MRI Slice Assisted Ischemic-Stroke Detection using Deep-Learning Scheme

The brain plays a crucial role in physiology and is in charge of information processing and making decisions. Any abnormality in the brain is a medical emergency and needs prompt identification and intervention. Worldwide, one of the main causes of disability and mortality is brain stroke (BS). It is categorized as: ischemic- and hemorrhagic-stroke, depending on the cause. Magnetic Resonance Imaging (MRI) is frequently used for clinical level stroke detection. A selected computer algorithm is used to analyze the image captured using a chosen modality in order to detect strokes. This study aims to create a deep learning system to categorize the selected MRI slices. Phases of this study involve; MRI collection and 2D axial slice extraction, image enlargement to 224x224x3 pixels, feature extraction with a selected deep-learning (DL) model, and binary classification and three-fold cross validation. In this study, 1600 images (800 normal and 800 stroke) are examined. The DL-model with the SoftMax classifier is selected, and the classification task is carried out. The study's experimental results demonstrate that the DenseNet121 scheme's two-fold training helped to achieve a superior detection accuracy of 98.12% when compared to the other DL-models used in the study.