Type 2 diabetes diagnosis assisted by machine learning techniques through the analysis of FTIR spectra of saliva

• Published in: Biomedical signal processing and control 2021-08, Vol.69, p.102855, Article 102855
• Main Authors: Sanchez-Brito, Miguel, Luna-Rosas, Francisco J., Mendoza-Gonzalez, Ricardo, Vazquez-Zapien, Gustavo J., Martinez-Romo, Julio C., Mata-Miranda, Monica M.
• Format: Article
• Language: English
• Subjects: Artificial intelligence techniques; Artificial neural network; Diabetes; Fourier Transform Infrared (FTIR) spectroscopy; Human saliva
• ISSN: 1746-8094; 1746-8108
• DOI: 10.1016/j.bspc.2021.102855

[Display omitted] •Analyzing FTIR spectra of saliva with machine learning techniques, it is possible to characterize patients with and without diabetes.•The proposed methodology is more agile than the blood analysis and does not require reagents, which is why it is a lower cost option.•ANNr and SVMr turn out to be the best options to characterize the FTIR spectra of patients with and without diabetes.•The changes in amide A and lipids turn out to be the ones that best allow to discriminate patients with and without diabetes. Diabetes is one of the four main non-communicable diseases worldwide. Despite not being a fatal disease, many complications derive from this illness that causes a drastic deterioration in the patient's health. Diabetes is a silent disease that, on many occasions, causes symptoms until the disease is already advanced, and due to the lack of education in health prevention, only a small part of the population undergoes routine laboratory studies. If this disease is detected on time, the quality of life could be improved. However, the simple facts of taking a blood sample, control studies are omitted. Besides, there is a need to sample the patient many times according to its severity and control. In the present work, we provide a novel technique based on the FTIR spectra of saliva samples to diagnose this disease. After analyzing the samples of 1,000 people, we found that it is possible to identify patients with this pathology through artificial neural networks and SVMr reliably. As it is not invasive and does not require reagents or complex processes, the proposed technique could be more agile and cheaper than traditional ones.