| Abstract [eng] |
Computed tomography images are used by the medical experts to diagnose and monitor abdominal aortic aneurysm by analysing slices of computed tomography images. As there is a shortage of automatic software tools for the detection and diagnostics of abdominal aortic aneurysm, the process is often performed manually or semi-automatic, which results in long analysis time and increases the risk of the patient not getting treatment on time. In this work literature analysis is performed, which describes the most frequently used types of medical imaging in the abdominal aortic aneurysm research and the methods of preprocessing and marking of images. The segmentation of abdominal aortic aneurysm is performed by training three neural networks with several loss functions. Various data processing methods are used, including cropping, changing scale and ratio of original image. Dropout is used to increase the accuracy of the segmentations. In addition to this, a method for the segmentation of the abdominal aortic aneurism is suggested, which combines two neural networks that were trained to segment different regions. This way similar segmentation results can be achieved to those results, that were achieved using twice as few patients. |
