Predictive nomogram for post-stroke motor dysfunction using fNIRS

Background: There is a lack of objective evaluation tools for assessing upper limb motor dysfunction in ischemic stroke patients (ULMD-IS). This study aimed to develop and validate a diagnostic nomogram for diagnosing the severity of ULMD-IS using functional near-infrared spectroscopy (fNIRS) data.

Methods: This retrospective analysis included 275 ULMD-IS patients at Tianjin Medical University General Hospital. Patients were randomly assigned to a training group (n = 193) or a validation group (n = 82). The data were preprocessed using HOMER2. In the training group, least absolute shrinkage and selection operator (LASSO) and multivariate logistic regression were employed to identify predictive variables and construct the nomogram. The nomogram’s performance was validated using the area under the receiver operating characteristic curve (AUC), the Hosmer-Lemeshow goodness-of-fit test, calibration curves, and decision curve analysis (DCA).

Results: No significant differences in baseline characteristics, including sex, age, lesion hemisphere, or medical history, were observed between the training and validation groups. LASSO regression analysis identified three independent risk factors: deoxyhemoglobin (HbD) levels in the affected temporal region, total hemoglobin (HbT) levels in the total region, and HbT levels in the unaffected frontopolar region. These factors showed good differentiation ability (training group AUC: 0.766, verification group AUC: 0.861). The model’s goodness-of-fit was confirmed, and it demonstrated a favorable net clinical benefit. Additionally, correlation analysis between these model variables and activities of daily living (ADL) scores revealed no significant relationships (p > 0.05 for all variables), indicating that the identified risk factors may not directly influence ADL performance.

Conclusion: This study identified HbD in the affected temporal region, Total HbT levels, and HbT in the unaffected frontopolar region as independent risk factors for diagnosing the severity of ULMD-IS, and a corresponding predictive model was constructed. Given the model’s limited sensitivity, the nomogram should be regarded only as a supplementary reference for objectively assessing post-stroke motor dysfunction; its utility in predicting treatment outcomes and guiding therapeutic choices remains modest and warrants cautious interpretation.

1 Introduction

Individuals face an elevated risk of developing cerebrovascular diseases (Chiu et al., 2014), with significantly higher morbidity, mortality, and disability rates observed in those who have suffered an ischemic stroke (He et al., 2014). Motor dysfunction is a prevalent sequela in stroke patients, exerting a substantial negative impact on their ability to perform daily self-care activities and diminishing their overall quality of life. Currently, the clinical evaluation of poststroke motor dysfunction predominantly relies on various assessment scales (Longhi et al., 2016), including the Fugl-Meyer Assessment Scale and the Modified Ashworth Scale (Van Der Velden et al., 2022, Kongwattanakul et al., 2020). Nevertheless, these scale assessment methods have significant subjectivity, and the assessment results for the same patient may vary greatly among different evaluators, affecting the accuracy and reliability of the diagnosis. Moreover, scale assessments generally offer only a general functional score, posing a challenge for in-depth exploration of the specific pathophysiological mechanisms and subtle characteristics of motor dysfunction (Subramanian et al., 2010). Consequently, there is a pressing need for a diagnostic approach that is more objective, precise, and comprehensive in capturing the full extent of poststroke motor dysfunction.

Functional Near-Infrared Spectroscopy (fNIRS) represents an innovative neuroimaging technology that relies primarily on the principle of neurovascular coupling to monitor the activity state of the brain (Teo et al., 2016, Kinder et al., 2022), offering various advantages including non-invasiveness, real-time continuous monitoring capabilities, and relatively high spatial and temporal resolution (Li et al., 2022, Xie et al., 2022, Eastmond et al., 2022). fNIRS measures changes in the concentration of oxygenated hemoglobin (HbO), deoxygenated hemoglobin (HbD), and total hemoglobin (HbT) in the cerebral cortex, reflecting the hemodynamic response to neural activity (Scholkmann et al., 2014). Studies have shown that changes in HbO and HbT are closely related to the intensity and duration of neural activity, while changes in HbD reflect the metabolic demands of local brain regions (Huppert et al., 2009). These hemodynamic indicators not only provide information about regional brain activation but also reveal patterns of functional connectivity between brain regions. Therefore, fNIRS has unique advantages in assessing post-stroke motor dysfunction, providing detailed information about regional brain activation and functional connectivity patterns (Pinti et al., 2020). Importantly, fNIRS is particularly suitable for patients with upper limb motor dysfunction following ischemic stroke (ULMD-IS), who often have limited mobility. Traditional imaging techniques such as CT and MRI can be quite challenging for patients and their families, especially since MRI has a low tolerance for motion artifacts and requires high patient compliance. In contrast, fNIRS is portable, easy to operate, and has a high tolerance for motion artifacts, making it more suitable for imaging in patients with ULMD-IS.

Current fNIRS research on poststroke motor function tends to focus on functional connectivity between regions of interest (ROIs), while neglecting the functional connectivity of the ROIs themselves (Cai et al., 2024, Liu et al., 2015, Almeida et al., 2017). Research has shown that the recovery of motor function after stroke is closely related to the hemodynamic response in the cerebral cortex. For example, increases in HbO and HbT are typically associated with enhanced regional brain activation, while decreases in HbD reflect reduced metabolic demands in local brain regions (Huppert et al., 2009). Additionally, patterns of functional connectivity between brain regions play a crucial role in motor recovery. Studies have shown that the recovery of motor function after stroke is closely related to increased strength and complexity of functional connectivity between brain regions (Grefkes and Fink, 2014). Therefore, monitoring changes in HbO, HbT, and HbD using fNIRS can provide a deeper understanding of the pathophysiological mechanisms of post-stroke motor dysfunction and offer important biomarkers for motor recovery (Pinti et al., 2020).

Our research focused not only on the entire brain but also pays attention to the correlation between functional connectivity within the ROI and the severity of ULMD-IS. Previous studies have shown that the recovery of motor function after stroke is closely related to increased strength and complexity of functional connectivity between brain regions (Grefkes and Fink, 2014). For example, Grefkes (Grefkes et al., 2008) found that the recovery of motor function after stroke was associated with enhanced functional connectivity between the affected and unaffected hemispheres. Additionally, Pinti (Pinti et al., 2020) demonstrated that changes in HbO and HbT monitored by fNIRS could effectively predict motor recovery in stroke patients. Therefore, monitoring changes in HbO, HbT, and HbD using fNIRS can provide a deeper understanding of the pathophysiological mechanisms of post-stroke motor dysfunction and offer important biomarkers for motor recovery.

