Menegas W, Schneider S, Nath T, et al. Multi-animal pose estimation, identification and tracking with DeepLabCut. Nat Methods. (2022). doi: 10.1101/2021.04.30.442096

Crossref Full Text | Google Scholar

14. Li Z, Liu J, Zhang Z, Xu S, Yan Y. CLIFF: carrying location information in full frames into human pose and shape estimation. In: European Conference on Computer Vision. Cham: Springer (2022). doi: 10.1007/978-3-031-20065-6_34

Crossref Full Text | Google Scholar

15. Thite S, Ravishankar J, Tomeo-Reyes I, Ortiz AM. Design of a simple rubric to peer-evaluate the teamwork skills of engineering students. Eur. J. Eng. Educ. (2024) 49:623–46. doi: 10.1080/03043797.2024.2338239

Crossref Full Text | Google Scholar

16. Martinelli G, Diprima F, Bisagno N, Conci N. Ski pose estimation. Star. (2024) 120–125. doi: 10.1109/STAR62027.2024.10635966

Crossref Full Text | Google Scholar

17. Lin YC, Florence PR, Barron J, Rodriguez A, Isola P, Lin TY. iNeRF: inverting neural radiance fields for pose estimation. In: IEEE/RJS International Conference on Intelligent Robots and Systems. (2020). Available online at: https://ieeexplore.ieee.org/abstract/document/9636708/

Google Scholar

18. Li Y, Zhang S, Wang Z, Yang S, Yang W, Xia S, et al. TokenPose: learning keypoint tokens for human pose estimation. In: IEEE International Conference on Computer Vision. Montreal, QC: IEEE (2021). doi: 10.1109/ICCV48922.2021.01112

Crossref Full Text | Google Scholar

19. França LGS, Miranda JGV, Leite M, Sharma NK, Walker MC, Lemieux L, et al. Fractal and multifractal properties of electrographic recordings of human brain activity: toward its use as a signal feature for machine learning in clinical applications. Front Physiol. (2018) 9:1767. doi: 10.3389/fphys.2018.01767

PubMed Abstract | Crossref Full Text | Google Scholar

20. Chen T, Fang C, Shen X, Zhu Y, Chen Z, Luo J. Anatomy-aware 3D human pose estimation with bone-based pose decomposition. IEEE Trans Circ Syst Video Technol. (2021) 32:198–209. doi: 10.1109/TCSVT.2021.3057267

Crossref Full Text | Google Scholar

21. Rong Y, Shiratori T, Joo H. FrankMocap: a monocular 3D whole-body pose estimation system via regression and integration. In: 2021 IEEE/CVF International Conference on Computer Vision Workshops (ICCVW). Montreal, BC: IEEE (2021). doi: 10.1109/ICCVW54120.2021.00201

Crossref Full Text | Google Scholar

22. Moon G, Yu SI, Wen H, Shiratori T, Lee KM. InterHand2.6M: a dataset and baseline for 3D interacting hand pose estimation from a single RGB image. In: European Conference on Computer Vision. Cham: IEEE (2020). doi: 10.1007/978-3-030-58565-5_33

Crossref Full Text | Google Scholar

23. Chen W, Jia X, Chang H, Duan J, Shen L, Leonardis A. FS-Net: fast shape-based network for category-level 6D object pose estimation with decoupled rotation mechanism. In: Computer Vision and Pattern Recognition. Nashville, TN: IEEE (2021). doi: 10.1109/CVPR46437.2021.00163

Crossref Full Text | Google Scholar

24. Palermi S, Brugin E, Schiavon M, Tulipano Di Franco F, Sartori P, Baioccato V, et al. Sport-related pneumomediastinum in a synchronized swimmer: from diagnosis to return to play. Phys Sportsmed. (2025) 23:1–4. doi: 10.1080/00913847.2025.2457313

