Noninvasive diabetes mellitus detection using facial block color with a sparse representation classifier

Area of Science:

• Medical imaging
• Biomedical engineering
• Computational diagnostics

Background:

• Diabetes mellitus (DM) is a growing global epidemic, posing significant healthcare challenges.
• Current diagnostic methods for DM can be invasive or burdensome.
• There is a need for accessible, noninvasive diagnostic tools for early DM detection.

Purpose of the Study:

• To propose and evaluate a novel noninvasive method for diabetes mellitus detection.
• To utilize facial block color features and sparse representation classification (SRC) for DM diagnosis.
• To assess the accuracy and efficacy of the proposed facial-based detection system.

Main Methods:

• A noninvasive capture device was used to acquire facial images, segmented into four facial blocks.
• Facial color features were extracted from each block using six centroids from a facial color gamut.
• A sparse representation classifier (SRC) with healthy and DM subdictionaries was employed to classify individuals.

Main Results:

• The system analyzed a dataset of 142 healthy and 284 diabetes mellitus samples.
• The SRC model, utilizing combined facial blocks, achieved an average accuracy of 97.54%.
• The method demonstrated high efficacy in distinguishing between healthy and DM classes.

Conclusions:

• Facial block color features combined with SRC offer a promising noninvasive approach for diabetes mellitus detection.
• This method presents a potential low-cost, accessible tool for widespread DM screening.
• Further research could validate this technique in diverse populations and clinical settings.