Thesis on clinically reliable AI for prostate cancer pathology
Prostate cancer is one of the most common cancers in men globally. Diagnosing prostate cancer involves examining tissue samples under a microscope, a process that can be inconsistent and yield different results by different pathologists for the same sample, potentially affecting patient treatment. In some unclear cases, extra testing called immunohistochemistry (IHC) is used, but it adds cost, time, and lab work.
In a new thesis from Karolinska Institutet, PhD student Nita Mulliqi at the Department of Medical Epidemiology and Biostatistics, developed and tested artificial intelligence (AI) models that can help pathologists to analyse digital images of prostate biopsies. These AI systems were trained and tested on nearly 100,000 biopsies from over 7,000 patients across 11 countries.
What are the most important results in your thesis?
"We found that our AI systems can detect cancer, grade its severity, and reduce the need for IHC testing, all while performing on par with expert pathologists. The results show that these AI models work well even when the images come from different hospitals and scanners or represent rare tissue morphologies, making them suitable for real-world use. Our work helps bring AI one step closer to everyday clinical practice, offering reliable and accurate prostate cancer diagnosis for patients worldwide.”
Why did you become interested in this topic?
"To be honest, it was a bit of luck combined with a strong interest. I’ve always been drawn to the intersection of AI and medicine. As a computer engineer, I was always looking for ways to apply my skills to areas where they could truly make a difference. I first had the opportunity to work on AI applications in colorectal cancer, which sparked my fascination with digital pathology. Then, by chance, I came across this position focusing on AI for prostate cancer, and I knew it was the right next step. I’m grateful I took it, it has been an incredibly rewarding journey."
What do you think should be done in future research?
"I believe future healthcare should focus on prospective clinical trials and integrating AI tools into clinical workflows in a responsible, transparent, and evidence-based way. It’s not just about creating high-performing models, it’s about ensuring they are robust, generalisable, and accepted by clinicians. Research should also prioritize collaboration across countries and institutions, so that AI systems are tested in diverse settings and serve global populations. Finally, we need to invest in digital infrastructure and training so that healthcare professionals are empowered to work effectively alongside AI," says Nita Mulliqi.