Application of artificial neural network for type 2 diabetes mellitus detection using buccal cell images

Diabetes mellitus (DM) is metabolic disease causing hyperglicemia due to insulin action anomali. DM can cause cellular changes, including buccal cell. Blood tests are used to diagnosis diabetes, so non-invasive test is required for diagnosis of diabetes. Accordingly, this research aims to design non-invasive system based on artificial neural network for type 2 DM detection using buccal cell images. Buccal cells smears were obtained from 30 subjects suffering from type 2 DM and 30 normal subjects. The smears were stained by using Papanicolaou method. Each slide were observed under digital microscope and were evaluated. The system was designed by using MATLAB with image processing and Probabilistic Neural Network (PNN) algorithm to classify features. Buccal cell images were segmented to get features. The features used in this study were nucleus area, nucleus perimeter and nucleus circularity. Nucleus areas and perimeters in type 2 DM group were higher than those in control group with similar nucleus roundness in both groups. Forty nucleus feature datasets were used for training process, while 20 nucleus feature datasets were used for testing process. The optimal PNN value was 0.4 g constant. The optimal accuracy of training was 92.5%, while the optimal accuracy of testing was 90%.