Efficiently Identifying Non-FoG, Pre-FoG, Pre-FoG Transition, and FoG in Parkinson's Disease Patients Using Window...

Area of Science:

• Neurology
• Biomedical Engineering
• Data Science

Background:

• Freezing of gait (FoG) is a debilitating walking disorder in Parkinson's disease (PD).
• Accurate identification of pre-FoG stages is critical for timely intervention but challenging with current methods.
• Existing sliding window techniques lack sensitivity to PD severity and disease progression.

Purpose of the Study:

• To develop and validate a novel algorithm for dynamic label assignment in gait data to improve FoG and pre-FoG detection.
• To assess the impact of overlapping windows and overlap rates on classification performance.
• To identify key features for efficient and accurate FoG stage classification.

Main Methods:

• Utilized accelerometer data from 10 high-risk PD participants, collected from ankle, thigh, and trunk sensors.
• Implemented a novel algorithm for dynamic label assignment based on gait data characteristics.
• Employed overlapping windows for data augmentation and performed sensitivity analysis on overlap rates.

Main Results:

• Achieved 89% sensitivity and 92% specificity for identifying all FoG stages.
• Demonstrated 96% sensitivity and 88% specificity for detecting the pre-FoG state (2 seconds prior to FoG).
• Effective classification was achieved using only 25 key features, with low overfitting risk below 25% overlap.

Conclusions:

• Dynamic label assignment is crucial for accurate classification of FoG and pre-FoG stages in PD.
• The proposed algorithm offers a sensitive and specific method for detecting FoG episodes and their precursors.
• The study identified essential features for FoG detection, reducing computational load and potential overfitting.