Aleksandra Osowska-Kurczab
supervisor: Tomasz Markiewicz
Though the development of modern medicine has helped lead to a surge in average life expectancy, there are still numerous obstacles waiting to be solved. One of them is the diagnosis of renal cancers. This neoplastic disease seems perplexing in detection and treatment, mainly because of unspecific symptoms. Due to this fact, almost 50% of cases are diagnosed accidentally, during imaging tests conducted towards other diseases. Early diagnosis is a crucial factor determining survival prospects and chances for minimally invasive procedures. Therefore there is a need for new methods supporting the diagnosis and treatment of renal cancers.
In my doctoral research project, I'm devoted to the development of an automated system aiming at supporting the medical doctors in detecting renal tumours basing on Computed Tomography scans. My latest works were mainly focused on methods of differentiation of 8 subtypes of renal tumours. Identification of lesion type is critical from the point of view of surgery planning and one of the most important research problems to be solved in my doctoral thesis. Deep learning, texture analysis, as well as ensemble learning, were investigated as potential building components of the detection system. The final performance of the system reached 94% of the weighted F1-score. During the talk, I present a full classification component of the system and discuss research outcomes introduced in my publications in IJCNN, CPEE and BPAST.