Machine learning algorithms can analyze complex sensor data to provide objective measures of spasticity, addressing the fundamental limitations of traditional assessment methods. Recent research by Yee and colleagues has led to a comprehensive sensor-based system that integrates wireless wearable sensors, including sophisticated goniometers, myometers, and surface electromyography sensors, to create a holistic assessment framework. Their innovative approach extracted 18 distinct features from comprehensive sensor data and used a combined logical support vector machine (SVM)-random forest (RF) approach that achieved an impressive 91% accuracy in predicting Modified Ashworth Scale scores, substantially outperforming individual classifiers and demonstrating the power of ensemble learning approaches.

Artificial neural networks have shown particular promise in learning complex patterns from biomechanical data, with the ability to capture subtle relationships that may not be apparent through traditional analytical approaches. Park and colleagues demonstrated that neural networks could be trained to mimic spasticity assessment performed by human clinicians, achieving satisfactory agreement with multiple expert raters while providing consistent, reproducible evaluations that eliminate inter-rater variability concerns.

With sophisticated algorithms trained to detect abnormal muscular activity patterns that characterize spasticity, AI technologies could analyze electromyographic signals during gait or upper extremity movements, whether passive or active. Implementing such tools is technically feasible, considering that semi-automated electrocardiogram analysis has been successfully used for several years, while recent studies in electroencephalogram analysis have shown that AI-based evaluation equals or exceeds human expert evaluation capabilities. Because most currently available data focus on gait and lower extremity movement patterns, applications in gait analysis appear most feasible for immediate implementation, offering the potential for semi-automated gait analysis while enabling longitudinal tracking of patient modifications over time.

Machine learning applications in gait analysis include sophisticated joint moment prediction from kinematic data, representing a significant advancement in biomechanical assessment capabilities. Using video-based, markerless assessment associated with AI-based moment prediction, clinicians could obtain comprehensive gait analysis using portable devices, resulting in substantial time savings and significant cost reductions compared with traditional laboratory-based approaches. However, several technical challenges currently prevent commercial implementation, particularly difficulties in accurately predicting hip moments from video-based kinematic data, although ongoing research continues to address these limitations.

Patient movements could be evaluated, rated automatically, and compared with a dataset from healthy subjects. Again, there are currently more data for gait than upper limb movements. One domain in which AI-based assessment is relatively advanced is passive and active range of motion (ROM) measurement. With automated AI-based optical systems, measuring hand movements (discriminating each finger and metacarpophalangeal joints and interphalangeal joints) in less than a second is possible. As for tremor analysis, deep learning systems could assist in evaluating patterned movements to classify and/or identify modifications in spasticity.

Applying machine learning to electromyographic signal analysis has yielded promising results in distinguishing between voluntary and involuntary muscle contractions. Lonini and colleagues developed a sophisticated system using skin-mounted, flexible wireless sensors to collect electromyographic and motion data from individuals with stroke, computing 15 distinct features from electromyographic data and using linear discriminant analysis to distinguish between voluntary muscle contractions and spastic contractions with remarkable accuracy, achieving an area under the curve of 0.94. In another recent paper, Merlo and colleagues demonstrated in 10 individuals with central nervous system lesions and complete paralysis of the upper extremity that an automated algorithm had the potential to detect involuntary muscle activity in the elbow flexors, combining accelerometer and electromyographic (EMG) signals in all patients.

Computer vision technologies have fundamentally changed movement analysis by enabling comprehensive, markerless motion capture and the automated assessment of movement patterns, without the constraints of traditional marker-based systems. A systematic review by Lam and colleagues examined 65 studies on markerless motion capture technology for clinical measurement in rehabilitation settings, identifying Microsoft Kinect as the most frequently used system. However, recent trends demonstrate increasing adoption of smartphone-based motion analysis systems because of their enhanced accessibility and portability.

