In recent years, Deep Learning has made many advances in a variety of fields. This has resulted in it establishing itself as the new state-of-the-art for several artificial intelligence related tasks. Within the field of Computer Vision, applications of Deep Learning for medical imaging have shown encouraging results.
The assessment of vascular structures is a crucial step in the diagnosis of a range of cardiovascular-related conditions. And currently, we rely heavily on the evaluations of expert clinicians whom manually dissect each image. This process is not only time-consuming and expensive but also error-prone due to operator fatigue. Deep Learning then stands as the best candidate to aid in the automation of this task, building tools to assist clinicians with their decisions.
However, in a recent research study, a deep 3D convolutional neural network was outperformed by conventional machine learning filters when applied for vessel segmentation in 3D magnetic resonance (MR) images. This project focuses on extending its novel implementation, improving upon the methods used for deep learning and later evaluating its performance.