Segmentation Approaches for Diabetic Foot Disorders.

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Diabetic Care

Background:

• Diabetic foot disorders are a significant cause of morbidity.
• Thermography provides a non-invasive method for early detection and monitoring.
• Accurate foot sole segmentation is crucial for reliable thermographic analysis.

Purpose of the Study:

• To implement and evaluate conventional and Deep Learning segmentation methods for diabetic foot sole analysis.
• To optimize segmentation using multimodal imaging data (visual-light, infrared, depth).
• To determine the most robust and accurate segmentation approach for clinical application.

Main Methods:

• Acquisition of multimodal images (visual-light, infrared, depth) from 37 healthy subjects.
• Implementation and comparison of various segmentation techniques, including conventional and Deep Learning (U-Net).
• Optimization of segmentation using depth image spatial information and manual ground truth establishment.

Main Results:

• All evaluated segmentation methods achieved satisfactory performance.
• Skin and U-Net approaches, optimized with depth data, showed superior spatial overlap, accuracy, and precision.
• The U-Net approach demonstrated preferred robustness for clinical implementation.

Conclusions:

• Deep Learning methods, particularly U-Net, show significant promise for automated foot sole segmentation in thermography.
• Optimizing segmentation with depth image data enhances accuracy and reliability.
• Robust segmentation techniques are essential for integrating thermography into standard diabetic care protocols for preventing foot disorders.