Using AI Systems to Optimize the Clinical Outcome of Stroke Patients

Not Applicable

Not yet recruiting

• Conditions: Acute Stroke Intervention; Acute Stroke; Acute Ischemic Stroke

• Registration Number: NCT06828679
• Lead Sponsor: Chinese University of Hong Kong

Brief Summary

This project addresses the imminent challenge of providing adequate motor rehabilitation to a growing number of stroke survivors amidst the ageing population, decreasing age of stroke, and shortage of physical/occupational therapists in Hong Kong through AI and precision rehabilitation. To reduce the socioeconomic burden from the stroke survivors' loss of independence and their care (\>HK$15 billion/year), the efficacy of rehabilitation and efficiency of its delivery must be improved. These goals can be achieved by prescribing them with individually tailored rehabilitations predicted to yield maximal functional return. Defining a predictive model for such personalization remains challenging given the immense heterogeneity of stroke. The investigators aim to build an explainable AI system that predicts a subject's recovery potential and the treatment option that may realize this potential based on multi-modal pre-rehab assessments. Data from clinical, neuroimaging, neurophysiological, and multi-omic evaluations will be collected from stroke survivors (N≥400) before they undergo upper limb rehab with usual care, neuromuscular stimulation, robotic training, or acupuncture. Machine learning-extracted data features will be used to train decision-tree and neural-network AI algorithms for robust predictions. As soon as the model is validated, the investigators will deploy it to implement a personalized rehab program in the community. Our model's ability to predict the optimal intervention from a wide spectrum of input modalities distinguishes ours from previous less-than-accurate models. Our interdisciplinary team of 13 PIs with expertise in neurology, PT/OT, acupuncture, electrical/biomed. engineering, robotics, neuroscience, neuroimaging, multi-omics, data science, and clinical trial management will put us in a world-unique position to execute this project successfully and generate opportunities of interdisciplinary education. In the long run, our prediction system will accelerate marketization of new rehab strategies by facilitating their clinical-trial evaluations in more targeted subjects, thereby leading Hong Kong to be a future global hub of innovative rehabilitation.

Detailed Description

DELIVERABLE 1: Overview of Data Collection Plan. Subacute stroke survivors (N=400) will be recruited and randomly assigned to one of four groups, which will receive usual rehab care only, and usual care plus acupuncture, robotic, or neuromuscular electrical stimulation (NMES) training, respectively. Each subject will be evaluated before (A0), after (A1) and 6 months after the start of intervention (A2). At A0, thorough clinical, neurophysiological (EEG, EMG, TMS), MRI, and blood-based assessments will be made for deriving predictive recovery markers. At A1 and A2, only the clinical scores and EMG will be assessed for characterizing post-rehab functional gain and long-term recovery. Also, in all groups clinical scores will be recorded midway through intervention (A½) to monitor treatment progress. Treatments will last ≥1 month with ≥20 training sessions. All A0, A1, and A2 data will be analyzed offline with machine learning and other methods, and then used to train the AI model, PRAISE-HK (Precision Rehabilitation AI System for Enhancing recovery in Hong Kong and beyond).

DELIVERABLE 2: Explainable AI Models of Recovery. Overview. Our AI system receives high-dimensional data inputs from diverse modalities, each of which requires unique processing techniques tailored to the specific characteristics of the data type for analysis. Neuroimaging data, for instance, demand specialized image processing algorithms for deriving brain structural and functional measures, while time series data such as EEG and EMG must first be analyzed with signal processing tools that capture embedded patterns across varying temporal resolutions. The need for domain-specific preprocessing of diverse data types engenders a level of analytic complexity that defies standard deep learning techniques, which may not accommodate the unique challenges presented by each modality. To obtain actionable, clinically relevant insights while ensuring interpretability and integration of the multimodal data, more sophisticated algorithms beyond standard deep learning are needed.

As such, the investigators propose the construction of a sophisticated AI fusion model as the core of PRAISE-HK. Fusion model is a machine learning approach that integrates input features from multiple data modalities in a way that leverages their unique contributions to make a prediction more accurate and robust than any single modality could produce. Although the number of subjects here (N=400) is, to the best of our knowledge, the highest among all similar studies, for high-dimensional data this number is small enough to impose rather stringent constraints on the choice of analytical methods should overfitting be avoided, and model generalization ensured. With these considerations, the investigators will construct our fusion model in two phases. Phase 1 involves extraction of predictors from the individual modalities through their independent processing with domain-specific AI models. This phase is necessary since each modality can yield the most valuable information only with distinct processing pipelines. Phase 2 involves the fusion of these processed modalities. The investigators will formulate a specialized fusion model that accommodate the relatively small sample size, the predictors' high dimensionality, and potential instances of missing data. The final model will capitalize on the strengths of the domain-specific phase-1 outputs and combine their insights for improved prediction accuracy.

