Stroke Classification in Brain CT Images Using Transfer Learning With Fastai

To identify strokes, which rank among the leading global causes of death, doctors commonly employ computed tomography (CT) imaging. Researchers are increasingly applying deep learning (DL) and machine learning (ML) methods to improve the accuracy of stroke classification and detection. In particular, DL methods have demonstrated significant potential in segmenting stroke regions within brain medical images. However, training convolutional neural networks from scratch is challenging due to dataset size constraints. Achieving timely and precise stroke diagnosis remains a significant hurdle. The purpose of this research is to classify brain stroke cases using CT scan images through transfer learning strategies. Pretrained architectures including ResNet, VGG, Inception, and EfficientNet are utilized, with optimization carried out in the fastai framework to improve computational efficiency and reduce training time. We developed these artifical intelligence (AI) systems by training them with data, making adjustments throughout the process to ultimately categorize patients as having a stroke or not. We evaluated different models using measures such as accuracy, speed, and how well they differentiate normal and stroke-affected CT scans of the brain. This helped us choose the most effective model. In this research, ResNet50 outperformed other models in accuracy (88%), and DCNN was the most time-efficient model.