Classifying CT images of brain infarction with 3D CNN

Brain infarction occurs as a result of a blockage in the arteries which supply blood and oxygen to the brain. The restricted oxygen causes to stroke that can result in an infarction if the blood flow is not normalized in a short period of time. Approximately, 0.6% of people suffer from stroke every year. About one third is fatal. In addition, stroke is a third leading cause of death. For diagnosis, doctors want to see MRI (Magnetic Resonance Imaging) results to ensure if the patient has infarction or not. However, this process takes a long time while patients require immediate intervention. Losing time while waiting for an exact diagnosis might have fatal consequences. On the other hand, doctors can have CT (Computed Tomography) scan results in a short period of time but is not enough to tell the exact diagnosis for infarction due to uncertainty and the low quality of the imaging technique. This work aims to classify CT scans if the patient has infarction or not. This study uses 3D Convolutional Neural Network (3D-CNN) methodology. It achieves 93% as a maximum accuracy and 74% accuracy after 10-fold testing. We believe that this method could be used as a decision support system to detect the patients with a higher risk of infarction, and prioritize utilization to MRI for them to make the final diagnosis quickly.