Diabetic foot ulcer classification using mapped binary patterns and convolutional neural networks.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision
• Diabetology

Background:

• Diabetic foot ulcer (DFU) is a severe diabetes complication, often leading to amputation.
• Automated classification of DFU using machine learning shows promise.
• Current methods primarily rely on RGB images with convolutional neural networks (CNNs).

Purpose of the Study:

• To develop and evaluate a CNN-based DFU classification method incorporating texture information.
• To investigate the performance enhancement by combining RGB images with texture features.
• To improve the accuracy and effectiveness of automated DFU detection.

Main Methods:

• A two-stage approach was proposed: texture feature extraction using mapped binary patterns and subsequent CNN classification.
• Texture information was extracted from RGB images using the mapped binary patterns technique.
• The CNN model received stacked RGB and texture images or a fused image as input.

Main Results:

• The proposed method, utilizing both RGB and texture features, outperformed existing CNN-based approaches.
• Achieved high performance metrics: 0.981 AUC and 0.952 F-Measure on the Part-A DFU dataset.
• Demonstrated strong results on Part-B datasets: 0.995 AUC (ischaemia) and 0.820 AUC (infection).

Conclusions:

• Integrating texture features with RGB images significantly enhances CNN-based DFU classification.
• The proposed method offers a more accurate and robust tool for automated DFU diagnosis.
• This approach holds potential for earlier and more precise detection, potentially reducing amputation rates.