Flatfeet Severity-Level Detection Based on Alignment Measuring.

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Orthopedics

Background:

• Flat foot is a common postural deformity affecting foot-ground contact.
• X-ray radiographs offer an affordable alternative to specialized devices for flat foot detection.
• Automated detection can improve diagnostic efficiency and accessibility.

Purpose of the Study:

• To develop an automated model for detecting flat foot and its severity from lateral foot X-ray images.
• To measure and analyze three critical foot angles: Arch Angle, Meary's Angle, and Calcaneal Inclination Angle.
• To compare the performance of Convolutional Neural Network (CNN) and Random Forest (RF) models for this task.

Main Methods:

• Template Matching was employed to identify potential anatomical landmarks for angle measurement.
• A classifier was used to select the most likely points for accurate angle calculation.
• Two models, CNN and Random Forest, were trained on 8000 images and tested on 240 cases, incorporating transfer learning (VGG-16) and data augmentation.

Main Results:

• The Random Forest model achieved the highest overall accuracy of 93.13%.
• The Random Forest model demonstrated strong performance metrics: 93.38% precision, 92.56% recall, 96.46% specificity, 95.42% accuracy, and 92.90% F-Score.
• Analysis of three foot angles provided more accurate estimations compared to single-angle measurements.

Conclusions:

• Transfer learning (VGG-16) as a feature extractor, combined with image augmentation, significantly enhances model accuracy.
• Utilizing multiple foot angle measurements (Arch, Meary's, Calcaneal Inclination) improves the precision of flat foot detection and severity assessment.
• The developed automated model shows promise for efficient and accurate clinical diagnosis of flat feet using X-ray images.