In this study, we specifically focused on different types of functional connectivity edges to more finely describe the structure of brain networks. Specifically, “Short1 of edges” refers to intrahemispheric connections, that is, the number of edges between nodes within the same region of interest (or subnetwork). “Short2 of edges” refers to intrahemispheric connections, that is, the number of edges between nodes in different regions of interest (or subnetworks). “Long1 of edges” refers to interhemispheric connections, that is, the number of edges between nodes in symmetric regions of interest (or subnetworks) between the left and right hemispheres. “Long2 of edges” refers to interhemispheric connections, that is, the number of edges between nodes in asymmetric regions of interest (or subnetworks) between the left and right hemispheres (Zhu et al., 2017). These classifications help us gain a deeper understanding of the organizational structure of brain networks and how they are related to motor dysfunction after stroke.

The goal of this study was to develop a nomogram based on resting-state brain functional connectivity for diagnosing motor dysfunction after stroke. We comprehensively collected resting-state fNIRS data and relevant clinical information, including age, sex, disease progression, and lesion location, from stroke patients. The innovation of this study lies in the fact that we not only identified key indicators with significant diagnostic value, but also developed a diagnostic model using statistical methods and transformed it into an intuitive and visual nomogram. The clinical value of this visual predictive model (nomogram) lies in its ability to simplify complex diagnostic models into an easily understandable and operable format for use by clinicians in practical settings. We carefully validated and optimized the model, which is expected to significantly improve the accuracy and efficiency of diagnosis, advancing the application and development of neuroimaging technology in clinical practice.

2 Materials and methods

2.1 Participant recruitment

This study received approval from the Ethics Committee of Tianjin Medical University General Hospital (Ethics Number: IRB2022-YX-054-01) and adheres to the principles outlined in the Declaration of Helsinki. Given its retrospective research design, the study was granted a waiver for obtaining informed consent from the participants. All patient medical data were anonymized and deidentified prior to analysis. To ensure patient privacy and ethical compliance, the following measures were taken: (1) Data anonymization: All patient identifiers were removed from the dataset, and the data were coded using unique identifiers that could not be traced back to the patients. This process was supervised by a dedicated individual to ensure that no personal information was inadvertently disclosed. (2) Data access control: Access to the anonymized dataset was strictly limited to research team members who had undergone ethics training and signed confidentiality agreements. The data were stored on a secure server with restricted access, and all data transfers were encrypted to prevent data breaches. (3) Ethical oversight: The study was subject to ongoing ethical oversight by the hospital’s ethics committee, which reviewed the study’s progress and ensured that all ethical guidelines were followed throughout the research process. Any potential ethical issues were promptly addressed and resolved. (4) Patient benefits: The study aimed to improve the diagnosis of upper limb motor dysfunction in ischemic stroke patients, which could potentially benefit future patients. The waiver of informed consent was justified on the grounds that the study posed minimal risk to participants and had the potential to significantly contribute to medical knowledge and patient care. The study involved screening data from all patients who were diagnosed with ischemic stroke and admitted to the Rehabilitation Department of Tianjin Medical University General Hospital between October 2023 and August 2024; 392 patients were included. The inclusion criteria for the study were as follows: (1) A confirmed diagnosis of ischemic stroke established through MRI or CT imaging; (2) Provision of informed consent by the patient or their authorized representative; and (3). Age of 18 years or older. The exclusion criteria were as follows: (1) Patients with severe aphasia, cognitive impairment, or consciousness disturbance who are unable to cooperate with the examination: These conditions can affect patients’ cooperation and the accuracy of the examination results. Therefore, to ensure the reliability of the research data and the safety of the patients, we excluded these patients. (2) Patients with incomplete or poor-quality fNIRS data: To ensure the accuracy and reliability of the research results, we excluded patients with incomplete or poor-quality fNIRS data. These situations include missing signal integrity, a low signal-to-noise ratio, data consistency issues, correctable artifacts, and data that do not meet the requirements for standardized processing. We assessed data quality through visual inspection, automated quality control metrics, and expert evaluation to ensure that only high-quality data were included in the analysis, thereby enhancing the credibility and effectiveness of the research results. On the basis of these inclusion and exclusion criteria, a total of 275 patients were enrolled in the study. Among these cases, 91 cases were classified as mild to moderate impairment (MI), and 184 cases were classified as severe impairment (SI). The patients were randomly allocated into a training group (n = 193, with 63 having MI and 130 with SI) and a validation group (n = 82, with 28 having MI and 54 with SI) based on a 7:3 ratio. Figure 1 illustrates the study’s flow diagram.

FIGURE 1

Figure 1. Flow diagram of the study.

2.2 Data collection

The data collected for this study comprise two primary components. The initial component involves the retrieval and extraction of patients’ demographic characteristics from the electronic medical records system. These characteristics included age, sex, blood pressure at admission, duration poststroke, body mass index (BMI), presence of diabetes, hypertension, affected hemisphere, coronary heart disease, and Fugl-Meyer Assessment for Upper Extremity (FMA-UE) scores at admission. The second component pertains to the acquisition of 5-minute resting-state brain functional connectivity data via fNIRS. This includes the quantification of whole-brain functional connection edges (total edges of deoxyhemoglobin [HbD], total edges of oxyhemoglobin [HbO], total edges of total hemoglobin [HbT]) and the measurement of Short1 of edges (Short1 of edges for HbD/HbO/HbT), Short2 of edges (Short2 of edges for HbD/HbO/HbT), Long1 of edges (Long1 of edges for HbD/HbO/HbT), Long2 of edges (Long2 of edges for HbD/HbO/HbT), and the number of functional connection edges within 12 regions of interest (HbD/HbO/HbT in affected/unaffected primary motor cortex [PMC]/supplementary motor area [SMA]/dorsolateral prefrontal cortex [DLPFC]/temporal/frontopolar) and the number of connection edges between bilaterally symmetrical regions of interest (HbD/HbO/HbT in affected PMC/SMA/DLPFC/Temporal/Frontopolar-unaffected PMC/SMA/DLPFC/Temporal/Frontopolar), yielding a total of 69 fNIRS variables. Deoxyhemoglobin (HbD) and oxyhemoglobin (HbO) are key indicators of brain oxygenation and blood flow, respectively. Their measurements can provide insights into brain function and connectivity. Total hemoglobin (HbT) represents the total amount of hemoglobin in the blood, which can be used to assess overall cerebral blood volume.

