Borghild started on her master thesis focusing on digital pathology at PiV

I have recently started on my master thesis focusing on digital pathology at PiV. My project focuses on Deep learning-based multi-compartment segmentation in digital nephropathology.

There has been conducted quite a bit of research on glomeruli segmentation, but not a lot on multiclass segmentation of kidney structures. Multicompartment segmentation means simultaneous identification of all relevant/selected structures (e.g. glomeruli, tubuli, etc). This is an important task to tackle, as there is a lot of important pathological information that can be easily accessed with automatic segmentation of these structures. An automated approach will also be more time efficient, increase reproducible results and decrease personal bias.

My work until now has consisted of exploring pilot cases and annotating them manually in QuPath to figure out how many classes we wish to classify and to learn to differentiate between them myself. I have also exported the polygon coordinates from QuPath to create masked patches of tubules using Python. These patches will be used to determine which tubule-classes we need.

I find math and physics interesting and therefore chose the physics-direction in my study program. However, it was when I started taking data science and machine learning courses I found my true passion. I could combine many fields I am interested in, and it allowed me to dive into the mathematical foundation of machine learning and deep learning as well as applying it for medical usage. That is when I found out that this was what I wanted to focus on in my master thesis.

I had knowledge of PiV from before, as I had an internship period here in my 3^rd year. The choice of where I should write my master thesis was therefore easy as I already knew I would have fantastic supervisors as well as a great social and working environment with PiV at Eitri.

I only just started working on my project and have a long way ahead of me. One challenge I have faced so far is my lack of knowledge of kidney structures and I struggle to identify them on digital WSI. Nevertheless, I have learned a lot already and I am looking forward to the road ahead 😊