Classification of brain ischemia and hemorrhagic stroke using a hybrid method

Abstract

The availability of artificial intelligence applications has expanded with the development of extremely powerful computers. This speeds up the diagnosis of many ailments and allows for the application of effective treatments. This facilitates the work of numerous experts in the field of health, and more trustworthy outcomes are attained. Many researchers have started working in this topic as a result of how swiftly artificial intelligence has entered human life and the promising outcomes that have been obtained. The detection of brain strokes from CT images, which are the most prevalent in the world and a major threat to human life, is a field that is currently seeing a flurry of deep learning and machine learning-based approaches. This study attempted to distinguish between ischemic and hemorrhagic brain stroke types using CT images. Utilizing a hybrid methodology, multi-classification was carried out utilizing EfficientNetB0 and SVM model from ischemic, hemorrhagic, and normal brain CT images in our dataset. The purpose of this study is to demonstrate a higher classification performance by combining the feature extraction skills of deep learning networks with the classification capability of machine learning models. The ischemic, hemorrhagic, and normal CT images in our dataset were used to train the EfficientNet-B0 + SVM, EfficientNet-B0 + KNN, ResNet50 + SVM, ResNet50 + KNN, ResNet50, VGG19 + SVM, VGG19 + KNN, and VGG19 models. There was a triple classification. According to the study, the hybrid EfficientNet-B0 + SVM architecture had the highest load success rate. The model had 95.13%, 94.93, 95.06, and 94.94% accuracy, recall, precision, and F1 values, respectively. These findings demonstrated that the EfficientNet-B0 + SVM architecture is a highly effective method for classifying different types of brain strokes from CT images. © 2023 IEEE.