A machine learning approach to microwave sensing for non-invasive alzheimer’s disease early detection

Early Alzheimer’s disease detection can greatly benefit patients, caregivers, and clinicians. Unfortunately, current diagnostic procedures are invasive, expensive, and not easily portable. To overcome these limitations, microwave sensing is emerging as an alternative non-invasive approach to distinguish between healthy and pathological conditions, based on the variation of permittivity in cerebrospinal fluid in Alzheimer’s patients. In this framework, our paper explores the use of machine learning applied to microwave sensing data, by means of a multilayer perceptron classifier. Different architectures have been considered and appraised by relying on experimental data collected with controlled experiments involving a multi-tissue head phantom that can be filled with tissue-mimicking liquids simulating different stages of the pathology. The initial results confirm the potential of the proposed non-invasive approach to early-stage Alzheimer's disease diagnosis.