Computer vision applications in spasticity assessment focus on the automated analysis of movement patterns during passive and active movements, using sophisticated algorithms that can detect subtle changes in movement quality, timing, and coordination that may not be apparent to human observers. Current algorithms can analyze video data, extracting precise kinematic parameters, including joint angles, velocities, and accelerations, and provide objective measures of movement characteristics associated with spasticity, while eliminating subjective interpretation variability. Recent advancements in deep learning have significantly improved the capabilities of automated video recording for movement analysis. Neural networks can now precisely identify specific movement patterns associated with spasticity. These systems can process video data in real time, providing clinicians and patients with immediate feedback.

Muscle modifications in patients with spasticity are primarily characterized by fibrofatty infiltration of muscles alongside fascial thickening and progressive muscle atrophy, representing essential markers of disease progression and treatment response. These modifications can be clinically observed using ultrasound imaging, with evaluation currently performed through qualitative assessment of muscle echo intensity using the modified Heckmatt scale. AI technologies, combined with the appropriate ultrasound settings, can facilitate the automated assessment of muscle characteristics and the identification of regions of interest within muscle tissue.

Recent research has demonstrated that AI-powered software can successfully automate muscle thickness and echo intensity measurements in healthy subjects and patients hospitalized in intensive care units, taking 99.8% less time than human experts while maintaining similar consistency and accuracy levels. Relatively spared structures within muscle tissue, such as hypoechoic pouches, could theoretically be automatically identified and highlighted in AI-enhanced ultrasound systems, potentially assisting clinicians in identifying optimal targets for botulinum toxin injection procedures.

Natural language processing (NLP) represents an emerging and particularly promising application of AI in spasticity assessment, with specific relevance to goal setting, clinical documentation, and comprehensive outcome evaluation. These sophisticated technologies can analyze unstructured clinical text to extract meaningful information, standardize terminology across different providers and settings, and provide valuable assistance in clinical decision-making processes.

The application of NLP to goal setting addresses one of the most challenging aspects of patient-centered care delivery. NLP systems can analyze patient interviews and read comprehensive clinical notes and historical documentation to automatically extract patient (or caregiver) priorities and translate them into structured treatment goals aligned with established frameworks. These systems may assist in formatting goals according to SMART criteria, transforming patient expressions into objectives that facilitate effective treatment planning and outcome measurement.

This represents one of the most clinically relevant AI-based tools for spasticity assessment in the immediate future, particularly as clinical practice shifts from impairment-based evaluation relying primarily on tone scales such as the Modified Ashworth Scale and Tardieu Scale toward goal-based approaches that guide comprehensive spasticity treatment. Patient experience, individual needs, and personal preferences are critical determinants in choosing treatment (indication and modalities) and outcome selection, representing a fundamental shift toward patient-centered care delivery.

With NLP technologies, it is possible to automate the analysis of clinical notes; patient and caregiver experiences can be captured through audio recordings during clinical visits, and multidimensional data from electronic medical records can be transcribed and analyzed. These systems can extract priorities and assist clinicians in translating them into relevant treatment goals while ensuring consistency and comprehensiveness. This capability has been evaluated and tested in health care situations sharing similar backgrounds to rehabilitation medicine, including trained NLP systems capable of detecting goals of care relevant to end-of-life decisions in palliative care medicine through analysis of electronic medical records and clinical notes.

Another NLP system was able to accurately extract headache frequency from clinical notes, showing that even nonstructured written notes could be examined, decoded, and evaluated automatically.

NLP could also help format goals in SMART criteria, translating a nonstructured goal into a structured and measurable one (eg, “I want to reduce pain in my shoulder” into “The goal is to reduce pain at the shoulder at rest of at least 40% at 4 weeks after botulinum toxin injection”).

NLP could also review electronic patients’ documentation, identify response patterns to previous treatments, and suggest goal modifications.

Another aspect that could be improved in spasticity assessment using NLP is consistency in documentation. Because of time constraints and variability among more and less experienced clinicians, NLP could ensure proper terminology use and standardize patient descriptions.