Functional Near-Infrared Spectroscopy (fNIRS) data were acquired with the BS-2000 near-infrared spectroscopy system (Wuhan Zilian Hongkang Technology Co., Ltd., Wuhan, China). The instrument employs dual-wavelength laser diodes (690 nm and 830 nm) operating at a sampling rate of 10 Hz and is equipped with 32 sources and 32 detectors, forming 106 channels. Each channel consists of a source–detector pair separated by 3 cm. Optode placement followed the international 10–20 system, with source S2 positioned at Fpz. A 3-D digitizer (NirMap, Wuhan Zilian Hongkang Technology Co., Ltd.) was used to record the reference landmarks (Nz, Cz, AL, RL) and the spatial coordinates of all optodes. Channel locations were projected onto the cortical surface and assigned to the corresponding Brodmann areas using the NIRS-SPM approach (Liu et al., 2025).

2.3 Data analysis

The raw fNIRS data were preprocessed using the HOMER2 toolbox (version 2.8) integrated with MATLAB R2014b (MathWorks, Natick, MA, USA). The preprocessing steps included data conversion, segmentation, quality control, and signal processing. For detailed information on the preprocessing steps and parameter settings, including motion correction thresholds, please refer to the Supplementary Methods.

2.4 Model construction

The statistical analysis was conducted utilizing SPSS (IBM Corp, version 27.0) and R (version 4.2.1). This study undertook a comprehensive examination to identify predictors of upper limb motor dysfunction poststroke, employing fNIRS data. The normality of continuous variables was initially evaluated using the Kolmogorov-Smirnov test. The results indicated that none of the variables followed a normal distribution. Consequently, these variables were described using their median and interquartile range, and group comparisons were performed using the Mann-Whitney U test. Categorical variables are represented as frequencies (percentages) and were analyzed using chi-square tests, with statistical significance established at a p value of less than 0.05 (two-tailed).

The LASSO (Least Absolute Shrinkage and Selection Operator) regression model was used to identify variables correlated with poststroke upper limb motor dysfunction scores within the training dataset. LASSO regression is a statistical method that performs both variable selection and regularization to enhance the prediction accuracy and interpretability of the model. To determine the optimal penalty parameter (λ) for LASSO regression, we employed the method of K-fold cross-validation (commonly K = 10). In this process, the dataset was divided into K equal parts. K-1 parts were used for model training, and the remaining part was used for validating the model’s performance. This process is repeated K times, with a different part selected as the validation set each time, ensuring that every data point is used for validation once. Using this method, we assessed the model performance for different λ values and selected the λ value that minimizes the cross-validation error, thereby balancing the model’s bias and variance. After that, a logistic regression analysis was conducted, with poststroke upper limb motor dysfunction scores as the dependent variable and the variables selected by the LASSO regression as independent variables, to identify independent risk factors. To ascertain the robustness of the model, multicollinearity among the variables was evaluated through the application of Variance Inflation Factor (VIF) analysis.

A predictive nomogram was developed utilizing the independent risk factors identified through logistic regression analysis to estimate the likelihood of poststroke upper limb motor dysfunction. This nomogram functions as a visual tool for clinicians to assess the risk of motor dysfunction on the basis of individual patient characteristics. The calibration of the model was evaluated through calibration curve analysis to verify that the predicted probabilities aligned with the observed outcomes. Additionally, the clinical utility of the model was assessed using Decision Curve Analysis (DCA), which quantifies the net benefit by balancing potential harms against benefits. DCA is a method that evaluates the net benefit of a predictive model by comparing the benefits of true positive predictions against the harms of false positive predictions. The sensitivity, specificity, and negative predictive value are used to evaluate the performance of the model, whereas the Hosmer-Lemeshow test is used to detect the calibration of the model.

All analyses were conducted utilizing R version 4.2.1, which employs a range of specialized packages for distinct analytical tasks. The “glmnet” package was used for LASSO regression to identify predictive variables, whereas the “car” package facilitated VIF analysis to assess multicollinearity. The “rms” package was employed in the construction of the predictive nomogram, and “pROC” was used to calculate the C-index for evaluating model discrimination. Additionally, “PredictABEL” was utilized to assess improvements in predictive performance, including Integrated Discrimination Improvement (IDI) and Net Reclassification Improvement (NRI). The “rmda” package was applied for DCA to evaluate the clinical utility of the developed nomogram.

3 Results

3.1 Baseline characteristics

After applying rigorous inclusion and exclusion criteria, 275 patients were deemed eligible for study inclusion. The patients’ baseline characteristics are detailed in Table 1. The participants were subsequently divided into “(n=193, see Supplementary Table 2) and a validation group (n=82, see Supplementary Table 1).” The comparative baseline characteristics of these two groups are presented in Table 2. Statistical analyses revealed no significant differences in demographic variables, including sex, age, disease progression, medical history, or other relevant factors, between the training and validation groups (P > 0.05). This result highlights the methodological rigor of the randomization process, ensuring that the allocation of patients across groups was both scientifically valid and equitable.

TABLE 1

Table 1. Baseline characteristics of subjects.

TABLE 2

Table 2. The baseline characteristics of the patients in the training and validation cohort.

3.2 Nomogram construction

Figures 2A, B display the outcomes of the LASSO regression feature selection process. Figure 2A provides a visualization of the LASSO regression coefficient plot for all risk factors, with each curve representing a unique risk factor. The vertical axis indicates the regression coefficient of the predictor, whereas the horizontal axis represents the logarithm of λ. In contrast, Figure 2B illustrates the binomial deviance, where the lowest point on the curve signifies the optimal parameter λ. In this study, the application of LASSO regression resulted in the identification of five variables integral to the evaluation of poststroke motor dysfunction. These variables include Changes in HbD in the affected temporal region, Total HbT levels, HbT affecting the DLPFC, HbT levels in the unaffected frontopolar region, and HbT Short1 of the edges.

