A Hybrid Dn-CNN and Hough Transform-Based Framework for Acute Ischemic Stroke Classification in CT Images

Abstract

Stroke is a life-threatening condition that involves a lack of blood flow to brain cells. Non-Contrast Computed Tomography (NCCT) is commonly used to diagnose stroke, but detecting Acute Ischemic Stroke (AIS) through imaging techniques poses a difficult task due to issues related to contrast and noise interference. This study presents a hybrid system that combines pixel-correlation-based image enhancement, ROI extraction using patches, and geometric structure-based patch extraction using the Hough transformation, followed by a Denoising Convolutional Neural Network (Dn-CNN) to classify brain CT images of patients with AIS. The uniqueness of this method stems from the fusion of an asymmetry-driven patch extraction technique with structural features of the brain. The performance of the proposed approach was tested on a brain image dataset consisting of 10 patients. Experimental results were evaluated using Peak Signal-to-Noise Ratio (PSNR), accuracy, sensitivity, and specificity metrics, demonstrating that the suggested system achieved an average classification accuracy of 96%, surpassing traditional classifiers such as SVM, CNN, and DNN.