A New Approach for Aneurysm Detection Based on CNNs

An intracranial aneurysm (IA) is an abnormal bulging of a blood vessel caused by a weakening of its wall. This bulge can rupture and cause internal bleeding. Rupture of an internal blood vessel causes subarachnoid haemorrhage. Without detection and treatment, such damage to the artery leads to death. The development of an automated system to help doctors accurately diagnose IA can save many lives. For this reason, we focused on neural networks that would be able to recognize IA. In our work, we propose a CNNs (convolutional neural networks), which is mainly used to classify different types of images, both in transportation and medicine. Conv2D layers are most used in image processing since there is convolution over the image, which causes better classification. Our proposed network got an overall accuracy of 97.36% during the classification of the dataset into vessels and aneurysms, and the accuracy of recognizing aneurysms directly was 84%. We compare the results with other freely available experiments, where the authors obtained the highest overall classification accuracy on POINTCNN models of 90.44% and directly on aneurysms on PN++ models of 88.55%.