Non-invasive microwave sensing system for early Alzheimer’s disease diagnosis

Alzheimer's disease (AD) is the most common form of neurodegenerative disease and one of the leading causes of death in the Western world. The pathological changes in the brain due to AD include brain atrophy (progressive shrinkage of the brain) and enlargement of the lateral ventricle. From a histological perspective, the main change in the brain during AD is the progressive accumulation of beta-amyloid plaques and tau neurofibrillary tangles. Beta-amyloid proteins form irregular clumps in the brain, which combine to create plaques that disrupt the signals between synapses. These amyloid plaques initially develop in the neocortex and spread throughout the brain in the early phase of the disease. Additionally, neurofibrillary tangles,formed by intracellular clusters of tau protein, become increasingly prevalent in the brain as AD progresses. These tangles hinder the transport of cell nutrients and are believed to contribute to neuron death. Early detection of Alzheimer's disease can significantly benefit patients, caregivers, and clinicians involved in managing the disease. Presently, crucial assessments such as cognitive testing are essential for diagnosis. Alongside these assessments, the most common techniques used include positron emission tomography (PET) and cerebrospinal fluid (CSF) sampling that however can be considered invasive and may not be convenient for patients. This thesis focuses on the integration of a microwave sensing system designed for early AD diagnosis, aiming to address the limitations associated with ongoing approaches. The system is based on the variation of permittivity in CSF of AD patients and operates by gathering data from a multi-tissue head phantom. In order to collect the data, the head phantom, which mimics the dielectric properties of different tissues, was filled with liquid solutions that simulate both healthy and AD condition CSF. To accurately represent the disease progression, four mock CSFs were created, corresponding to various stages of the disease (poor, mild, mild-high, and severe). The acquired data from controlled experiments were utilized as inputs for a machine learning (ML) binary classifier employing a multilayer perceptron structure (MPL). The Scattering parameters (S-parameters) were obtained using two different setups: firstly, a configuration involving a two-port vector network analyzer (VNA) connected to a switching-matrix (SW) that controls six flexible antennas; secondly, an alternative setup utilizing a four-port VNA, eliminating the need for the SW and greatly enhancing the signal-to-noise ratio (SNR) during data recording with four flexible antennas. The frequency range analyzed spanned from 500 MHz to 6.5 GHz. Within this range, it was observed that, unlike in healthy patients, the permittivity exhibited a decrease while the conductivity values remained relatively constant. Various approaches were tested in constructing a MLP binary classifier, and the results demonstrate the high accuracy of our innovative Alzheimer's disease detection method (over 90% accuracy). These promising findings pave the way for potential future applications in real-world settings and instill optimism for significant advancements in related areas.