An Effective Application of Deep Learning in the Early Diagnosis of Stroke

Stroke constitutes one of the primary reasons for death and severe weakness that affects millions of people worldwide and has a significant chance of developing into a pandemic in the coming years. Early diagnosis and fast treatment decisions are crucial in preventing paralysis, brain death, and other severe consequences of this condition. ML algorithms have been widely used to diagnose, analyze, and predict this disease, but there are data quality difficulties because the resources utilized to obtain the data are spread across multiple institutions. Hence, the deep learning framework with CNN is demonstrated in this paper. This technique makes use of a multimodal stroke dataset that is accessible to everyone in the Kaggle repository. 6100 MRI images are gathered for this process. The collected images contain noise and are of bad quality. So, the Weiner filter is utilized for pre-processing phase to remove the noise. Then, the watershed technique is employed for the segmentation process. Subsequently, Lion Optimization Algorithm is utilized for extracting the features. Lastly, CNN has been applied for the classification. By validating the performance measures, the LOA-CNN technique seems to be higher than the current techniques. The results demonstrated that the presented method achieves precision, accuracy, and recall rates of 99.32%, 99.52%, and 99.67%, respectively.