Multimodal Gait Recognition for Neurodegenerative Diseases.

Area of Science:

• Biomedical Engineering
• Neurology
• Computer Science

Background:

• Single modality gait recognition has limitations in capturing complex gait disturbances.
• Gait analysis is crucial for diagnosing and evaluating neurodegenerative diseases.
• Multimodality analysis offers a more comprehensive approach to gait pattern recognition.

Purpose of the Study:

• To develop a novel hybrid model for gait recognition using multi-sensor data fusion.
• To accurately differentiate between three neurodegenerative diseases.
• To classify varying severity levels of Parkinson's disease and distinguish patients from healthy individuals.

Main Methods:

• A hybrid model integrating a spatial feature extractor (SFE) and a correlative memory neural network (CorrMNN).
• Fusion and aggregation of data from multiple sensors to capture both spatial and temporal gait features.
• Implementation of a multiswitch discriminator for state estimation and classification.

Main Results:

• The proposed hybrid model achieved higher classification accuracy compared to state-of-the-art techniques.
• Demonstrated effectiveness in identifying gait differences across various neurodegenerative conditions.
• Successfully distinguished between different Parkinson's disease severity levels.

Conclusions:

• Multimodality gait analysis using the proposed hybrid model significantly enhances diagnostic accuracy for neurodegenerative diseases.
• The novel SFE and CorrMNN architecture effectively extracts discriminative spatial and temporal gait features.
• This approach offers a promising tool for objective clinical assessment and disease monitoring.