Gait-Based Diplegia Classification Using LSMT Networks.

Area of Science:

• Biomedical Engineering
• Clinical Biomechanics
• Computational Neuroscience

Background:

• Diplegia, a subtype of cerebral palsy (CP), presents diverse motor challenges.
• Current classification relies on expert interpretation of gait patterns.
• Automated gait analysis offers potential for objective and efficient diplegia subtyping.

Purpose of the Study:

• To develop and evaluate deep learning models for automated classification of diplegia into four clinical forms.
• To compare the performance of multilayer perceptron (MLP) and recurrent neural networks (RNNs) for this task.
• To assess the accuracy of AI-driven classification against expert clinical judgment.

Main Methods:

• Utilized a dataset of gait data from 174 diplegic patients, collected via an optoelectronic system.
• Extracted 27 angular parameters from gait measurements to represent walking patterns.
• Trained and tested MLP and RNN deep learning models for automated classification.

Main Results:

• Both MLP and RNN models demonstrated capability in classifying diplegia subtypes.
• The deep learning classification accuracy was comparable to expert assessments for 3 out of the 4 clinical forms.
• This indicates a high potential for automated diplegia classification using gait analysis.

Conclusions:

• Deep learning models, specifically MLPs and RNNs, can effectively automate the classification of diplegia into distinct clinical forms.
• Automated gait analysis shows promise in supporting clinical diagnosis and potentially improving patient management for cerebral palsy.
• Further research can refine these models for broader clinical application in movement disorder assessment.