Improving the detection of hemorrhages with image processing and deep learning method in brain aneurysms

Cerebral aneurysms are an important disease that threatens human life. Rupture of these aneurysms causes bleeding in the brain arteries. Clinically, computed tomography angiography (CTA) is widely used in the diagnosis of cerebral aneurysms. Determining the bleeding risk for cerebral aneurysms that do not bleed from the CTA images obtained helps in determining how to proceed in the treatment process of the disease. In this thesis study, Convolutional Neural Networks, one of the deep learning architectures, were used to distinguish whether a person is unhealthy or healthy from the data set created using CTA images of patients treated at the Neurosurgery Polyclinic at Fırat University Hospital. As a result, the presence of an aneurysm in the patient was examined. In this way, radiologists' errors in interpreting CTA images were kept to a minimum. This model was developed using the MatLAB program, which is the widely preferred programming language in the literature. The study also examined radiology reports of patients using machine learning algorithms. Thus, a data set containing the morphological and hemodynamic characteristics of the patients was created. For patients with cerebral aneurysms without bleeding, hemodynamic characteristics were classified as age, gender, and comorbidities. Similarly, the risk of rupture was estimated for patients with aneurysms by examining morphological features, location, width, length and type parameters. It was examined whether these features made a significant difference in the risk of bleeding. According to the results obtained in the study, although it was stated that there was no rupture in the radiology reports of 3 patients, it was predicted that rupture would occur in these patients according to the machine learning classifier model. When the disease history was examined, it was determined that the aneurysms had ruptured in these 3 patients, according to the radiology report results on different dates.