Accessible, At-Home Detection of Parkinson's Disease via Multi-task Video Analysis

2024-06-21

AAAI Conference on Artificial Intelligence

AI Summary

Early Parkinson’s disease (PD) diagnosis is hindered by limited clinical resources and the modality limitation of existing AI methods—typically relying on single-modal data (e.g., motor or speech)—which fail to capture PD’s multidimensional manifestations. To address this, we construct a large-scale, multi-task video dataset comprising 1,102 subjects performing finger-tapping, facial-expression, and speech tasks. We propose UFNet, an uncertainty-calibrated fusion network enabling contactless, at-home screening using only a standard camera and microphone. Key contributions include: (i) a novel task-uncertainty-guided dynamic weighting mechanism for self-attention-based feature fusion; (ii) uncertainty-driven automatic rejection, achieving high sensitivity (79.3±0.9%) and specificity (92.6±0.3%); and (iii) the first demonstration of cross-gender and cross-ethnicity fairness in multi-task video-based PD analysis. UFNet attains 88.0±0.3% accuracy, 93.0±0.2% AUROC, with only 2.3±0.3% of samples rejected due to high predictive uncertainty.

Technology Category

• Computer Vision: Video Understanding & Activity Analysis
• Machine Learning: Calibration & Uncertainty Quantification
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Economics, Online Markets and Human Computation: Fairness and ethical considerations in crowd work and in human-in-the-loop AI systems
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Limited accessibility to neurological care leads to under-diagnosed Parkinson's Disease (PD), preventing early intervention. Existing AI-based PD detection methods primarily focus on unimodal analysis of motor or speech tasks, overlooking the multifaceted nature of the disease. To address this, we introduce a large-scale, multi-task video dataset of 1102 sessions (each containing videos of finger tapping, facial expression, and speech tasks captured via webcam) from 845 participants (272 with PD). We propose a novel Uncertainty-calibrated Fusion Network (UFNet) that leverages this multimodal data to enhance diagnostic accuracy. UFNet employs independent task-specific networks, trained with Monte Carlo Dropout for uncertainty quantification, followed by self-attended fusion of features, with attention weights dynamically adjusted based on task-specific uncertainties. We randomly split the participants into training (60%), validation (20%), and test (20%) sets to ensure patient-centered evaluation. UFNet significantly outperformed single-task models in terms of accuracy, area under the ROC curve (AUROC), and sensitivity while maintaining non-inferior specificity. Withholding uncertain predictions further boosted the performance, achieving 88.0 +- 0.3% accuracy, 93.0 +- 0.2% AUROC, 79.3 +- 0.9% sensitivity, and 92.6 +- 0.3% specificity, at the expense of not being able to predict for 2.3 +- 0.3% data (+- denotes 95% confidence interval). Further analysis suggests that the trained model does not exhibit any detectable bias across sex and ethnic subgroups and is most effective for individuals aged between 50 and 80. By merely requiring a webcam and microphone, our approach facilitates accessible home-based PD screening, especially in regions with limited healthcare resources.

Problem

• Limited access to neurological care causes underdiagnosed Parkinson's Disease
• Existing AI methods overlook multifaceted nature of Parkinson's Disease
• Need for accurate, accessible home-based Parkinson's Disease screening

Innovation

• Multi-task video dataset for Parkinson's detection
• Uncertainty-calibrated Fusion Network (UFNet) model
• Self-attended fusion with dynamic uncertainty adjustment