FIGURE 2

Figure 2. LASSO regression analysis with 10-fold cross-validation for predicting motor functional impairment post-stroke. (A) Use LASSO regression to make a map ofrisk factor coefficients. (B) Preliminary screening of clinical risk factors using LASSO regression.

A logistic regression analysis was performed to determine the weights and p-values of each predictive variable identified through LASSO regression, using a significance threshold of p < 0.05. As presented in Table 3, the variables HbT affected DLPFC and HbT Short1 of edges did not meet the significance criteria (P > 0.05) and were therefore excluded from further consideration. The exclusion of these variables was based on their lack of statistical significance, which suggests that they may not contribute substantially to the prediction of upper limb motor dysfunction scores following ischemic stroke. However, it is important to note that the choice of p < 0.05 is a common convention in medical research. Consequently, three independent risk factors associated with upper limb motor dysfunction scores following ischemic stroke were identified: Changes in HbD in the affected temporal region, Total HbT levels, and HbT levels in the unaffected frontopolar region, all of which satisfied the condition of p < 0.05 (refer to Table 4). On the basis of these three risk factors, a nomogram was developed for predictive purposes, as detailed in Figure 3.

TABLE 3

Table 3. Original model coefficients.

TABLE 4

Table 4. Final model coefficients.

FIGURE 3

Figure 3. The nomogram. This nomogram is based on three independent risk factors: HbD levels in the affected temporal region, total HbT levels, and HbT levels in the unaffected frontopolar region. To use the nomogram, follow these steps: (1) Locate the value of each risk factor on the corresponding axis (for example, HbD_affected_Temporal = 13, HbT_total = 500, HbT_unaffected_Frontopolar = 2). (2) Draw a vertical line upward from each value to the “Points” axis to determine the score for each factor. (3) Add up the scores for all three factors (for example, HbD score = 0, HbT_total score = 10, HbT_frontopolar score = 10; total score = 20). (4) Draw a vertical line upward from the total score to the “diagnose rate” axis to obtain the predicted probability of upper limb motor dysfunction (for example, the predicted probability corresponding to a total score of 20 is 0.004). Example: For a patient with HbD_affected_Temporal = 13, HbT_total = 500, and HbT_unaffected_Frontopolar = 2, the nomogram predicts a probability of upper limb motor dysfunction of 0.4%.

3.3 Model validation

To assess the discriminative ability of our model, we computed the area under the receiver operating characteristic (ROC) curve (AUC). As depicted in Figure 4A, the AUC for the training dataset was 0.766 (95% CI: 0.951–5.422). Figure 4B shows that the AUC for the testing dataset was 0.861 (95% CI: 0.950–3.89). For the training set, the sensitivity was 0.429, the specificity was 0.915, the negative predictive value was 0.768, and the p value for the Hosmer-Lemeshow test was 0.125. For the validation set, the sensitivity was 0.357, the specificity was 0.889, the negative predictive value was 0.727, and the Hosmer-Lemeshow test p-value was 0.211. Overall, the model exhibited satisfactory performance across both the training and validation datasets. It demonstrated high specificity, reflecting a strong ability to accurately identify non-target classes, and maintained a reasonable negative predictive value. The p values from the Hosmer-Lemeshow test exceeded 0.05 for both datasets, indicating a good fit of the model. Refer to Figure 5 for further details. Figure 6 illustrates the results of the DCA for the model. These results of the DCA demonstrated a substantial clinical net benefit in both the validation and development groups. This is observed across threshold probability ranges of 0.05–0.95 for the training dataset and 0.05–0.95 for the testing dataset (refer to Figure 6).

FIGURE 4

Figure 4. Receiver operating characteristic (ROC) curves for nomogram. (A) Training ROC. This graph illustrates the ROC curve of the model on the training dataset. The area under the curve (AUC) is 0.766 (95% confidence interval: 0.951–5.422), indicating a moderately high discriminative power of the model in distinguishing between the presence and absence of poststroke upper limb motor dysfunction. This suggests that the model can identify patients with motor dysfunction with relatively good accuracy. The x-axis represents the false positive rate (1 - specificity), and the y-axis represents the true positive rate (sensitivity). The diagonal line indicates a model with no discriminative ability (AUC = 0.5). The closer the ROC curve is to the top-left corner, the better the model’s performance. (B) Validation ROC. This graph displays the ROC curve of the model on the test dataset. The AUC is 0.861 (95% confidence interval: 0.950–3.89), demonstrating that the model maintains high discriminative power on new, unseen data. This result indicates that the model has good generalization ability and can effectively identify poststroke upper limb motor dysfunction in practical applications.

FIGURE 5

Figure 5. Calibration curves for the training cohort and the validation cohort. (A) Training cohort. The calibration curve for the training cohort shows the agreement between the predicted probabilities of upper limb motor dysfunction and the observed outcomes. The x-axis represents the predicted probability, and the y-axis represents the actual observed probability. The dashed diagonal line represents perfect calibration, where predicted probabilities perfectly match observed outcomes. The solid line represents the model’s calibration, with closer alignment to the diagonal indicating better calibration. (B) Validation cohort. The calibration curve for the validation cohort similarly compares predicted and observed probabilities. The x-axis and y-axis are labeled as in (A). The close alignment of the solid line to the diagonal indicates that the model is well-calibrated and generalizes well to new data.

FIGURE 6

Figure 6. Decision-curve analysis.

3.4 Correlation with ADL scores

We conducted a correlation analysis between the identified model variables and the activities of daily living (ADL) scores of the patients (refer to Table 5 Correlation Results between Model Variables and ADL). The results showed no significant correlation between the model variables and ADL scores (p > 0.05 for all variables). Specifically, the HbD levels in the affected temporal region (rs = −0.061, p = 0.315), the total HbT levels (rs = −0.022, p = 0.721), and the HbT levels in the unaffected frontopolar region (rs = 0.077, p = 0.202) did not exhibit significant associations with ADL scores. This suggests that while these variables are predictive of upper limb motor dysfunction severity, they may not directly influence the patients’ ability to perform daily living activities.

TABLE 5

Table 5. Correlation results between model variables and ADL.

