SNIC 2017/1-326


SNAC Medium

Principal Investigator:

Carolina Wählby


Uppsala universitet

Start Date:


End Date:


Primary Classification:

20603: Medical Image Processing

Secondary Classification:

10203: Bioinformatics (Computational Biology) (applications to be 10610)



When a prostate cancer is diagnosed, the Gleason score is the single most important marker of prognosis and for treatment decisions. The Gleason score is assigned by a pathologist based on how prostate cells look under a microscope. The inter-pathologist variance in Gleason scoring is high and there is a shortage of trained uro-pathologists. A high-quality decision aid for Gleason scoring of prostate biopsies, based on image analysis and genomic profiling, would reduce both inter-pathology variance and pathology workload. The STHLM3 study was a prospective and population-based prostate cancer diagnostic trial conducted in 2012-2015 and including almost 60,000 men of age 50-69 from Stockholm, Sweden. 7417 of the study participants received a prostate biopsy (10-12 cores depending on prostate volume), all Gleason graded by one single, world-class pathologist (Professor Lars Egevad). In total, we have more than 80,000 biopsy cores from STHLM3 being scanned using whole slide imaging. We are currently also developing a prognostic genomic profile for prostate cancer. The combination of the size and quality of the STHLM3 data and the prostate cancer genomic profile is unique and provides the foundation for taking prostate cancer digital pathology to the next level. Specifically, the following clinical aims are interesting to tackle: * Use image analysis and machine learning/AI to reliably identify biopsy cores without cancerous cells. * Identify areas on the biopsy core where cancer cells are present, so that those areas can be worked up for genomic profiling. * Support and augment Gleason scoring by combining image analysis, genomic profiling and pathology assessment, using PSA relapse after initial treatment as an endpoint.