Introducing Radwan - one of our dedicated medical students in the PiV project

From the start of my medical studies I had a strong interest in technology and medical technology specifically. Over the years I have watched new technological breakthroughs and the growth of "Artificial intelligence", and it made me quite inquisitive about the future. When I saw that PiV "Pathology services in the Western Norway Health Region – a centre for applied digitization" had an initiative for medical students to perform research in digital pathology, it rekindled that interest in me.

I was especially pleased to find that one of my mentors, Sabine, was my teacher in the kidney pathology course on the fourth year of medical school. I can't express how grateful I am to have Hrafn and Sabine as my supervisors. They always try to help as much as they can and also urge me to push myself and do new things, such as giving presentations at various internal and external meetings.

In my thesis, I assess the performance of the open-source software "QuPath" in detecting and calculating the tumour marker “Ki-67” in Gastrointestinal Neuroendocrine Neoplasms “GI NENs”. Since the classification of GI-NEN classification is based on Ki67 index, accurate detection and calculation of Ki67 are pivotal in distinguishing between different grades of the cancer.
What was challenging was learning something completely new to me. To use QuPath, I had to watch numerous lessons and video tutorials, and read various articles. It was, nevertheless, not impossible.

In my research, I used 28 training whole-slide images “WSI” with NEN from various locations in the GI tract. First, I used a pre-trained cell detection model called “Stardist” for cell detection and then I manually trained a classifier to differentiate between detected tumor and non-tumor cells, and Ki67 positive tumor cells and the Ki67 negative ones. After training and creating a classifier, I will be using “Stardist” along with the trained classifier to test the model on 10 test cases with 80 region of interest “ROIs”, that’s 8 ROIs from each case. 
The results will be  compared with the performance of other previously tested commercially available systems like Aiforia and ImageScope.

In general, I hope that my project yields useful results that will aid in the advancement of digital pathology implementation in clinical practice, and help in future discussions or decisions about the implementation of QuPath in clinical practice.