Video-Based Data-Driven Models for Diagnosing Movement Disorders: Review and Future Directions.

Area of Science:

• Neurology
• Computer Science
• Biomedical Engineering

Background:

• Movement disorders significantly impact quality of life.
• Current diagnosis relies on subjective clinical assessments.
• Clinical videos offer objective data for analysis.

Purpose of the Study:

• To comprehensively review video-based, data-driven models for movement disorders.
• To examine machine learning (ML) and deep learning (DL) applications in this field.
• To identify trends, limitations, and future directions for AI in movement disorder analysis.

Main Methods:

• Literature review of studies from 2006-2024 across scientific databases.
• Analysis of various video modalities (RGB, depth, marker-based, multi-perspective, multimodal).
• Focus on pose estimation, real-time methods, and end-to-end DL architectures.

Main Results:

• Significant trend towards pose estimation methods in recent studies.
• Increasing usability and performance of AI models, nearing expert levels.
• Identification of limitations: data scarcity, lack of standardized metrics, privacy concerns.

Conclusions:

• Video-based AI models are a promising advancement for movement disorder diagnosis and assessment.
• Future research should focus on explainable AI, privacy-preserving techniques, and standardized metrics.
• Continued development can enhance clinical workflows and patient care.