Deep O-SegNet-Based Stroke Lesion Segmentation and HX-ShuffleNet with Optimization for Stroke Classification Using MRI Image

Presently, stroke is the world’s biggest underlying cause of mortality, and is caused because of inadequate blood flow to the brain. Stroke leads to cell death, and it has various risk factors if not treated at an initial stage. Therefore, a new method called Chronological Gold Rush Optimization-based Hybrid Xception ShuffleNet (CGRO-based HX-ShuffleNet) is developed to detect stroke using magnetic resonance images (MRIs). As a first step, the MRI input is denoised by the adaptive median filter (AMF), and then the stroke lesion is segmented by employing the deep O-SegNet, which is formed by combining the outputs obtained from DeepJoint segmentation and object segmentation network (O-SegNet) on the basis of the RV coefficient. Thereafter, augmentation is effectuated, and the features are retrieved. Subsequently, stroke is detected by employing the proposed CGRO-based HX-ShuffleNet technique. Here, HX-ShuffleNet is obtained by combining the ShuffleNet and Xception methods, and it is tuned using CGRO. Moreover, CGRO is constructed through the integration of gold rush optimization (GRO) and the chronological concept. Furthermore, the CGRO-based HX-ShuffleNet showed superior performance with a true positive rate (TPR) of 94.78%, an accuracy of 93.67%, and a true negative rate (TNR) of 93.68%.