PCAI clinical results: AI that outperforms experienced pathologists on all external test cohorts

The study validated PCAI against three independent external test cohorts: Uppsala, Malmö and Hamburg.

• Uppsala (UPP): biopsies from Akademiska sjukhuset (123 patients, 683 images)
• Malmö (MMX): biopsies from a Swedish clinic (269 patients, 578 images)
• Hamburg (UKE-sealed): tissue samples (TMA) graded by pathologists at UKE Hamburg

Martini-Klinik in Hamburg is a training data partner and is therefore not included as a test cohort. That is an important distinction when the results are to be assessed.

• 21.2 percentage points better predictive ability than the weakest individual expert on the MMX cohort (AUROC5)
• 9.9 percentage points better than the expert pathologists' average on the MMX cohort (AUROC5 0.869 versus 0.770)
• Outperformed ISUP-grading pathologists on all external test cohorts
• Consistently reproducible, the same biopsy always gives the same PCAI index
• Robust to variations in laboratory routines (section thickness, staining, scanner)

Pilot studies show that PCAI outperforms experienced pathologists in diagnostic precision on all external test cohorts and reduces the "grey zones" that previously created uncertainty in decision-making.

The grey zone in question is above all the ISUP 2 group, where today's method leaves the most room for differing assessments Limitations of the Gleason method. It is also where an incorrect decision costs the most: either an operation that was not needed, or a tumour that was allowed to grow Surveillance vs. surgery.

A method of analysis that gives different answers depending on which laboratory prepared the sample is not good enough as a basis for decisions. PCAI retains its accuracy despite variation in staining protocol, section thickness and scanner model. It is one of the most underestimated parts of the result, and the one that makes the method usable in real clinical operation Implementation.

Would you like to understand how the analysis is carried out behind the figures? Read how PCAI works and how the method differs from Gleason grading PCAI vs. Gleason. The complete references are available under research and publications Research & publications.