The progression of artificial intelligence in healthcare is largely occurring in silos, and innovation occurs within specific medical domains rather than in all of them. However, applying technological breakthroughs from one field to another reveals some of the most promising advances.
This was precisely what was originally developed for behavioral health in the case of AI screening techniques, and now shows the prominent potential for assessment and monitoring of movement disorders.
Parallel challenges
Behavioral health and motor disorders share something fundamental. They manifest through observable patterns that are difficult to objectively quantify using traditional clinical evaluation methods. Both domains have historically relied on subjective rating measures, regular clinical observations, and patient self-reports – all worthwhile, but are essentially limited by the nature and potential variability of their episodes.
Get a Parkinson's disease assessment. This uses a unified Parkinson's Disease Rating Scale (UPDR) during rare clinical visits. Compare this with depression screening. This screening often uses a PHQ-9 survey at similar intervals. Both approaches capture only a short window of patient experience and may lack the important patterns that appear during appointments.
First success in behavioral health of AI
Behavioral health breakthroughs occurred when we began applying computer vision, natural language processing, and acoustic analysis to detect subtle patterns of patient expression, speech, and behavior that correlate with patient expression, speech, and conditions such as depression, anxiety, and cognitive impairment. By analyzing facial microexpression, speech modulation, linguistic patterns, and even delayed responses, AI systems are now able to detect indicators of mental health with sensitivity that climb or exceed traditional screening methods.
Studies from recent studies have demonstrated a promising application of AI-driven assessment in the detection of depression in adults. The hybrid deep learning model combining text and audio features achieved 98% accuracy for audio-based depression detection and 92% accuracy for text-based detection in adult participants. These results show that audio CNNs are a good model for depression detection and perform better than text CNN models alone.
The ability of techniques to analyse modalities of multiple data, including speech patterns, language functions, and behavioral markers, provides a comprehensive approach to mental health screening that may enhance traditional clinical assessments.
The logical extension of motor disorders
The leap towards applying these techniques to motor disorders is logical and scientifically sound. Conditions such as Parkinson's disease, essential tremor, Huntington's disease, and various dystonias exhibit observable motor symptoms, such as being equipped with unique equipment for quantification by AI systems.
Recent advances in computer vision and machine learning demonstrate the potential for early Parkinson's disease detection. Speech biomarker analysis achieved 92% accuracy with SVM and 94% with a random forest algorithm in distinguishing patients from healthy individuals. For neuroimaging, nonlinear kernel SVM achieved a detection rate of 96.14% using SPECT data, and another study reported an accuracy of 97.86%. These AI approaches analyze multiple biomarkers simultaneously and significantly outperform traditional clinical evaluations at early stages of disease.
Beyond detection: Continuous monitoring and outcome measurement
The true value of this cross-application extends beyond initial screening. The same AI system that can detect behavioral health can now be reused to continuously monitor the progression of motor disorders and treatment response.
Consider patients with Parkinson's disease using a smartphone application to analyze.
- Fine motor control during routine smartphone use
- Expression during a video call
- Speech modulation and speech patterns during conversation
- Walking and balance irregularities using accelerometer data
This continuous data collection creates a longitudinal profile that is much more detailed than regular clinical assessments alone. For pharmaceutical companies developing new therapies, this represents an unprecedented opportunity to measure drug efficacy through objective, continuous data rather than relying solely on subjective patient reporting and frequent clinical measurements.
Clinical trials and impact on drug development
For pharmaceutical researchers and clinical trial sponsors, the technology offers several transformative benefits.
- Enhanced endpoint measurement:AI-driven continuous ratings provide a more sensitive measure of treatment response than traditional rating scales.
- Previous Efficacy Signals: Subtle improvements in exercise patterns may be detectable via AI before they become apparent on clinical rating scales and may shorten the duration of the trial.
- Reducing sample size requirements: Using more sensitive measurement tools can reduce recruitment challenges as fewer participants may achieve statistically significant results.
- Remote distributed trial: Participants can be monitored continuously from home, expanding geographical reach and reducing dropout rates.
A recent phase 3 trial incorporates a smartphone application as an exploratory endpoint along with traditional MDS-ddrs scoring, with participants completing finger tapping, walking, and cognitive tests (both at home and intra-clinical). In another example, Roche developed a smartphone application that incorporates motor tasks such as finger tapping that can distinguish individuals with Parkinson's disease from those without, and included it as an exploratory endpoint in a phase 2 clinical trial. In a six-month Phase 1B clinical trial, participants completed six daily motor active tests (continuous vocalization, resting tremor, postural tremor, finger tapping, balance, and walking) via smartphone, demonstrating that smartphone sensor technology provides reliable, effective, clinically meaningful, highly sensitive phenotypic data in Parkinson's disease.
Implementation challenges and ethical considerations
Despite that promise, this cross-approach approach faces several hurdles.
- Regulation uncertainty: The FDA has begun developing a framework for AI/ML as a medical device, but the regulatory pathway for new digital biomarkers remains complex.
- Data Privacy: Continuous surveillance raises important questions about patient privacy and data security that need to be addressed through robust safeguards.
- Clinical verification: New digital biomarkers need to be thoroughly tested against gold standard measurements to ensure reliability and clinical relevance.
- Health Equity: Access to smartphones and broadband internet can vary widely across demographic groups and create disparities for those who benefit from these technologies.
Future landscape
The convergence of behavioral health and movement disorder assessment techniques is just the beginning of a broader trend towards a unified digital biomarker platform. As more sophisticated AI systems are developed, further cross-pollination between medical domains that have traditionally been operating independently may be seen.
For pharmaceutical companies and clinical researchers, early adoption of these cross-application technologies offers significant competitive advantages in drug development efficiency, cost reduction, and the ability to demonstrate treatment efficacy through new and more sensitive endpoints.
Conclusion
Cross-application of AI screening technology from behavioral health to movement disorders exemplifies ways to promote innovation by breaking silos between healthcare specialties. Recognizing the fundamental similarities of how these seemingly different conditions are measured and monitored, opens up new pathways for the detection, treatment, and understanding of complex neurological and psychiatric disorders.
For those involved in drug development and clinical research, this new approach offers not only gradual improvements, but also potentially transformative changes in the way disease assessment and measurement of treatment response are conceptualized. Companies and researchers who recognize and utilize these connections will be at the forefront of the next wave of therapeutic innovation.
Brett Talbot is the co-founder and CCO of Videra Health, a leading AI-driven mental health assessment platform. Talbot is a renowned clinical psychologist, innovator and respected figure in the behavioral health community. Prior to Videra Health, Talbot was the Chief Clinical Officer and Executive Director of several prestigious healthcare organizations. His pioneering efforts have led to the creation of pioneering video-based clinical assessments of depression, anxiety and trauma.
