Classification of Cerebral Infarction Using Convolutional Neural Networks

Cerebral infarction is a serious condition that occurs when blood flow to a part of the brain is suddenly interrupted, resulting in damage to or death of nerve cells due to lack of oxygen. Early diagnosis and treatment are crucial to minimize the permanent neurological damage that cerebral infarction can cause. The magnetic resonance imaging (MRI) technology is widely used for diagnosing and assessing cerebral infarction. In this study, an original dataset is created from images obtained from the Radiology Department of Yozgat Bozok University Research Hospital. The prepared unique dataset is subjected to classification by labelling it into two classes (Normal and Infarct) by expert clinicians. The classification process is performed using a novel proposed convolutional neural network model and a pre-trained EfficientNetB0 model. The proposed model achieved average training and validation accuracies of 99.43% and 99.09%, respectively, with a test accuracy of 99.2%. Furthermore, with an F1-score of 99.2%, the model's classification performance is also successful. Looking at the results obtained for the EfficientNetB0 model, training and validation accuracies are observed as 96.2% and 94.74%, respectively, while the test accuracy is 94.91 % and the F1-score is 95%. In conclusion, the proposed model exhibits high performance despite having less complexity and offers better classification capability with fewer parameters. This makes it a preferable option, especially in situations with limited computational power.