Automatic identification and segmentation of infarct lesions from diffusion weighted MR images and ADC maps

Introduction There has been considerable interest in the use of diffusion weighted imaging (DWI) to detect the site and size of ischemic lesions in acute stroke patients [1, 2]. This information assists in classifying the stroke age and may also be useful in predicting the clinical outcome [3]. The measurement of infarct volume has an important role to play in: (a) decision-making in acute stroke, and (b) assessment of stroke therapy response in clinical trials. Previous methods for infarct segmentation from DWI images had many false positives, as they did not use the complementary information from apparent diffusion coefficient (ADC) maps (free from T2-shine through and coil-shading effects) available with DWI images [4,5,8]. In the present work, we have combined the contrast information provided by DWI and physiological information provided by the ADC maps to segment the infarct core automatically. A T2 image is part of the standard DWI protocol and hence ADC map is available for use. Previous research [6,11,12] shows that ADC values of acute stroke lesions fall in a range and we use this to get rid of susceptibility artifacts and some of in-homogeneity artifacts, which may mimic acute stroke. Methods The algorithm was run on 22 data sets with cerebral stroke of which 15 subjects were imaged using GE Signa HDx 3T system and 7 with GE Signa Excite 1.5T system. DWI images were acquired using a spin-echo EPI sequence, with diffusion encoding applied (b-factor = 1000 s/mm) along the three principal axes and trace maps generated to remove the effect of anisotropy. In-plane resolution was either 0.9375 mm or 1.01 mm, slice thickness varied from 3.5 mm to 6 mm and spacing between slices varied from 0-1 mm. ADC images were calculated for each slice on pixel-by-pixel basis as: ( ) ( ) ( ) 1000 b 0 b / y , x S / y , x S ln ) y , x ( ADC 1000 = . As shown in fig.1, the method primarily consists of four steps: skull stripping, automatic seed(s)