4 Discussions

In this study, three independent risk factors were identified as being associated with the severity of upper limb motor dysfunction after ischemic stroke: (1) Changes in HbD in the affected temporal region, (2) Total HbT levels, and (3) HbT levels in the unaffected frontopolar region. Our model demonstrated good discriminative power in both the training and validation datasets, with AUC values reaching 0.766 and 0.861, respectively. This result underscores the efficacy and reliability of the model in clinical settings. Moreover, the nomogram we developed provides clinicians with an intuitive tool for objectively assessing the functional status of stroke patients, which can aid in devising personalized treatment plans and offering patients more precise diagnostic information.

However, it is important to note that our analysis of the correlation between model variables and ADL scores showed no significant correlation (with all variables having p > 0.05). This indicates that although the identified risk factors can be used to predict the severity of upper limb motor dysfunction, they may not be directly related to patients’ functional outcomes in daily activities. This finding highlights the complexity of post-stroke rehabilitation, where many other factors, in addition to hemodynamic changes in specific brain regions, may influence ADL performance. This phenomenon may be influenced by several factors: Firstly, the complexity of neuroplasticity itself cannot be ignored. After a stroke, the brain undergoes a highly complex process of neuroplasticity (Xi et al., 2023, Chen et al., 2023, Cramer et al., 2023), with dynamic changes occurring in multiple brain regions and neural networks (Yao et al., 2020, Liu et al., 2025). The model variables included in this study mainly reflect the correlation signals between brain regions related to function. These signal changes may not correspond completely with the improvement in patients’ motor function or daily living abilities as assessed by clinical scales (Arun et al., 2020, Huo et al., 2023). Secondly, there are significant individual differences among stroke patients. Factors such as lesion location, lesion size, and neural recovery potential vary from person to person, which may lead to inconsistencies in the relationship between brain functional connectivity and clinical scales (Liu et al., 2025).

Lastly, the limitations of measurement methods may also be a reason. Brain functional connectivity is measured using fNIRS technology to assess hemodynamic changes in the cerebral cortex, while the clinical scale (ADL) evaluates patients’ motor function and daily living abilities through behavioral assessment. The differences in sensitivity and specificity between these two measurement methods may result in a lack of significant correlation between them (Liu et al., 2025, Yang and Wang, 2025).

The HbD and HbT levels identified in this study as important predictive indicators hold significant neurophysiological importance. The changes in HbD levels in the affected temporal region may reflect the metabolic demands of local brain areas associated with motor function. Neural activity and energy metabolism in the brain are closely related to changes in HbD. For example, Highton’s article (Highton et al., 2012) indicates that deoxyhemoglobin serves as a biomarker of cerebral autoregulation. The total HbT level, on the other hand, reflects overall cerebral blood volume, which is crucial for maintaining normal neurological function. Additionally, the HbT level in the unaffected frontal pole region may be associated with compensatory mechanisms in the brain following a stroke. Research has shown that the frontal pole plays a role in motor coordination and recovery.

Regarding the calibration curves and decision curve analysis (DCA), our results indicate that the model is well-calibrated, with predicted probabilities closely aligning with observed outcomes. The DCA further supports the model’s clinical utility by demonstrating a favorable net benefit across a range of threshold probabilities (0.05–0.95). This suggests that the model could be clinically useful in guiding treatment decisions, particularly in identifying patients who may benefit from targeted rehabilitation programs. However, we recognize that the DCA results are based on a single-center dataset, and future studies should validate these findings in multi-center settings to ensure broader applicability.

To ensure the robustness of the model, multicollinearity among the variables was evaluated using Variance Inflation Factor (VIF) analysis. VIF analysis is a statistical method used to detect the presence of multicollinearity in a set of multiple regression variables. A VIF value greater than 10 is generally considered to indicate high multicollinearity, which can distort the regression coefficients and make the model less reliable. In our study, all VIF values were less than 10, indicating that multicollinearity was not a significant issue in our model. This ensures that the predictors included in the model are independent and contribute uniquely to the prediction of upper limb motor dysfunction scores.

Unlike Ye (Ye et al., 2024), who used a linear regression model combining sEMG and fNIRS signals, we employed LASSO regression and logistic regression analysis to identify biomarkers with higher discriminative power. Compared to Zhou (Zhou et al., 2024), who used machine learning models to predict recovery, we used statistical methods to identify risk factors. These differences highlight the innovation in our assessment methods and model construction, offering new perspectives and tools for stroke rehabilitation assessment.

To address potential concerns regarding the adequacy of our sample size, we conducted power analyses based on the AUC values obtained from both the training and validation datasets. The results indicated that our sample sizes were adequate to support the model’s discriminative ability, with power nearing 100% in both datasets. Specifically, in the training dataset, which comprised 130 positives and 63 negatives with an AUC of 0.766, the power was nearly 100%. Similarly, in the validation dataset, consisting of 54 positives and 28 negatives with an AUC of 0.861, the power was also close to 100%. These findings confirm that our chosen sample size was appropriate for detecting the expected effect sizes, ensuring robust and statistically significant results for both model development and validation.

Our findings reveal the key roles of the temporal lobe and the frontal pole in stroke recovery. Regarding the role of the temporal lobe in the recovery of motor function in stroke patients, studies by Kenzie (Kenzie et al., 2019) and Liu (Liu et al., 2021) have shown that the temporal lobe plays a significant role in the recovery of proprioception, whereas the frontal pole is crucial for improving motor coordination, especially in walking ability (Berger et al., 2019, Chatterjee et al., 2019). This is closely related to the recovery of motor function in stroke patients. This finding is also consistent with the results of our study. Our findings further support that changes in functional connectivity are not limited to between brain regions but also include within brain regions, reflecting that recovery after a stroke is a multifaceted process.

The current model has an AUC of 0.766 in the training set and 0.861 in the validation set, with sensitivity at 0.429 in the training set and 0.357 in the validation set, which indeed indicates potential for optimization. Considering the focus and current progress of this study, we did not attempt to use other machine learning algorithms (such as random forest and support vector machines) to optimize the model performance this time. However, we have acknowledged this as a limitation of our study in the discussion section and plan to explore these algorithms in future research to see if they can improve the model’s predictive accuracy and sensitivity. Meanwhile, to enhance the robustness of the model, we have used stricter cross-validation techniques to improve feature selection. In determining the optimal penalty parameter (λ) for LASSO regression, we employed a more rigorous repeated sampling 10-fold cross-validation method, which was repeated multiple times to ensure the stability of the selected λ value.