PubMed Abstract | Crossref Full Text | Google Scholar

25. Hanson CS, Outhred T, Brunoni AR, Malhi GS, Kemp AH. The impact of escitalopram on vagally mediated cardiovascular function to stress and the moderating effects of vigorous physical activity: a randomized controlled treatment study in healthy participants. Front Physiol. (2013) 4:259. doi: 10.3389/fphys.2013.00259

PubMed Abstract | Crossref Full Text | Google Scholar

26. Zheng C, Wu W, Yang T, Zhu S, Chen C, Liu R, et al. Deep learning-based human pose estimation: a survey. ACM Comput Surv. (2020) 56:1–37. doi: 10.1145/3603618

Crossref Full Text | Google Scholar

27. Maji D, Nagori S, Mathew M, Poddar D. YOLO-pose: enhancing YOLO for multi person pose estimation using object keypoint similarity loss. In: 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). New Orleans, LA: IEEE (2022). doi: 10.1109/CVPRW56347.2022.00297

Crossref Full Text | Google Scholar

28. Liu H, Fang S, Zhang Z, Li D, Lin K, Wang J. MFDNet: collaborative poses perception and matrix Fisher distribution for head pose estimation. IEEE Trans Multimed. (2022) 24:2449–60. doi: 10.1109/TMM.2021.3081873

Crossref Full Text | Google Scholar

29. Chen H, Wang P, Wang F, Tian W, Xiong L, Li H. EPro-PnP: generalized end-to-end probabilistic perspective-n-points for monocular object pose estimation. In: Computer Vision and Pattern Recognition. New Orleans, LA: IEEE (2022). doi: 10.1109/CVPR52688.2022.00280

PubMed Abstract | Crossref Full Text | Google Scholar

30. Palermi S, Vittadini F, Pasta G, Zappia M, Corsini A, Pedret C, et al. The challenge of thigh tendon reinjuries: an expert opinion. J Basic Clin Physiol Pharmacol. (2024) 35:335–45. doi: 10.1515/jbcpp-2024-0177

PubMed Abstract | Crossref Full Text | Google Scholar

31. Xu T, Takano W. Graph stacked hourglass networks for 3D human pose estimation. In: Computer Vision and Pattern Recognition. Nashville, TN: IEEE (2021). doi: 10.1109/CVPR46437.2021.01584

Crossref Full Text | Google Scholar

32. Liu H, Liu T, Zhang Z, Sangaiah AK, Yang B, Li Y. ARHPE: asymmetric relation-aware representation learning for head pose estimation in industrial human–computer interaction. IEEE Trans Ind Inf. (2022) 18:107–17. doi: 10.1109/TII.2022.3143605

Crossref Full Text | Google Scholar

33. Sun J, Wang Z, Zhang S, He XH, Zhao H, Zhang G, et al. OnePose: one-shot object pose Estimation without CAD models. In: Computer Vision and Pattern Recognition. New Orleans, LA: IEEE (2022). doi: 10.1109/CVPR52688.2022.00670

Crossref Full Text | Google Scholar

34. Serir A, Tuff J, Rook N, Fongaro E, Schreiber T, Peus E, et al. Balanced anesthesia in pigeons (Columba livia): a protocol that ensures stable vital parameters and feasibility during long surgeries in cognitive neuroscience. Front Physiol. (2024) 15:1437890. doi: 10.3389/fphys.2024.1437890

PubMed Abstract | Crossref Full Text | Google Scholar

35. Li W, Liu H, Tang H, Wang P, Gool L. MHFormer: multi-hypothesis transformer for 3D human pose estimation. In: Computer Vision and Pattern Recognition. New Orleans, LA: IEEE (2021). doi: 10.1109/CVPR52688.2022.01280

Crossref Full Text | Google Scholar

36. Palermi S, Vecchiato M, Brusiani S, De Martino L, Cavarretta E, Biffi A. Age-related distribution of cardiovascular risk factors in a Corporate Wellness Program: a case study of occupational cardiology at Ferrari. Eur J Prev Cardiol. (2024) 31(Supplement_1): 175–309. doi: 10.1093/eurjpc/zwae175.309

