Convolutional Neural Network-based Diabetes Diagnostic System via Iridology Technique


Iridology is a sort of complementary medicine using the patterns, colors, and other properties of the iris to gather systemic information about a person’s health status. To put it another way, iridology is the examination of the iris’ sensitive structures. Iridology is employed in conjunction with other approaches by physicians to better understand their patients’ needs. Examination and diagnosis, on the other hand, are highly subjective and dependent on medical experience of the physicians. It’s also a time-consuming and exhausting process for physicians. This study proposed a hybrid method of deep learning and image processing for a more objective examination and diabetes diagnosis based on iris images. The suggested method initially detected the iris boundary and then automatically extracted the pancreatic region in the iridology chart. Image processing steps allowed for the detection of the pancreatic region on the iris and its automatic segmentation from the eye image. Afterward, diabetes was diagnosed using convolutional neural networks on images, and the results were compared with different convolutional neural network architectures. It was concluded that the proposed method combined with VGG-16 architecture and automatic segmentation of the pancreatic region resulted in an accuracy of 80%, a sensitivity of 100%, a precision of 71.42%, a specificity of 60%, and an f1 score performance of 83.33%.

Authors: Merve Nur Önal, Gür Emre Güraksin & Reşat Duman