This study employed LASSO regression for variable selection because of its ability to reduce the risk of model overfitting and enhance the accuracy of predictions. However, this method may be sensitive to outliers and has a certain dependence on parameter settings. On the other hand, logistic regression analysis, on the other hand, provides us with a robust framework to assess the relationships between risk factors and ULMD-IS.

Although this study utilized MATLAB R2014b for data processing, the algorithms and methods employed remain universally applicable and reliable in subsequent versions. Furthermore, we will explore the use of updated software versions for data processing in future studies to verify the consistency and stability of the results, as well as to enhance the timeliness of our research.

Our model not only performs well statistically but also has practical clinical application value. By collecting resting-state fNIRS data from admitted patients and subsequently identifying the three key data points—HbD levels in the affected temporal region, HbT levels in the total region, and HbT levels in the unaffected frontopolar region—clinicians can more accurately and objectively differentiate stroke patients with upper limb motor dysfunction at different levels, formulate personalized diagnosis and rehabilitation plans, and potentially improve patients’ long-term outcomes.

Although our study provides valuable insights, it is not without limitations. For example, although the AUC value is acceptable, it is not particularly high, indicating that there is room for further optimization. One potential reason for this situation could be the insufficient sample size and reliance on data from a single source, which may undermine the model’s predictive accuracy and reliability. To increase the model’s universality and robustness, future research should expand the sample size and incorporate data from multiple research centers. Additionally, the limitations of the current study include its retrospective design, which may restrict the broad applicability of the results. The small sample size may affect the model’s stability and predictive power, and potential biases may also interfere with the interpretation of the results. To address issues such as sample size and single - source data, we propose the following plans for future research. First, for multi - center validation, we will collaborate with multiple research centers to collect a larger and more diverse dataset. This will not only increase the sample size but also enhance the representativeness of the data, thereby improving the model’s universality and robustness. Second, for model optimization, we will explore different machine - learning algorithms and techniques. For example, we may attempt neural network - based algorithms to see if we can further improve the model’s predictive accuracy. Additionally, we will conduct sensitivity analyses to better understand the impact of different variables on the model’s performance and carry out targeted optimizations. Regarding the clinical value, we plan to integrate the model into the clinical workflow. We will cooperate with developers of electronic medical record systems to create an interface that can automatically import relevant fNIRS data from patients’ records. This will enable clinicians to quickly obtain the model’s prediction results and incorporate them into their daily decision - making processes. For instance, when a stroke patient is admitted, the system can immediately calculate and assess the severity of upper limb motor dysfunction based on the collected fNIRS data, providing clinicians with timely and useful information for formulating treatment plans.

The results of this study may have a positive impact on patient management by providing more accurate diagnostic information, which can help patients and doctors make more effective treatment decisions. However, any decision based on predictive models should take into account individual differences and be used under the guidance of clinical judgment. Our findings provide novel insights into the assessment and management of upper limb motor dysfunction subsequent to ischemic stroke, thereby serving as valuable references for future research and clinical practice.

5 Conclusion

In this investigation, the application of LASSO regression and logistic regression analyses enabled the identification of three independent risk factors correlated with the severity of upper limb motor dysfunction subsequent to ischemic stroke: HbD levels in the affected temporal region, HbT levels in the total region, and HbT levels in the unaffected frontopolar region. A nomogram model was subsequently developed based on the basis of these identified factors. Although the model exhibited strong discriminative ability, with high specificity and an acceptable negative predictive value, its sensitivity was low. The model’s reliability and clinical net benefit were validated using the Hosmer-Lemeshow test and decision curve analysis, confirming its effectiveness in predicting upper limb motor function scores in stroke patients. In settings where high sensitivity is not critical, this model may assist clinicians in risk assessment and treatment planning and could, if used judiciously, contribute to modest improvements in patient outcomes and some reduction in long-term disability-related costs. It also lays the groundwork for future research, suggesting the need for model validation across different populations and exploring its integration into electronic medical records to enhance the clinical workflow and the model’s utility in everyday practice. Overall, our study not only advances the understanding of poststroke motor dysfunction but also provides a practical tool that could influence patient management and guide future research directions in the field of stroke rehabilitation.

Data availability statement

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

Ethics statement

The studies involving humans were approved by Ethics Committee of Tianjin Medical University General Hospital. The studies were conducted in accordance with the local legislation and institutional requirements. The ethics committee/institutional review board waived the requirement of written informed consent for participation from the participants or the participants’ legal guardians/next of kin because given its retrospective research design, the study was granted a waiver for obtaining informed consent from the participants.

Author contributions

ML: Conceptualization, Methodology, Software, Validation, Visualization, Writing – original draft, Writing – review & editing. CXW: Conceptualization, Methodology, Supervision, Project administration, Funding acquisition, Writing – review & editing. CYW: Data Curation, Formal analysis, Investigation, Resources, Writing – review & editing. XL: Conceptualization, Supervision, Project administration, Writing – review & editing. MS: Writing – review & editing.

Funding

The author(s) declare financial support was received for the research and/or publication of this article. This research was funded by the Tianjin Key Medical Discipline (Specialty) Construction Project (Grant No. TJYXZDXK-060B), the Key Discipline of Tianjin Health Science and Technology Project (Grant No. TJWJ2022XK007), and the General Project of Tianjin Multi-Investment Fund for Applied Basic Research (Grant No. 21JCYBJC01610).

Acknowledgments

We would like to express our gratitude to Qiu Shen and Ze Chen for their assistance during the submission process. We are also thankful to Yifan Bai, Lin Lv, and Zhenyan Zhang for their contributions to data processing. Lastly, we extend our appreciation to all the patients who participated in this study.

