In this article: What do the sceptics say? · What does the research show about PCAI? · How can AI perform better? · What limitations are there? · What does it mean for you?
The answer is more nuanced than the intuitive 'no'. In clinical studies published in Medical Image Analysis 2025, the AI system PCAI outperformed all three experienced pathologists in every single comparison on external test datasets from Malmö. That is not a claim, but a statistically significant result from peer-reviewed scientific evidence.
What do the sceptics say – and what is the answer?
The most common objection to AI in diagnostics is that it lacks clinical context and experience. That is a valid observation for many diagnostic situations. But AI also lacks bias, fatigue and variability. An AI model trained on sufficient data always gives the same answer for the same input. And it can be trained on a body of material no single pathologist could accumulate in a working lifetime. The background to the question is gathered on the page about the limitations of the Gleason method.
What does the research show about PCAI?
In a study carried out at UKE Hamburg, Uppsala University Hospital and Spearpoint Analytics, PCAI was evaluated against experienced specialists on three external test datasets. On the MMX cohort from Malmö, PCAI achieved an AUROC5 of 0.869, compared with the expert average of 0.770, a difference of 9.9 percentage points. PCAI outperformed all three individual pathologists in every comparison. On the UPP cohort from Uppsala, PCAI achieved an AUROC5 of 0.672 against 0.660 for ISUP grading. On the UKE-sealed TMA, PCAI outperformed ISUP by 3.3 percentage points in the EOC-Index. All cohorts and margins are gathered under clinical results.
How can AI perform better?
Three factors explain it. Data volume: PCAI is built on a study with 25,591 patients and up to 23 years of follow-up data. Objective training: PCAI is trained directly on patient outcomes, not on human assessment that may in turn be wrong. Consistency: AI does not make mistakes from fatigue or variation in daily form, but gives the same answer for the same input every time. Read more about how PCAI works.
What limitations are there?
PCAI is trained mainly on TMA samples from patients who have undergone a radical prostatectomy. Generalisability to all clinical situations requires continued verification. PCAI is therefore positioned as a second opinion: a complement to, not a replacement for, clinical judgement.
What does it mean for you as a patient?
It means that there is now a scientifically validated tool that gives an independent assessment of the aggressiveness of your cancer, an assessment that does not depend on who is looking into the microscope. Information that can strengthen the basis for your decision, whether it confirms your current treatment plan or raises questions worth taking up with your urologist.
You may also find answers here:
- PCAI vs. Gleason score – a direct comparison PCAI vs. Gleason
- How certain is your prostate cancer diagnosis, really? Article 2 – How certain is the diagnosis
- What a second opinion involves – and what it is not What is a second opinion?
What would an AI analysis say about your sample?
We go through what PCAI can and cannot answer in your case. The meeting is free of charge. If you feel afterwards that this is right for you, you can simply order a second opinion.