Towards Microwave Based Monitoring of Brain Thermo-Fluid Dynamics

The human brain is a central organ whose complex biological processes closely reflect the health and well-being of an individual. The temporal and spatial variations of thermal and hemodynamic states of the brain are of utmost clinical importance for both intensive care patients as well as healthy subjects. Microwave-based techniques are promising for such applications because they provide non-ionizing, non-invasive sensing with low system cost, compact and portable hardware, and tunable tissue penetration capabilities.

The primary objective of this thesis is to develop a Machine Learning (ML) powered microwave system for non-invasive thermal and hemodynamic monitoring of the brain. The quantitative and automatic method proposed in this work is based on two-level features extracted from the magnitude and phase response of microwave sensors. The proposed method ensures an automatic, near-real-time operation using a unique Ordered Selection Scheme (OSS).

The secondary objective is to optimize the proposed microwave system in terms of inter-antenna distance, bandwidth, sampling rate, and a specific regression model suitable for a particular application. The proposed method is tested on seven different microwave sensors with varying strategies of placement, using dynamic human head phantoms developed in this work and healthy human subjects. The response of the proposed microwave sensors is also evaluated using simulations in CST Studio Suite with planar layered models and realistically shaped voxel models.

A strong analogy is observed between the thermo-fluid response of the realistic brain phantom model and the response of the proposed microwave system. The proposed system can measure these changes externally from the scalp of the subject, without requiring surgical intervention or direct tissue penetration, thereby ensuring safety and patient comfort. Further, the system offers precise, reproducible numerical measurements, achieved through rigorous calibration of measured signals against tissue-mimicking phantoms and anatomically realistic voxel models.