Deep Learning based Gait Abnormality Detection using Wearable Sensor System.

Area of Science:

• Biomedical Engineering
• Machine Learning in Healthcare
• Human Movement Analysis

Background:

• Human gait is a complex function susceptible to abnormalities, increasing fall risk.
• Early diagnosis and treatment of gait disorders are crucial for patient outcomes.
• Machine learning offers advanced capabilities for analyzing biomedical data compared to traditional methods.

Purpose of the Study:

• To present a wearable sensor system integrated with a stacked Long short-term memory (LSTM) model for detecting gait abnormalities.
• To assess the system's effectiveness in identifying gait patterns associated with fall risk.
• To demonstrate the application of advanced technology in gait diagnosis and treatment assistance.

Main Methods:

• Development of a wearable sensor system combining plantar pressure measurement and Inertial Measurement Units (IMUs).
• Integration of a stacked Long short-term memory (LSTM) model for analyzing sensor data.
• Simulation and validation of the model using three specific gait abnormalities: Hemiplegic, Parkinsonian, and Sensory-Ataxic gaits.

Main Results:

• The computed gait metrics and parameters showed significant differences between normal and abnormal gait patterns.
• The proposed model demonstrated promising results in detecting simulated gait abnormalities.
• The system effectively identified distinct features of Hemiplegic, Parkinsonian, and Sensory-Ataxic gaits.

Conclusions:

• The integrated wearable sensor and LSTM model system shows potential for accurate gait abnormality detection.
• This technology can serve as a valuable tool for gait diagnosis and as a component in treatment assistant systems.
• Advanced technological integration is key to improving the diagnosis and management of human gait disorders.