Detection and Classification of Stroke Using Texture Analysis on CT Images

Abstract

Detection and diagnosis of various types of ailments and fractures is one of the fields that is highly dependent on medical image processing. The need for ccorrect diagnosis of the ailment type is very important for proper medication as any delay or wrong diagnosis may become fatal to the patient. Stroke is a disease that is caused by obstruction of the blood supply to the brain, or bleeding of blood into the brain. Any delay in administering right medicine may result in permanent disability or sudden death. Many methods have been developed to diagnose stroke using MRI images. In this work we have used Computed Tomography (CT) images for diagnosis stroke using texture features and classifiers. Five different classifiers are used and they are combined to get better diagnosis accuracy. The accuracy of classifier ensemble output is found to be 95% and the area under ROC (AUC) was found to be about greater than 0.95 for all the classes. The method proves very effective for diagnosis of stroke with good accuracy and able to differentiate acute, chronic and hemorrhage successfully.