Computer Aided Diagnosis of Stroke from Brain CT Images

This paper presents an automated method to detect and classify an abnormality into infarct and hemorrhagic or normal in brain CT images. Firstly, the original image is converted into gray scale and noise is removed by median filter. Then skull regions are removed by a morphological function. Image is classified into infarct or hemorrhagic stroke or normal image based on features from gray-level co-occurrence matrix (GLCM), mean, standard deviation and area. The abnormality contained hemisphere is also detected based on these features. The different brain parts are segmented by using modified canny edge detection method. Histogram-based thresholding method is applied to extract region of interest (i.e., abnormality) from the parts of brain. The accuracy of the classification results can be compared with the specialists’ decision. The results are segmented images for abnormal region and its abnormal type. Experiments done on real CT images show the efficiency and accuracy of the proposed classification method.