Master thesis presentation – 3D finite difference time domain simulations of microwave measurements of stroke patients and analysis of stroke patients undergoing thrombolytic treatment

Presented by Antti Stålnacke

Date:  Thursday, February 27th, 10.00

Location:  Conference room Änggården, Sahlgrenska Science Park, Medicinaregatan 8 A

Examiner:  Andreas Fhager, Chalmers S2

Stroke is a big issue for today’s health care and society. It causes large direct costs for health care, large indirect costs for society and is also the fourth largest cause of disease burden measured in disability-adjusted life years. Stroke means that the blood supply to parts of the brain is cut off or limited leading to reductions in cognitive or physical functions. Stroke is most often due to a blood clot, a so called ischemic stroke (IS), but can also in about 15\% of patients be due to a bleeding, a so called intracerebral hemorrhage (ICH). The treatment path differs between these two conditions and an earlier diagnosis than possible today would be beneficial. Medfield Diagnostics is developing a microwave based diagnostic tool for early diagnosis of stroke. This tool relies on a classification algorithm and the first purpose of this thesis is to increase the understanding of how this algorithm handles microwave based data related to stroke. The second purpose is to develop a variable 3D microwave model of patients with ICH for finite difference time domain (FDTD) simulations to investigate how the ICH effects the transmissions and reflections and also classification results.

The thesis consist of two parts. Part one is to look at data from patients with an IS while they receive a clot dissolver. The goal here being to see a correlation between the National Institute of Health Stroke Scale (NIHSS) score and a value output by the classification algorithm over time. The second part consists of using an FDTD solver to simulate microwave measurements on a large number of simulated patients. These simulations are then analyzed to observe how the classification algorithm handles different cases. In total 1000 patients are simulated of which 500 represented healthy patients and 500 are patients with ICH. The parameters that could vary were head size , hair thickness, bleeding position and bleeding size.

A correlation between NIHSS and output of the classification algorithm could not be seen with any of the approaches used in this thesis. Therefore a number of other approaches are suggested in the future work section. The simulation model and the resulting simulations showed promising results for being a valid model of the clinical measurement of the patient. The size of the bleeding together with the thickness of the hair had a large impact on the outcome of the classification. The best classification was also expected to improve with added number of patients. It was found that this was not true for patient sets smaller than 200 but true once the set became larger and the best classification for the whole data set had a maximum area under curve (AUC) value of 0.93.