Brain Stroke Detection and Classification System: A Hybrid Approach using Deep Learning Techniques

Abstract

Brain stroke is a major health concern around the globe, which causes significant disability and loss of life. Effective treatment can be ensured only with accurate and timely diagnosis. However, traditional methods for stroke diagnosis heavily rely on manual evaluation of patient symptoms and CT scans, which prove to be time-consuming and prone to errors. Our major objective of this study is to develop a reliable stroke detection and classification system that can automate the diagnosis process using deep learning techniques. The proposed pipeline integrates a MLP model analysis textual patient health records and a CNN model for image based classification of CT scans. The study aims to provide a fast, reliable and systematic tool to aid medical professionals in detecting stroke accurately and removing any hindrance in treatment. In addition to the proposed methods, other models including ResNet50 and VGG for stroke classification and Random Forest for text analysis were also explored. The MLP model achieved accuracy of 94.67% for stroke detection based on patient data while the CNN displayed superior performance with accuracy of 98.6%. Relative analysis with accuracy, precision-recall and F1 score confirmed CNN beat pre-trained models like VGG and ResNet50, while the MLP model outperformed the Random Forest Classifier.