Phase 1: Modality-specific Feature Extractions. For the neuroimaging and multi-omics modalities, the investigators will leverage pre-trained models to map high-dimensional inputs to a reduced set of informative features or embeddings. The investigators will start by fine-tuning foundation models based on each dataset, and then develop modality-specific models using the pre-trained and fine-tuned foundation models as feature extractors. For time-series data such as EMG, EEG, and motion capture data, both standard and custom-derived procedures will be used to filter noise and identify suitable predictors. From the EMG, non-negative matrix factorization and our recently proposed rectified latent variable model will be used to extract muscle synergy indices, which, from our preliminary results (see below), contain predictive stroke recovery information. From the EEG and EMG, coherence between these two signals (cortico- muscular coherence) and that between EEG and muscle synergy activations (cortico-synergy coherence) will be computed through spectral analysis. From the motion capture data, movement parameters such as joint motion range can be extracted with standard methods. All parameters above will be used as predictors in phase 2.

Phase 2: AI Fusion Model. Once the investigators have extracted the predictors from each modality, the investigators will develop a fusion model, one for each treatment group and each clinical score, that integrates these A0 predictor inputs for a prediction of the A1 and A2 scores . Since the investigators do not have prior information on the suitability of the different methods to be used in fusion models, the investigators will evaluate the performance of various models. Artificial neural networks with meta-learning will be used for their capacity to learn from a rich set of data features. Their performance will be benchmarked against other models such as logistic regression, Bayesian methods, and random forests, methods that are advantageous for smaller datasets such as ours. For each treatment group, the fusion model will be trained on the phase-1 predictors with the labelled data from the 100 subjects as inputs to predict the score improvements at A1 and A2.

Realistically, the investigators anticipate that our final database will have occasional missing data points. The investigators will employ mutual information-based imputation techniques to estimate missing heterogenous values, thereby making full use of the available data and preventing biases in model training. Sensitivity analyses will be conducted to assess the impact of the missing data on model predictions, thereby ensuring the reliability of our findings.

Through the entire system construction process, the investigators will utilize and compare different AI algorithms for phase-1 feature extractions and phase-2 fusion. The models and approaches employed will be iteratively refined based on test performance metrics and clinical feedback. Historical data will be utilized to validate the models through retrospective assessment of the effectiveness of the recommended interventions in previous clinical scenarios.

Additional Phase 3: Building Decision-making Framework. For every subject, by comparing the phase-2 outcome predictions across the treatment-specific models, one may already decide on an ideal intervention for the subject simply by selecting the treatment whose model yields the best A2 improvement. But this selection process is not transparent, because the fusion models do not explicitly indicate how the many predictors interact to lead the models to collectively arrive at the recommended treatment. To enhance the AI system's explanability, the investigators will implement a phase 3 for constructing an explicit decision-making framework that guides intervention selection, one analogous to the clinically successful PREP decision-tree algorithm of Stinear et al. A classifier will be trained with the phase-1 predictors and the phase-2 score predictions of all subjects (N=400) serving as inputs, and the ranked intervention preferences inferred from phase-2 outcome comparison as outputs. The trained classifier can serve as a decision support tool that reveals the most parsimonious set of evaluations needed for reaching decisions, and provides transparent reasoning behind its recommendations, thereby enhancing our understanding of why particular subjects are assigned to each treatment. To maximize interpretability, the investigators will implement decision-tree algorithms such as the random forests, which are capable of discerning complex, nonlinear relationships among the variables.

Study Timeline

• Study First Submitted: 2024-12-13
• Study First Submitted QC: 2025-02-11
• Study First Posted: 2025-02-14
• Status Verified: 2025-04
• Last Update Submitted: 2025-04-17
• Last Update Posted: 2025-04-18
• Study Start: 2025-05-01
• Primary Completion: 2027-12-31
• Study Completion: 2028-06-30

Recruitment & Eligibility

• Status: NOT_YET_RECRUITING
• Sex: All
• Target Recruitment: 400

Inclusion Criteria

• Age 65-80
• 1-6 months after onset of a first-time unilateral stroke rostral to midbrain
• Moderate-to-severe motor impairment of one upper limb (Fugl-Meyer Assessment for Upper Extremity of 10-50 out of 66);
• Able to provide written informed consent;
• Detectable electromyographic (EMG) activities in flexor digitorum-flexor carpi radialis and extensor digitorum-extensor carpi ulnaris muscle groups, with EMG from each muscle group exceeding 3 standard deviations above baseline mean. This last criterion is essential for successful NMES training

Exclusion Criteria

• Unconscious or bed-bound;
• Uncontrollable diabetes;
• Anticipated non-adherence to treatment schedule;
• On cardiac pacemaker;
• Other severe comorbidities (heart/kidney failure, deranged liver function).

