Automated Prediction of Diabetes Mellitus using Infrared Thermal Foot Images: Recurrent Neural Network Approach

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Medical Imaging Analysis

Background:

• Diabetes Mellitus (DM) affects millions globally, necessitating regular monitoring for complication prevention.
• Current blood glucose monitoring methods are invasive and uncomfortable for patients.
• Noninvasive monitoring techniques are crucial for consistent DM management and improved patient outcomes.

Purpose of the Study:

• To develop an automated, noninvasive system for predicting Diabetes Mellitus (DM) using thermal foot images.
• To leverage Recurrent Neural Network (RNN) and Convolutional Neural Network (CNN) for accurate DM prediction.
• To offer a more comfortable and consistent alternative to traditional invasive glucose monitoring methods.

Main Methods:

• Acquisition of thermal foot images from diabetic and non-diabetic participants using a FLIR E-60 thermal camera.
• Feature extraction from thermal images utilizing a Convolutional Neural Network (CNN).
• Prediction of DM presence using a Recurrent Neural Network (RNN) fed with extracted features.

Main Results:

• The proposed automated system achieved a high prediction accuracy of (97.14 ± 1.5) % for Diabetes Mellitus.
• This performance significantly outperformed the light-weight Convolutional Neural Network (Lw-CNN) method, which attained (82.9 ± 3) % accuracy.
• The developed framework demonstrated superior predictive capabilities compared to existing state-of-the-art methods.

Conclusions:

• The automated thermal foot imaging system shows significant potential as a noninvasive tool for Diabetes Mellitus prediction.
• This approach can enhance patient outcomes through timely intervention and personalized care.
• Future research should involve larger datasets and integration with clinical decision support systems for broader application.