Neural network-based diagnosis of type 2 diabetes using an iridology approach

Abstract

The growing global occurrence of type 2 diabetes requires the development of non-invasive and effective diagnostic methods. This work proposes a novel approach to detecting type 2 diabetes using iridology and machine learning (ML) techniques. By analyzing the iris of the right eye, a single region of interest (ROI) corresponding to the head of the pancreas is recognized for feature extraction. A total of 112 statistical and texture features are extracted using gray-level co-occurrence matrix (GLCM) and discrete wavelet transform (DWT) algorithms. Five neural network (NN) models, narrow, medium, wide, bi-layered, and tri-layered are deployed to classify healthy and diabetic people. The models are trained and assessed using a range of k-fold values (2 to 20) to optimize performance. The highest classification accuracy of 83.2% was reached using the narrow neural network (NNN) model at 7-fold cross-validation. This work exhibts the potential of iridology-based ML approaches for non-invasive diabetes diagnosis, providing a promising substitute to traditional blood tests.