Study & Design

• Study Type: INTERVENTIONAL
• Study Design: PARALLEL

Primary Outcome Measures

Name	Time	Method
Surface EMG Recordings	From Pre-assessment stages (A0) to follow up sessions (A2), the whole time frame will be within 6 months	Surface EMG electrodes may be conveniently used to record myoelectric activities of multiple muscles during voluntary movement. This data can reveal the muscle coordination patterns utilized by the motor system for control. We have previously shown that muscle synergies, building blocks of muscle coordination decomposed from EMG using machine learning, may serve as markers of post-stroke functional assessment and prognosis. Here, we will record EMG from 16 affected-side muscles during 8-10 activities of daily living.
Kinematic recordings	From Pre-assessment stages (A0) to follow up sessions (A2), the whole time frame will be within 6 months	We will utilize motion capture technology to analyze stroke patients' movements by attaching reflective markers to their bodies. This setup will allow us to record their performance during 8 to 10 activities of daily living. The collected data will enable us to evaluate joint angles and movement patterns, providing insights into functional capabilities and informing rehabilitation strategies.
Cortico-muscular Coherence (CMC) from Electroencephalography (EEG) and EMG	From Pre-assessment stages (A0) to follow up sessions (A2), the whole time frame will be within 6 months	The CMC is a measure, derived from concurrent EEG and EMG during movement, that quantifies the overall capacity of the cortex to coordinate activities of multiple muscles. After stroke, the CMC between different brain regions and muscles are altered. Given that the structural connectome may predict recovery potential, it is likely that CMC should carry some predictive power. We will record EEG (64 channels) during the EMG assessment described above. The CMC between each EEG channel and each muscle may serve as inputs to the prediction system.

Secondary Outcome Measures

Name	Time	Method
Fugl-Meyer Assessment for Upper Extremity	From Pre-assessment stages (A0) to follow up sessions (A2), the whole time frame will be within 6 months	The Fugl-Meyer Assessment (FMA) for Upper Extremity is a standardized test used to evaluate motor function, sensation, and joint function in individuals with neurological conditions, particularly after a stroke. It assesses voluntary movements of the shoulder, elbow, wrist, and fingers, as well as sensory capabilities. Scoring ranges from 0 (no function) to 66 (full function), helping to identify impairments and monitor recovery progress in rehabilitation settings.
Modified Tardieu Scale	From Pre-assessment stages (A0) to follow up sessions (A2), the whole time frame will be within 6 months	The Modified Tardieu Scale is a clinical tool used to assess spasticity in individuals with neurological conditions. It measures the resistance of a muscle to passive stretch at different velocities, providing insights into the severity of spasticity. The scale includes observations of muscle reaction during slow and fast stretches, rating the response from 0 (no resistance) to 4 (very strong resistance). This assessment helps guide treatment and rehabilitation strategies for managing spasticity.
National Institutes of Health Stroke Scale	From Pre-assessment stages (A0) to follow up sessions (A2), the whole time frame will be within 6 months	The National Institutes of Health Stroke Scale (NIHSS) is a clinical assessment tool used to evaluate the severity of stroke symptoms. It consists of 15 items that assess functions such as consciousness, vision, movement, speech, and sensation. Each item is scored, with higher total scores indicating more severe impairment. The NIHSS is essential for diagnosing stroke, guiding treatment decisions, and predicting patient outcomes.
Brunnstrom Stages	From Pre-assessment stages (A0) to follow up sessions (A2), the whole time frame will be within 6 months	The Brunnstrom Stages are a framework for understanding the progression of motor recovery after a stroke. Developed by Signe Brunnstrom, the stages range from 1 to 7, describing the evolution from flaccidity (Stage 1) to the re-emergence of coordinated movement and normal function (Stage 7). Each stage reflects specific patterns of muscle tone and voluntary movement, helping clinicians plan rehabilitation strategies and track recovery progress.
Wolf Motor Function Test	From Pre-assessment stages (A0) to follow up sessions (A2), the whole time frame will be within 6 months	The Wolf Motor Function Test (WMFT) is a standardized assessment used to evaluate upper extremity motor function in individuals after a stroke or neurological injury. It measures the speed and quality of performing various tasks involving hand and arm movements, such as reaching, grasping, and manipulation. The test includes timed tasks and a functional score, providing insights into motor recovery and guiding rehabilitation interventions.
Barthel Index	From Pre-assessment stages (A0) to follow up sessions (A2), the whole time frame will be within 6 months	The Barthel Index is a measure of an individual's ability to perform activities of daily living (ADLs) independently. It assesses ten specific tasks, including feeding, bathing, dressing, and mobility, assigning scores that reflect the level of assistance required. The total score ranges from 0 (completely dependent) to 100 (fully independent), making it a valuable tool for evaluating functional status and progress in rehabilitation, particularly after stroke or other disabilities.

Trial Locations

Locations (1)

The Chinese University of Hong Kong; 🇭🇰 Hong Kong, Sha Tin, Hong Kong