The clinician could dictate or type notes during their assessment, and the NLP system directly processes this input. The system could identify key clinical findings and the patient’s expectations and goals directly during the clinical examination and history. The system could also suggest appropriate outcome measures for a specific goal (eg, a 6-minute walking test if the goal is to improve walking endurance).

NLP could also facilitate goal formatting according to SMART criteria, transforming nonstructured goals into structured and measurable objectives while reviewing electronic patient documentation to identify response patterns to previous treatments and suggest appropriate goal modifications. Another significant aspect that could be improved through NLP implementation is consistency in clinical documentation, addressing time constraints and variability among clinicians with different experience levels.

For over a decade, wearable sensor technologies have offered the potential for continuous monitoring of patients in real-world environments. Wearables could provide clinicians with unique insights into the variability and progression of spasticity symptoms across different contexts and periods, revealing modifications in a patient’s activities. Modern wearable sensors can simultaneously capture multiple physiologic and biomechanical parameters relevant to comprehensive spasticity assessment, including detailed movement patterns, muscle activation characteristics, and tremor characteristics with high temporal resolution. ^,

Recent advances in flexible, skin-mounted sensors have effectively overcome many practical limitations of traditional wearable devices by conforming to body contours and offering comfortable, long-term monitoring without disrupting daily activities. More advanced AI algorithms for wearable sensors will likely be able to identify specific patterns linked to spastic muscle contractions, distinguishing them from voluntary movements in real time.

The continuous monitoring capabilities enable detection of patterns and trends that would not be apparent from intermittent clinical assessments, including circadian variations in spasticity severity, individual responses to medication timing, and the impact of specific activities on symptom expression. This comprehensive information proves invaluable for optimizing treatment timing and dosing while facilitating personalized care approaches that account for individual variation in symptom presentation.

One of the biggest challenges in complex evaluations is combining vast amounts of data from different sources into a coherent picture. AI can provide the capability to integrate diverse data points of comprehensive assessment, combining traditional clinical scales, sophisticated biomechanical measurements, patient-reported outcomes, and continuous data from wearable devices. This integration can provide a more holistic view of individual patients than any single assessment method while leveraging the capability to handle large datasets to reveal trends and associations that remain invisible through traditional single-evaluation approaches.

Digital twin technology is the most advanced use of AI in personalized medicine, creating detailed virtual models of individual patients for treatment simulation, outcome prediction, and personalized care planning. Recent research by Mikołajewska and colleagues emphasized the potential of AI-based digital patient twins in rehabilitation, allowing customized treatment plans by simulating various rehabilitation scenarios with predictive modeling capabilities.

Digital twins in spasticity management can simulate the effects of other treatment methods before clinical use, potentially enhancing treatment choices while reducing trial-and-error approaches. Their predictive abilities go beyond immediate treatment results to advanced long-term outcome modeling, analyzing patterns from similar patients while including individual traits to forecast the chances of reaching specific functional goals with different treatment options.

AI-powered decision support systems provide clinicians with evidence-based recommendations for treatment selection, optimal dosing, and timing of interventions based on a comprehensive analysis of patient-specific data combined with large databases of treatment outcomes. These sophisticated systems analyze individual patient characteristics in conjunction with extensive outcome databases to generate personalized recommendations for comprehensive spasticity management, accounting for individual variation and treatment response patterns.

The success of AI implementation depends heavily on intuitive user interface design that integrates seamlessly with existing clinical workflows while minimizing disruption to established practices. Effective artificial intelligence systems should present information in formats familiar to clinicians while providing real-time feedback during clinical assessments that enhances rather than replaces clinical decision-making capabilities.

The integration of AI recommendations with clinical judgment represents a critical consideration. AI medical systems are designed to support rather than replace clinical decision making while providing evidence-based guidance that enhances clinical expertise. Comprehensive training programs are essential for successful implementation, addressing both the technical aspects of system operation and the clinical interpretation of AI-generated results, while ensuring appropriate integration with existing clinical knowledge and experience.