Electronic nose for detecting multilevel diabetes using optimized deep neural network

Diabetes is a chronic disease which is still a major issue in the world. The common testing methods generally used to detect diabetes are urine dipstick, laboratory blood tests, and blood glucose monitors. However, those testing procedures are often perceived as painful and inconvenient for the patients. In this context, this study proposes an electronic nose (e-nose) for detecting three classes of diabetes (healthy, prediabetes, and diabetes) based on a patient breath. The proposed e-nose system is called DENS, which utilizes an optimized deep neural network for the classiflcation. DENS also attempts to enhance accuracy and to reduce the error rate from previous studies. Therefore, this paper has three contributions: (i) the optimal gas sensors for capturing patient breaths; (ii) the optimal signal preprocessing; (iii) the fine-tuned parameters of deep neural network (DNN) for classifying multilevel diabetes. The proposed system successfully detected multilevel diabetes with an accuracy of 96.29% and showed a minimum classiflcation error of 0.050.