Methodological aspects for improving clinical value of SPECT and MRI

Abstract

Image processing methods were developed for SPECT and MR images. The methods were validated in clinical environment. Segmentation of SPECT images for region of interest (ROI) analysis was found to be unreliable without accurate attenuation and scatter correction for the original images.

The reliability of ROI analysis of brain SPECT images was enhanced using registration with MRI. The method was based on external markers. The registration error was studied using phantom tests and simulations. It was concluded that the registration accuracy was not the limiting factor in ROI analysis of the registered images provided that the external marker system was properly designed and attached.

Quality requirements for MRI data from patients with cerebral infarctions were evaluated in order to make segmentation as automatic as possible. Quantitative information from these images could be extracted with e.g. statistical and neural network classifiers, but required more manual work than expected due to the visible intensity nonuniformity in the images.

The third application consisted of developing a registration methodology for ictal and interictal SPECT, MRI and EEG for improved localization of the epileptogenic foci. The methodology was based on SPECT transmission imaging. The accuracy of registration was about 3-5 mm.

As a conclusion, improved analysis of SPECT and MR images was obtained with the carefully evaluated methodology presented in the thesis. The registration procedure for brain SPECT and MRI as well as the registration procedure for epilepsy surgery candidates are in clinical use for selected patients in Helsinki University Central Hospital (currently Health Care Region of Helsinki and Uusimaa).