AI Hand Pose Recognition for Parkinson's Dyskinesia Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current systems for detecting dyskinesias in Parkinson's disease patients face challenges due to variability in observer diagnostics and the inability to automatically quantify hand pose recognition, especially with background and illumination issues during video capturing.
Innovation Solution
The implementation of AI technology using deep learning for hand segmentation followed by a 3D convolutional neural network (CNN) classification, which removes background and illumination effects to accurately classify hand poses such as tremors and pronation-supination movements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual observation and diagnosis by qualified observers are used for UPDRS evaluation, then diagnostic expertise and interpretation accuracy are improved, but observer availability is limited and inter/intra observer variability increases
Solution Approach 1:
The system enables automatic self-diagnosis through AI-based hand pose recognition and dyskinesia detection, eliminating the need for qualified observers while maintaining diagnostic accuracy. The computer automatically evaluates video footage and provides UPDRS assessment without human intervention.
Solution Approach 2:
The patent replaces the mechanical system of manual observation and human diagnosis with an automated computer vision system using deep learning and 3D CNN algorithms. This substitution eliminates observer variability and makes the diagnostic system universally available without requiring specialized medical personnel.
2Adaptability or versatility
If automatic detection systems are implemented to improve accessibility and reduce variability, then observer availability and consistency are improved, but background and illumination issues reduce measurement precision
Solution Approach 1:
The system extracts and isolates the hand region from the background using image segmentation techniques. By separating the hand from the complex background and illumination variations, the system maintains high measurement precision while achieving universal accessibility through automatic detection.
Solution Approach 2:
The patent transforms the input data by applying image processing techniques that adjust for illumination variations and background differences. The system changes the parameters of the input images through normalization and segmentation, enabling consistent hand pose recognition across diverse recording conditions.
3Loss of information
If the entire capture range including background is analyzed, then context information is preserved, but background objects and movements increase complexity and reduce detection precision
Solution Approach 1:
The system segments the hand region from the background using deep learning-based image segmentation. This division allows the system to focus computational resources on analyzing hand movements while excluding irrelevant background objects and movements, thereby improving dyskinesia detection precision without losing essential hand-related context information.
Data Source
AI summary
There is included an apparatus and system including image segmentation code, configured to cause at least one hardware processor to segment an image of a person's hand from an input image, and classification code configured to cause the at least one processor to classify the segmented image of the person's hand according to at least one predefined pose.