Diabetes screening based on convolutional neural network and Raman spectroscopy

Type 2 diabetes mellitus (DM2) is one of the most widely prevalent diseases worldwide and is currently screened for in a laboratory setting by invasive techniques based on enzymatic determination of plasma glucose concentration. Therefore, studies to screen for diabetes by identifying molecular features in a non-invasive manner are of great importance. Raman spectroscopy has the advantages of rapid, in situ, non-invasive detection and is increasingly used in the biomedical field. Thus, convolutional neural networks are proposed to be combined with Raman spectroscopy technology for screening type 2 diabetes for different skin sites with Raman spectroscopy datasets. Compared to machine learning algorithms, convolutional neural networks do not require complex spectral pre-processing work. Compared with common machine learning algorithms such as support vector machines and random forests, convolutional neural networks have a higher classification accuracy of 95%.