Automated localization and segmentation of brain lesions due to hemiplegia using windowing-based entropy comparison

Abstract

Magnetic Resonance Imaging is the most popular imaging technique used to in brain lesion diagnosis. Brain lesions due to Stroke appear as a gray region similar in color to some normal tissues like gray matter. Manual extraction of brain lesion is time-consuming. On the other side, current automated methods require either multispectral MR images or extensive time of training. To avoid these problems, this paper suggests a novel automated brain lesion recognition method that uses single spectral MR images to efficiently extract brain lesions with a reasonable amount of time and with acceptable accuracy. By applying this method, it can distinguish brain lesions automatically. The principle of operation and mathematical characterization of the suggested algorithm are given in details. The results of the proposed algorithm using a single T1 weighted MR images for stroke subjects and for healthy subjects with simulated brain lesions are presented. Results showed that the suggested window-based entropy comparison method could identify a lesion with a minimum size of 10×10×10 mm and with an average accuracy of 3 voxels and success rate of 91%.