Thermal Change Index-Based Diabetic Foot Thermogram Image Classification Using Machine Learning Techniques.

Area of Science:

• Medical imaging
• Machine learning
• Diabetology

Background:

• Diabetes mellitus (DM) poses significant risks, including plantar ulcers and amputation.
• Infrared thermography can detect temperature variations linked to diabetic foot ulceration risk.
• Machine learning (ML) shows promise for early diagnosis of diabetic foot complications.

Purpose of the Study:

• To explore ML approaches for classifying diabetic foot thermograms.
• To identify the most effective ML model for early detection of diabetic foot complications using thermal imaging.
• To categorize thermograms based on the thermal change index (TCI) for risk stratification.

Main Methods:

• Utilized a publicly available dataset of plantar foot thermograms.
• Classified images into distinct categories using the thermal change index (TCI), validated by experts.
• Investigated classical ML algorithms with feature engineering and Convolutional Neural Networks (CNNs) with image enhancement.
• Evaluated multilayer perceptron (MLP) classifier performance.

Main Results:

• The multilayer perceptron (MLP) classifier, combined with extracted thermogram features, achieved 90.1% accuracy in multi-class classification.
• This performance surpassed previously reported metrics on the same dataset.
• Feature engineering with classical ML demonstrated strong classification capabilities.

Conclusions:

• Machine learning, particularly the MLP classifier with feature engineering, is effective for classifying diabetic foot thermograms.
• This approach shows significant potential for the early diagnosis and management of diabetic foot complications.
• Accurate classification of thermograms can aid in preventing severe outcomes like ulcers and amputations.