Editorial: The Applications of AI Techniques in Medical Data Processing

Pengfei Zhang^*

• Chengdu University of Traditional Chinese Medicine, Chengdu, China

A central theme of this collection is the utilization of ML to identify high-risk patients across diverse clinical scenarios. In acute care, studies on urosepsis emphasize the necessity of interpretability. By employing XGBoost and SHapley Additive exPlans (SHAP),Significant adverse prognostic events in patients with urosepsis: a machine learning based model development and validation study identifies high-risk individuals with an AUC of 0.904.This focus on clinical utility is supported by Preparing hospitals and health organizations for AI: practical guidelines for the required infrastructure, which argues that prognostic accuracy must be matched by robust organizational readiness. Predictive efficacy is further demonstrated in cardiovascular and respiratory conditions.Using machine learning models to predict post-revascularization thrombosis in PAD achieves an AUC of 0.76 by combining patient baseline characteristics with viscoelastic testing.Similarly, Risk prediction of stroke-associated pneumonia in acute ischemic stroke with atrial fibrillation using machine learning models utilizes a logistic regression-based nomogram to reach an accuracy of 90.13%. For heart transplantation, a meta-analysis in Mortality prediction of heart transplantation using machine learning models: a systematic review and meta-analysis identifies CatBoost as the superior algorithm (AUC 0.80) while cautioning against high risks of bias in current modeling flows.In infectious disease screening, Comparative analysis of frequentist, Bayesian, and machine learning models for predicting SARS-CoV-2 PCR positivity finds that random forest classifiers (AUC 0.947-0.963) significantly outperform traditional logistic regression for rapid symptom-based screening. Advanced imaging and multi-modal architectures represent a primary frontier in this Research Topic.The Universal medical image segmentation via in-context cross-attention introduces a novel framework that enhances accuracy across 135 different segmentation tasks by leveraging support-set pre-selection.More targeted approaches include An efficient method for early Alzheimer's disease detection based on MRI images using deep convolutional neural networks, which achieves a 99.68% accuracy in stage-specific AD detection. The integration of metadata is also crucial; Comparative analysis of multimodal architectures for effective skin lesion detection using clinical and image data highlights how cross-attention fusion between dermatoscopic images and clinical attributes outperforms image-only baselines.In ophthalmology, Multicenter evaluation of machine and deep learning methods to predict glaucoma surgical outcomes utilizes 1D-CNNs to predict surgical failure with an AUROC of 76.4%.Furthermore, genomic analysis is advanced by Construction of a diagnostic model for temporal lobe epilepsy using interpretable deep learning: disease-associated markers identification, which employs Kolmogorov-Arnold Networks (KAN) to map complex nonlinear relationships between genetic features like DEPDC5 and LGI1 and disease status. AI is deeply integrated into chronic disease and mental health management. Predicting the risk of depression in older adults with disability using machine learning: an analysis based on CHARLS data reveals that subjective perceptions, such as life satisfaction, contribute more to prediction (SHAP value 0.339) than traditional biomedical indicators.For pain management, Predicting therapy dropout in chronic pain management: a machine learning approach to cannabis treatment utilizes random forest classifiers to achieve 80% accuracy in predicting treatment adherence, identifying high THC dosage as a primary risk factor for dropout. The role of Large Language Models (LLMs) presents a critical dichotomy. Performance of GPT-4 for planning acupuncture treatment: comparison with human clinician performance shows a promising 51.3% overlap between AI and expert acupoint selection, suggesting educational potential.Conversely, Limitations of broadly trained LLMs in interpreting orthopedic Walch glenoid classifications reveals that generalist models like Claude 3.5 fail significantly in specialized visual tasks, with DeepSeek R1 achieving only 44% accuracy on orthopedic diagrams. Scaling medical AI requires specialized foundational systems.The development and use of data warehousing in clinical settings: a scoping review notes that while specialized data warehouses excel in decision support, they face unique scalability challenges compared to general-purpose architectures.This infrastructure gap is further emphasized in Evaluating the effectiveness of AI-enhanced "One Body, Two Wings" pharmacovigilance models in China: a nationwide survey on medication safety and risk management, which identifies information system effectiveness as the most critical predictor (mean importance 0.53) of pharmacovigilance success. Efficiency and explainability remain paramount. From data extraction to analysis: a comparative study of ELISE capabilities in scientific literature introduces ELISE as a tool specifically designed for regulatory and clinical traceability.For real-time applications, a novel interpretable and real-time dengue prediction framework using clinical blood parameters with genetic and GAN-based optimization demonstrates how GA-GAN-XAI frameworks can achieve 99.49% accuracy with minimal latency (0.0013s) . Furthermore, Pose estimation for health data analysis: advancing AI in neuroscience and psychology proposes Dynamic Medical Graph Frameworks (DMGF) to model temporal relationships in multi-modal neuroscience data. Crucially, the ethical deployment of these tools must address technical vulnerabilities. Evaluating XAI techniques under class imbalance using CPRD data provides a critical observation: class imbalance in clinical datasets significantly undermines the consistency of LIME and SHAP explanations, highlighting a potential risk for trustworthy AI deployment. This Research Topic illustrates that the application of AI in medical data processing is entering a more mature and critical stage of development. The emphasis is gradually shifting beyond isolated gains in predictive accuracy toward deeper inter-modal integration, ethical transparency, and the readiness of supporting clinical and computational infrastructures. As these technologies continue to evolve, the effective integration of expert human oversight with high-performing and interpretable algorithms will become a cornerstone for building safer, more reliable, and more efficient healthcare systems. The 21 contributions in this Research Topic collectively signal that medical AI is transitioning from "algorithmic novelty" to "clinical maturity". By synthesizing the findings of this diverse collection, four critical forward-looking observations for the field are summarized as follows.1) The Fragility of Interpretability in Skewed Data: A pivotal insight from our collection is the sensitivity of XAI tools like LIME and SHAP to class imbalance in clinical datasets. Future research must shift from merely applying post-hoc explanations to developing distribution-aware interpretability frameworks that remain consistent across varying disease prevalences, ensuring that AI-driven clinical reasoning does not misguide physicians in rare-disease scenarios.2) The Transition from Generalist LLMs to Specialized Vision-Language Models: Our findings contrast the promising diagnostic support of LLMs in textual domains with their stark failure in visual medical classification (e.g., orthopedic Walch classifications). This suggests that the future of medical AI lies in domain-specific foundation models that integrate expert medical-grade visual reasoning with textual knowledge, rather than reliance on broadly trained generalist architectures. 3) Infrastructure as the Primary Determinant of Scalability: The success of the "One Body, Two Wings" pharmacovigilance model and the insights from data warehousing reviews highlight that the bottleneck of clinical AI is increasingly structural rather than computational. Future investments must prioritize AI-native healthcare IT ecosystems that balance specialized analytical power (specialized warehouses) with the interoperability required for large-scale clinical impact. 4) Hyper-Multimodal and Temporal Synergy: The move toward dynamic medical graphs and in-context cross-attention for segmentation demonstrates that the standard for precision diagnostics is shifting toward temporal integration. Integrating longitudinal patient patterns with high-dimensional multi-modal data (genomics, clinical metrics, and imaging) will be the cornerstone for realizing truly personalized, data-driven healthcare.