Automated detection of ischemic stroke with brain MRI using machine learning and deep learning features

Recently, the occurrence rate of stroke in humans has been growing steadily for various reasons. Ischemic stroke lesion (ISL) is a stroke that arises due to an inadequate oxygen supply to the brain tissues, and appropriate diagnosis is necessary to treat the patient. This research aims to employ a deep learning (DL) structure to identify ISL in multimodality brain MRI slices. To implement a reliable ISL recognition scheme, this work employed convolutional neural network (CNN)-based joint segmentation and categorization system. The stages of this scheme include: (1) VGG-UNet-supported segmentation, (2) machine learning (ML) feature mining, (3) deep-feature extraction, (4) features ranking and concatenation (DL + ML), and (5) binary classifier-supported classification. During the classification, fivefold cross-validation is considered and the best outcome is chosen as the result. The test pictures for this research were obtained from the ISLES2015 and the experimental investigation was implemented using MATLAB. The achieved results are separately presented for DL and DL + ML and the classification accuracy achieved with DL + ML is superior to other methods employed in this work.