Brain Stroke Detection from CT Images using Transfer Learning Method

Recently, with the production of high-capacity computers, artificial intelligence methods have been used in many areas. In particular, in the field of health, artificial intelligence methods are used to detect diseases and determine the treatments to be applied. Deep learning methods, a sub-branch of artificial intelligence, show a high success in diagnosing many diseases thanks to its deep CNN networks. In this study, firstly, it was tried to determine which deep learning methods are more successful for the detection of brain stroke from computerized tomography images. This study is of great importance in terms of determining which deep learning architectures we will focus on in our future studies on brain stroke. In this study, brain stroke disease was detected from CT images by using the five most common used models in the field of image processing, one of the deep learning methods. The pre-trained ResNetl01, VGG19, EfficientNet-B0, MobileNet-V2 and GoogleNet models were run with the same dataset and same parameters. As a result, the EfficientNet-B0 architecture showed the highest performance with 97.93% accuracy. ResNetl01, VGG19, MobileNet-V2 and GoogleNet models showed 94.32%, 97.28%, 92.30% and 91.61% accuracy rates, respectively. With this study, it was concluded that the EfficientNet architecture is more suitable for the detection of brain stroke disease from CT images.