Automated Detection of Diabetes From Exhaled Human Breath Using Deep Hybrid Architecture

In this paper, we have proposed an automated medical system for detecting type 2 diabetes from exhaled breath. Human breath can be used as a diagnostic sample for detecting many diseases as it contains many gases that are dissolved in the blood. Breath-based analysis stands out among the different non-invasive ways of detection as it provides more accurate predictions and offers many advantages. In this work, the concentration of acetone in the exhaled breath is analysed to detect type 2 diabetes. A new sensing module consisting of an array of sensors is implemented for monitoring the acetone concentration to detect the disease. Deep learning algorithms like Convolutional Neural Networks (CNN) are normally used to automatically analyse medical data to make predictions. Even though the CNN performs well, a few modifications to the network layout can further improve the classification accuracy of the learning model. To analyse the sensor signals to generate predictions, a new deep hybrid Correlational Neural Network (CORNN) is designed and implemented in this research. The proposed detection approach and deep learning algorithm offer improved accuracy when compared to other non-invasive techniques.