Measurement of infarct volume in stroke patients using adaptive segmentation of diffusion weighted MR images

Abstract

This paper describes a semi-automatic method of determining the infarct volume, an important parameter in the assessment of stroke patients, from MRI Diffusion Weighted Images (DWI). An adaptive thresholding algorithm incorporating a spatial constraint was used to segment the images. The relationship between adjacent pixels was modeled using a Markov Random Field (MRF) and the Iterative Conditional Modes (ICM) method was used to find a locally optimum solution. In order to improve the robustness of the ICM method, initial threshold levels were determined automatically using a nonspatial method. Preliminary results showed that the completely automatic technique failed if the infarct was too small or if the contrast was too low. The operator was therefore given a choice of modifying the initial threshold levels manually. It was also necessary to edit the final segmentation results in some cases as nerve tracts may also appear as bright regions on the images. Simulation studies were used to determine the accuracy of the technique. Reproducibility studies were carried out to determine the effect of inter and intra observer variability and patient positioning. The semi-automatic technique was quicker and more reproducible than manual segmentation and allowed the infarct volumes to be measured with a repeatability coefficient of < 6 cc.