Freezing of Gait Detection and Prediction Based on a Dilated Temporal Encoder Approach.

Area of Science:

• Neurology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Freezing of Gait (FoG) is a major Parkinson's Disease symptom impacting mobility and fall risk.
• Current detection methods often lack predictive capabilities for timely intervention.

Purpose of the Study:

• To develop and validate a novel framework for accurate detection and prediction of Freezing of Gait (FoG) using wearable sensors.
• To optimize the prediction window for anticipatory interventions in Parkinson's Disease patients.

Main Methods:

• Utilized wearable inertial measurement unit (IMU) sensors from 40 Parkinson's patients.
• Developed a Dilated Temporal Encoder (DTE) integrating Bidirectional Dilated Temporal Convolutional Networks (Bi-DTCN) and attention mechanisms.
• Incorporated an auxiliary loss for the Pre-FoG segment and analyzed varying segment lengths (2-5 seconds).

Main Results:

• Achieved a high F1 score of 91.33 ± 0.98% for patient-dependent FoG prediction with a 5-second anticipatory window.
• Patient-independent prediction showed lower performance (F1 score 62.90 ± 1.88%), indicating generalization challenges.
• A performance-based clustering strategy improved cross-patient performance to 59.62 ± 1.92% by identifying patient subgroups.

Conclusions:

• The proposed DTE framework demonstrates significant potential for accurate and anticipatory FoG detection in Parkinson's Disease.
• A 5-second prediction window offers practical clinical utility for proactive management of FoG.
• Combining individualized modeling with subgroup generalization strategies is crucial for improving patient-independent FoG prediction and enhancing patient safety.