Development of a Novel Deep Convolutional Neural Network Model for Early Detection of Brain Stroke Using CT Scan Images

Introduction

Stroke is a leading cause of disability and mortality worldwide, claiming an estimated 15 million victims annually (Associate, 2021; Jing, 2020). Early stroke detection and treatment are essential for improving stroke outcomes and lowering the risk of fatality or permanent impairment (Dev et al., 2022; Matta-Solis et al., 2022). The capacity of computed tomography (CT) scans to create good images of the brain and quickly spot areas of bleeding or occlusion makes them a popular diagnostic tool for stroke detection (Sailasya & Kumari, 2021; Lin, 2021). However, the lengthy and prone to error manual interpretation of CT scan pictures by radiologists may postpone diagnosis and treatment. Deep learning techniques, particularly deep convolutional neural networks (DCNNs), have shown encouraging results in recent years when used for automated stroke identification from CT scan pictures. Unlike humans, DCNNs can learn complex properties from enormous amounts of data, which enables them to recognize subtle patterns and anomalies that could otherwise go undetected. Deep convolutional neural networks (DCNNs) are the foundation of current stroke detection methods, but they can only classify images as stroke- or non-stroke-related without taking the onset time into consideration. Early stroke detection can lead to quicker treatment and improved outcomes (Kshirsagar et al., 2021; American Stroke Association, 2016; Anisha, & Saranya, 2021). In this study, we develop a fresh deep convolutional neural network model for early stroke diagnosis and put it to use on CT scan pictures. The proposed methodology aims to detect strokes during the first six hours, when medical intervention is most successful.

Finding a solution to reduce the time and human error involved in manually interpreting CT scan pictures for the purpose of detecting strokes was one of the main drivers behind this project. Early detection is important in stroke cases because it allows for quicker diagnosis. Automated technology makes early detection possible. DCNNs have shown useful for stroke identification even though the majority of modern models don't take the actual time the stroke began into consideration. We want to build a deep convolutional neural network model that can precisely diagnose and predict stroke within the first six hours of symptom onset (Geeta et al., 2022; Ouyang & Davis, 2019).

A stroke is a potentially fatal medical emergency that happens when the blood flow to the brain is suddenly interrupted. About 87% of all strokes are ischemic strokes, which happen when blood supply to the brain is interrupted. Hemorrhagic strokes, which result in brain hemorrhage by rupturing a blood artery, account for around 15% of all strokes (Yu et al., 2022; Kaiyrbekov and Sezgin, 2020; Meier et al., 2016).