Using Expert-Cited Features to Detect Leg Dystonia in Cerebral Palsy.

Area of Science:

• Neurology
• Biomedical Engineering
• Pediatrics

Background:

• Leg dystonia in cerebral palsy (CP) is a significant challenge, often underdiagnosed due to limitations in routine clinical evaluations.
• Current diagnostic accuracy for leg dystonia via standard clinical assessment is only 12%, necessitating improved detection methods.

Purpose of the Study:

• To determine if quantifiable, expert-cited features of leg dystonia could be used to train machine learning (ML) models for improved detection in children with CP.
• To develop an objective tool for diagnosing leg dystonia in pediatric CP patients using video analysis.

Main Methods:

• 298 videos of children with CP performing a seated task were assessed by eight pediatric movement disorder physicians.
• Expert-identified leg dystonia features were quantified from videos to train 4,664 ML models.
• The best performing ML models were tested on a separate dataset of 135 videos.

Main Results:

• 69 quantifiable features were identified, corresponding to 12 expert-cited leg dystonia features.
• ML models achieved 82% accuracy, 88% sensitivity, and 74% specificity in identifying leg dystonia.
• The best models identified leg movement variability as a key diagnostic indicator, with 17 out of 25 features quantifying this aspect.

Conclusions:

• The developed ML models, integrated into the open-source software DxTonia, demonstrate an 82% accuracy rate for detecting leg dystonia in children with CP.
• DxTonia significantly outperforms routine clinical diagnosis (12% accuracy) by leveraging the quantification of leg movement variability.
• Quantifying leg movement variability during simple tasks offers a promising approach to enhance leg dystonia detection in pediatric CP.