Detection of Alzheimer's by Machine Learning-assisted Vibrational Spectroscopy in Human Cerebrospinal Fluid

Abstract

Nowadays, the diagnosis of Alzheimer's disease is a complex process that involves several clinical tests. Cerebrospinal fluid contains common Alzheimer-related biomarkers that include amyloid beta 1-42 (Aβ1-42) and tau proteins. In this work, we propose vibrational spectroscopy techniques supported by machine learning for the detection of biomarkers in cerebrospinal fluid that are related with Alzheimer's by prediction models. Vibrational spectroscopy provides the entire biochemical composition of the body fluid, and thus, small but typical physiological changes related with the pathology can be ascertained. Within a machine learning framework, Raman and FTIR spectra were analyzed, which were taken from samples of healthy volunteers in comparison with samples from patients clinically diagnosed with Alzheimer's. We find that a logistic regression model can discriminate between healthy control and Alzheimer's patients with a precision of 98%, when the input for the model combines data from both vibrational spectroscopy methods. Our approach shows high discriminative capabilities and constitutes a proof of concept for an alternative and accurate tool for the diagnosis of Alzheimer's disease.