Master’s Thesis Presentation • Health Informatics • Parameterizing the Spatial Distribution of Renal Tumors using Modified Spherical Coordinates

Please note: This master’s thesis presentation will take place in DC 2102.

Christopher West, Master’s candidate
David R. Cheriton School of Computer Science

Supervisors: Professors Anita Layton, Justin Wan

Kidney cancer is a well-studied and highly prevalent cancer in which tumors grow and obstruct the functional tissues of the kidney. Medical imaging is commonly used to diagnose and verify the presence of these kidney tumors and can give some qualitative intuition on the scope of their proliferation.

Although the overall size of a kidney tumor may indicate the scope of its impact on the function of the kidney, there is good reason to believe that the tumor’s location may be similarly important. Renal tissues are highly specialized and separated into unique compartments that perform different tasks, so a tumor in one area may have other downstream effects than a similar one in another location. Much is known about renal carcinoma’s cell- and tissue-level proliferation, but the distribution of tumors in the kidney at the macroscopic level needs to be better studied. This knowledge gap partly stems from a lack of common registration and standardization schemes. One can describe the location of tumors in absolute spatial terms, but this has poor carryover for patients with naturally different kidney sizes or shapes.

With this in mind, we develop a novel spatial parameterization scheme that takes advantage of the intrinsic structures and symmetries present in the kidney. We use boundary-based distance metrics to model the concentric structures of the kidney, such as the renal cortex and medulla. To manage the variability present in the kidney segmentation process, we developed a method for using convex hulls to smooth and standardize the analysis region. A bilateral spherical coordinate system is developed to describe tumor position and scope uniquely. Using an atlas and Cartesian projections allows us to map tumors from distinct patients to the same shared space.

Using these tools, we process the Kits21 kidney tumor dataset and generate the first-of-its-kind kidney tumor heat map. We also theorize and provide preliminary experimentation into using spatial features for simple risk evaluation. A biologically-motivated prototype occlusion metric is designed to model the relative impact of a tumor based on size and location. Overall, this research paper demonstrates and justifies using this spatial parameterization scheme as a foundation for future quantitative analyses. The researcher hopes future work will improve and extend the method for further use cases.