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 authors declare that no Generative AI was used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

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.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnhum.2025.1526455/full#supplementary-material

References

Almeida, S. R., Vicentini, J., Bonilha, L., De Campos, B. M., Casseb, R. F., and Min, L. L. (2017). Brain connectivity and functional recovery in patients with ischemic stroke. J. Neuroimag. 27, 65–70. doi: 10.1111/jon.12362

PubMed Abstract | Crossref Full Text | Google Scholar

Arun, K. M., Smitha, K. A., Sylaja, P. N., and Kesavadas, C. (2020). Identifying resting-state functional connectivity changes in the motor cortex using fNIRS during recovery from stroke. Brain Topogr. 33, 710–719. doi: 10.1007/s10548-020-00785-2

PubMed Abstract | Crossref Full Text | Google Scholar

Berger, A., Horst, F., Müller, S., Steinberg, F., and Doppelmayr, M. (2019). Current state and future prospects of EEG and fNIRS in robot-assisted gait rehabilitation: A brief review. Front. Hum. Neurosci. 13:172. doi: 10.3389/fnhum.2019.00172

PubMed Abstract | Crossref Full Text | Google Scholar

Cai, G., Xu, J., Zhang, C., Jiang, J., Chen, G., Chen, J., et al. (2024). Identifying biomarkers related to motor function in chronic stroke: A fNIRS and TMS study. CNS Neurosci. Ther. 30:e14889. doi: 10.1111/cns.14889

PubMed Abstract | Crossref Full Text | Google Scholar

Chatterjee, S. A., Fox, E. J., Daly, J. J., Rose, D. K., Wu, S. S., Christou, E. A., et al. (2019). Interpreting prefrontal recruitment during walking after stroke: Influence of individual differences in mobility and cognitive function. Front. Hum. Neurosci. 13:194. doi: 10.3389/fnhum.2019.00194

PubMed Abstract | Crossref Full Text | Google Scholar

Chen, N., Wang, Y. L., Sun, H. F., Wang, Z. Y., Zhang, Q., Fan, F. Y., et al. (2023). Potential regulatory effects of stem cell exosomes on inflammatory response in ischemic stroke treatment. World J. Stem Cells 15, 561–575. doi: 10.4252/wjsc.v15.i6.561

PubMed Abstract | Crossref Full Text | Google Scholar

Chiu, H. E., Hung, Y. C., Chang, K. C., Shih, C. C., Hung, J. W., Liu, C. W., et al. (2014). Favorable circulatory system outcomes as adjuvant traditional Chinese medicine (TCM) treatment for cerebrovascular diseases in Taiwan. PLoS One 9:e86351. doi: 10.1371/journal.pone.0086351

PubMed Abstract | Crossref Full Text | Google Scholar

Cramer, S. C., Lin, D. J., and Finklestein, S. P. (2023). Domain-Specific outcome measures in clinical trials of therapies promoting stroke recovery: A suggested blueprint. Stroke 54, e86–e90. doi: 10.1161/STROKEAHA.122.042313

PubMed Abstract | Crossref Full Text | Google Scholar

Eastmond, C., Subedi, A., De, S., and Intes, X. (2022). Deep learning in fNIRS: A review. Neurophotonics 9:041411. doi: 10.1117/1.NPh.9.4.041411

PubMed Abstract | Crossref Full Text | Google Scholar

Grefkes, C., and Fink, G. R. (2014). Connectivity-based approaches in stroke and recovery of function. Lancet Neurol. 13, 206–216. doi: 10.1016/S1474-4422(13)70264-3

PubMed Abstract | Crossref Full Text | Google Scholar

Grefkes, C., Nowak, D. A., Eickhoff, S. B., Dafotakis, M., Küst, J., Karbe, H., et al. (2008). Cortical connectivity after subcortical stroke assessed with functional magnetic resonance imaging. Ann. Neurol. 63, 236–246. doi: 10.1002/ana.21228

PubMed Abstract | Crossref Full Text | Google Scholar

He, Z., Wang, X., Wu, Y., Jia, J., Hu, Y., Yang, X., et al. (2014). Treadmill pre-training ameliorates brain edema in ischemic stroke via down-regulation of aquaporin-4: An MRI study in rats. PLoS One 9:e84602. doi: 10.1371/journal.pone.0084602

PubMed Abstract | Crossref Full Text | Google Scholar

Highton, D., Ghosh, A., Tachtsidis, I., Kolyva, C., Panovska, J., Elwell, C., et al. (2012). Deoxyhaemoglobin as a biomarker of cerebral autoregulation. Crit. Care 16(Suppl. 1):295. doi: 10.1186/cc10902

Crossref Full Text | Google Scholar

Huo, C., Xu, G., Sun, A., Xie, H., Hu, X., Li, W., et al. (2023). Cortical response induced by task-oriented training of the upper limb in subacute stroke patients as assessed by functional near-infrared spectroscopy. J. Biophoton. 16:e202200228. doi: 10.1002/jbio.202200228

PubMed Abstract | Crossref Full Text | Google Scholar

Huppert, T. J., Diamond, S. G., Franceschini, M. A., and Boas, D. A. (2009). HomER: A review of time-series analysis methods for near-infrared spectroscopy of the brain. Appl. Opt. 48, D280–D298. doi: 10.1364/ao.48.00d280

PubMed Abstract | Crossref Full Text | Google Scholar

Kenzie, J. M., Findlater, S. E., Pittman, D. J., Goodyear, B. G., and Dukelow, S. P. (2019). Errors in proprioceptive matching post-stroke are associated with impaired recruitment of parietal, supplementary motor, and temporal cortices. Brain Imag. Behav. 13, 1635–1649. doi: 10.1007/s11682-019-00149-w

PubMed Abstract | Crossref Full Text | Google Scholar

Kinder, K. T., Heim, H. L. R., Parker, J., Lowery, K., McCraw, A., Eddings, R. N., et al. (2022). Systematic review of fNIRS studies reveals inconsistent chromophore data reporting practices. Neurophotonics 9:040601. doi: 10.1117/1.NPh.9.4.040601

PubMed Abstract | Crossref Full Text | Google Scholar

Kongwattanakul, K., Hiengkaew, V., Jalayondeja, C., and Sawangdee, Y. (2020). A structural equation model of falls at home in individuals with chronic stroke, based on the international classification of function, disability, and health. PLoS One 15:e0231491. doi: 10.1371/journal.pone.0231491