PubMed Abstract | Crossref Full Text | Google Scholar

37. Cai M, Zhang S, Xiao G, Fan S. 3D face reconstruction and dense alignment with a new generated dataset. Displays. (2021) 70:102094. doi: 10.1016/j.displa.2021.102094

Crossref Full Text | Google Scholar

38. Bao Q, Liu W, Cheng Y, Zhou B, Mei T. Pose-guided tracking-by-detection: robust multi-person pose tracking. IEEE Trans Multimed. (2020) 23:161–75. doi: 10.1109/TMM.2020.2980194

Crossref Full Text | Google Scholar

39. Le H, Hsu CC, Lu CK, Wang WY, Kuo PYM. Human action recognition on edge devices: a novel light-weight model. In: 2024 IEEE 13th Global Conference on Consumer Electronics (GCCE). Kitakyushu: IEEE (2024). p. 910–11. doi: 10.1109/GCCE62371.2024.10760363

Crossref Full Text | Google Scholar

40. Zlocha M, Dou Q, Glocker B. Improving RetinaNet for CT lesion detection with dense masks from weak RECIST labels. In: Medical Image Computing and Computer Assisted Intervention-MICCAI 2019: 22nd International Conference, Shenzhen, China, October 13-17, 2019, Proceedings, Part VI 22. Cham: Springer (2019). p. 402–10. doi: 10.1007/978-3-030-32226-7_45

Crossref Full Text | Google Scholar

41. Wu H, Liang C, Liu M, Wen Z. Optimized HRNet for image semantic segmentation. Expert Syst Appl. (2021) 174:114532. doi: 10.1016/j.eswa.2020.114532

Crossref Full Text | Google Scholar

42. Wu Y, Jiang L, Yang Y. Revisiting embodiedqa: a simple baseline and beyond. IEEE Trans Image Process. (2020) 29:3984–92. doi: 10.1109/TIP.2020.2967584

PubMed Abstract | Crossref Full Text | Google Scholar

43. Liu H, Liu F, Fan X, Huang D. Polarized self-attention: towards high-quality pixel-wise mapping. Neurocomputing. (2022) 506:158–67. doi: 10.1016/j.neucom.2022.07.054

Crossref Full Text | Google Scholar

44. Viswakumar A, Rajagopalan V, Ray T, Parimi C. Human gait analysis using OpenPose. In: 2019 Fifth International Conference on Image Information Processing (ICIIP). Shimla: IEEE (2019). p. 310–4. doi: 10.1109/ICIIP47207.2019.8985781

Crossref Full Text | Google Scholar

45. Chao W, Duan F, Du P, Zhu W, Jia T, Li D. DEKRV2: more accurate or fast than DEKR. In: 2022 IEEE International Conference on Image Processing (ICIP). Bordeaux: IEEE. (2022). p. 1451–5. doi: 10.1109/ICIP46576.2022.9897550

Crossref Full Text | Google Scholar

46. Ono H. Study on Environmental Evaluation Using PRTR. In: 2018 7th International Congress on Advanced Applied Informatics (IIAI-AAI). Yonago: IEEE (2018). p. 823–8. doi: 10.1109/IIAI-AAI.2018.00168

Crossref Full Text | Google Scholar

47. Wu Z, Huang M, Zhao A, Ian Z. Traffic prediction based on GCN-LSTM model. J Phys Conf Ser. (2021). 1972:012107. doi: 10.1088/1742-6596/1972/1/012107

PubMed Abstract | Crossref Full Text | Google Scholar

48. Xi X, Zhang C, Jia W, Jiang R. Enhancing human pose estimation in sports training: Integrating spatiotemporal transformer for improved accuracy and real-time performance. Alex Eng J. (2024) 109:144–56. doi: 10.1016/j.aej.2024.08.072

Crossref Full Text | Google Scholar