| Abstract: | Brain MRI is an important method for examining the diseases caused by various cerebral pathologies, and the measurement of temporal lobe volume is useful for identifying dementia and temporal lobe abnormalities. However, no segmentation algorithm for the temporal lobe on coronal MR images has been established. Such an algorithm is needed because the shape of the temporal lobe on coronal images varies from area to area. The purpose of this research was to develop a segmentation method for the posterior portion of the temporal lobe on coronal MR images. The subjects were 11 normal patients, whose coronal T(1)-weighted images were selected for this study. The preprocessing algorithm for segmentation consists of smoothing, binarization, and thinning. The first step of the segmentation process consists of recognition techniques for the temporal lobe region on thinning images. The next step is distance transformation on identified thinning images. Finally, the temporal lobe was segmented by using the original images and distance transformation images and employing the newly developed algorithm. The rate of accuracy of automated recognition was over 74% for all cases, while the average rate of accuracy was 83.2+/-4.0%. These results suggest that this segmentation method can clearly segment the temporal lobe and has potential for clinical use. Based on this study, although it included only 11 normal patients, we have started applying this segmentation method to many patients, with or without temporal lobe disease. |
