Dense & Attention Convolutional Neural Networks for Toe Walking Recognition.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Pediatric Gait Analysis

Background:

• Idiopathic toe walking (ITW) is a common pediatric gait disorder characterized by limited heel strike during walking.
• ITW can lead to secondary complications including balance issues, pain, and impaired physical development.
• Early detection of ITW is crucial for timely and effective intervention strategies.

Purpose of the Study:

• To develop and evaluate a novel deep learning architecture for accurate detection of idiopathic toe walking.
• To improve upon existing methods for ITW identification using advanced convolutional neural network features.
• To provide a scalable and generalizable AI solution for diagnosing ITW.

Main Methods:

• A one-dimensional Dense & Attention convolutional network (DANet) architecture was proposed.
• Dense blocks were integrated for enhanced feature transfer, and attention modules were used to prioritize relevant features.
• The Focal Loss function was modified to address data imbalance issues in the dataset.

Main Results:

• The DANet achieved a test recall of 88.91% for ITW detection on a local dataset.
• Validation on public datasets yielded an average precision of 89.34%, recall of 91.50%, and F1-Score of 92.04%.
• The proposed DANet demonstrated superior performance compared to other existing methods.

Conclusions:

• The DANet model presents a valid and feasible approach for the automated detection of idiopathic toe walking.
• The AI-driven method shows significant potential for improving early diagnosis and management of ITW in children.
• The study highlights the efficacy of integrating dense blocks and attention mechanisms in deep learning for gait analysis.