Title image for AI grading of prostate cancer

AI grading of prostate cancer

Spearpoint Analytics has used the AI compute and data sharing services at the NBIS infrastructure unit AIDA Data Hub to improve the diagnostic tools available for prostate cancer using AI and machine learning.

The startup company Spearpoint Analytics develops an outcome-based analysis (PCAI – Prostate Cancer Aggressiveness Index) that predicts the severity of prostate cancer by analyzing microscopy images of tissue sections.

“We want to improve the analysis and develop a system that does not suffer from the subjectivity of today’s analysis. The system in general use today (Gleason) was developed in the 1950s and analyzes the deformation of prostate glands in tissue samples. With our approach, by training our algorithms directly against patient outcome rather than human labeling, the system is capable of adding more biomarkers available in the image. This way, we achieve superhuman performance,” says Peter Walhagen, a developer at Spearpoint Analytics.

In cancer, early diagnosis and treatment is key for patient prognosis and survival. “However, today, there is a shortage of pathologists and radiologists, and as a result, patients often have to wait longer than ideal. Tools like these will help ease the work of pathologists, and help increase survival of cancer patients,” says Joel Hedlund, head of AIDA Data Hub.

The idea for the outcome-based analysis was born from a collaboration between Uropathology Professor Christer Busch and Image Analysis Professor Ewert Bengtsson. The technology demands a lot of data and was developed using a dataset from Martini Klinik with tissue sections from more than 20,000 patients and the added component of 20 years of outcome data for these patients.

“Currently, we are in the research stage and we have newly recruited a team of researchers at the Universitätsklinikum Hamburg-Eppendorf in Germany, where we are further developing the algorithm to validate it against even larger datasets. We must validate our algorithm at at least three to four different centers to release a product – the algorithm must be robust enough to cope with different conditions on different sites,” says Peter Walhagen.

“It is crucial to understand when the algorithms work well and when they do not, in order to build trust within the medical profession so the technology can be taken into clinical use,” says Joel Hedlund.

The focus for the future is to understand what the analysis does and how it differs from the currently performed analysis, which measures how much the glands have been deformed. Preliminary results show that the disruption in the connective tissue – desmoplasia – between the prostate glands has an impact on the cancer risk.

The AIDA data sharing policy is a resource that describes not only the data sharing processes in AIDA but also the common practice in handling and sharing medical imaging data for research in Sweden and similar countries. Open data sharing and FAIR data principles are crucial for transparency, moving the research field forward faster.

“The collaboration with AIDA has been extremely helpful, and Joel has been key to server access. I have also appreciated the research collaboration – it is an incredibly rewarding community to be part of, where you share yourself even with those you compete with. That’s why we share data. The competition that exists for the research moves us forward, and we can further develop our analysis. We want to make validation data public to be able to compare against others”, says Peter Walhagen. “What we have done, would not have been possible to do on any other platform”.

Spearpoint Analytics’ work within the AIDA Data Hub has been a visible project, where true FAIR and open science spirit have permeated the work.

“It will always be the researchers and developers who are the best at collaboration who will win in global competitiveness. Large-scale consortium research always has a higher impact than smaller ‘data hugging’ projects”, finishes Joel Hedlund.