PubMed Abstract | Crossref Full Text | Google Scholar

Li, X., Luo, M., Gong, Y., Xu, N., Huo, C., Xie, H., et al. (2022). Altered brain activity and effective connectivity within the nonsensory cortex during stimulation of a latent myofascial trigger point. Neural Plast 2022:4416672. doi: 10.1155/2022/4416672

PubMed Abstract | Crossref Full Text | Google Scholar

Liu, H., Jiang, Y., Wang, N., Yan, H., Chen, L., Gao, J., et al. (2021). Scalp acupuncture enhances local brain regions functional activities and functional connections between cerebral hemispheres in acute ischemic stroke patients. Anat. Rec. 304, 2538–2551. doi: 10.1002/ar.24746

PubMed Abstract | Crossref Full Text | Google Scholar

Liu, J., Qin, W., Zhang, J., Zhang, X., and Yu, C. (2015). Enhanced interhemispheric functional connectivity compensates for anatomical connection damages in subcortical stroke. Stroke 46, 1045–1051. doi: 10.1161/STROKEAHA.114.007044

PubMed Abstract | Crossref Full Text | Google Scholar

Liu, M., Wan, C., Wang, C., and Li, X. (2025). Effects of intermittent theta burst stimulation on upper limb motor recovery in early stroke patients: An fNIRS study. Front. Neurol. 16:1542827. doi: 10.3389/fneur.2025.1542827

PubMed Abstract | Crossref Full Text | Google Scholar

Longhi, M., Merlo, A., Prati, P., Giacobbi, M., and Mazzoli, D. (2016). Instrumental indices for upper limb function assessment in stroke patients: A validation study. J. Neuroeng. Rehabil. 13:52. doi: 10.1186/s12984-016-0163-4

PubMed Abstract | Crossref Full Text | Google Scholar

Pinti, P., Tachtsidis, I., Hamilton, A., Hirsch, J., Aichelburg, C., Gilbert, S., et al. (2020). The present and future use of functional near-infrared spectroscopy (fNIRS) for cognitive neuroscience. Ann. N. Y. Acad. Sci. 1464, 5–29. doi: 10.1111/nyas.13948

PubMed Abstract | Crossref Full Text | Google Scholar

Scholkmann, F., Kleiser, S., Metz, A. J., Zimmermann, R., Mata Pavia, J., Wolf, U., et al. (2014). A review on continuous wave functional near-infrared spectroscopy and imaging instrumentation and methodology. Neuroimage 85(Pt 1), 6–27. doi: 10.1016/j.neuroimage.2013.05.004

PubMed Abstract | Crossref Full Text | Google Scholar

Subramanian, S. K., Yamanaka, J., Chilingaryan, G., and Levin, M. F. (2010). Validity of movement pattern kinematics as measures of arm motor impairment poststroke. Stroke 41, 2303–2308. doi: 10.1161/STROKEAHA.110.593368

PubMed Abstract | Crossref Full Text | Google Scholar

Teo, W. P., Muthalib, M., Yamin, S., Hendy, A. M., Bramstedt, K., Kotsopoulos, E., et al. (2016). Does a combination of virtual reality, neuromodulation and neuroimaging provide a comprehensive platform for neurorehabilitation? - a narrative review of the literature. Front. Hum. Neurosci. 10:284. doi: 10.3389/fnhum.2016.00284

PubMed Abstract | Crossref Full Text | Google Scholar

Van Der Velden, L. L., Onneweer, B., Haarman, C. J. W., Benner, J. L., Roebroeck, M. E., Ribbers, G. M., et al. (2022). Development of a single device to quantify motor impairments of the elbow: Proof of concept. J. Neuroeng. Rehabil. 19:77. doi: 10.1186/s12984-022-01050-2

PubMed Abstract | Crossref Full Text | Google Scholar

Xi, X., Wang, H., Han, L., Ding, M., Li, J., Qiao, C., et al. (2023). Meta-analysis of repetitive transcranial magnetic stimulation combined with task-oriented training on upper limb function in stroke patients with hemiplegia. Medicine 102:e33771. doi: 10.1097/MD.0000000000033771

PubMed Abstract | Crossref Full Text | Google Scholar

Xie, H., Jing, J., Ma, Y., Song, Y., Yin, J., Xu, G., et al. (2022). Effects of simultaneous use of m-NMES and language training on brain functional connectivity in stroke patients with aphasia: A randomized controlled clinical trial. Front. Aging Neurosci. 14:965486. doi: 10.3389/fnagi.2022.965486

PubMed Abstract | Crossref Full Text | Google Scholar

Yang, L., and Wang, Z. (2025). Applications and advances of combined fMRI-fNIRs techniques in brain functional research. Front. Neurol. 16:1542075. doi: 10.3389/fneur.2025.1542075

PubMed Abstract | Crossref Full Text | Google Scholar

Yao, X., Cui, L., Wang, J., Feng, W., Bao, Y., and Xie, Q. (2020). Effects of transcranial direct current stimulation with virtual reality on upper limb function in patients with ischemic stroke: A randomized controlled trial. J. Neuroeng. Rehabil. 17:73. doi: 10.1186/s12984-020-00699-x

PubMed Abstract | Crossref Full Text | Google Scholar

Ye, S., Tao, L., Gong, S., Ma, Y., Wu, J., Li, W., et al. (2024). Upper limb motor assessment for stroke with force, muscle activation and interhemispheric balance indices based on sEMG and fNIRS. Front. Neurol. 15:1337230. doi: 10.3389/fneur.2024.1337230

PubMed Abstract | Crossref Full Text | Google Scholar

Zhou, Y., Xie, H., Li, X., Huang, W., Wu, X., Zhang, X., et al. (2024). Predicting upper limb motor recovery in subacute stroke patients via fNIRS-measured cerebral functional responses induced by robotic training. J. Neuroeng. Rehabil. 21:226. doi: 10.1186/s12984-024-01523-6

PubMed Abstract | Crossref Full Text | Google Scholar

Zhu, H., Xu, J., Li, J., Peng, H., Cai, T., Li, X., et al. (2017). Decreased functional connectivity and disrupted neural network in the prefrontal cortex of affective disorders: A resting-state fNIRS study. J. Affect. Disord. 221, 132–144. doi: 10.1016/j.jad.2017.06.024

PubMed Abstract | Crossref Full Text | Google Scholar