Artificial intelligence for reproducible MRI diagnosis of prostate cancer

Diagnosis of prostate cancer (PCa), which is the most common cancer with the highest mortality among males in Sweden, has for decades been established by trans-rectal ultrasound (TRUS) guided systematic, i.e., non-targeted trans-rectal biopsies.

A scientifically validated paradigm shift has recently turned diagnostics towards pre-biopsy magnetic resonance imaging (MRI) followed by MRI-targeted transrectal biopsies of suspected lesions. Accordingly the policy "MRI first" has been adopted by the Swedish National Prostate Cancer Guidelines.

The development of Artificial Intelligence (AI) solutions for MRI may play an important role to improve diagnostic accuracy and in dealing with the significant increase in the number of exams that will come with future screening programs that follow this policy. This should also be viewed in the context of the limited number of adequately trained radiologists.

The proposed multi-institutional approach uses large databases, which ensures machine training with images obtained with different imaging protocols and different equipment, thus avoiding the significant bias that results with databases that do not reflect the variety that arises when many screening sites are involved.

Furthermore,  the improvement  from the integration of reproducible quantitative values will be investigated.

It is expected that the proposed research can make a signifcant contribution towards more reliable prostate cancer diagnostics with better outcome.