60/100

Moderate evidence

Cite
More
EmailLinkedInX

Prostate cancer study of 503 men finds AI model can flag biopsies likely to understate grade

Study title: Predicting ISUP upgrading from biopsy to radical prostatectomy using an interpretable machine learning model with mpMRI-diagnosed prostatitis: a retrospective study

Evidence type
Observational study
Sample
n = 503
Sources
1 sourcePeer-reviewed
Published
2 min read

Study title: Predicting ISUP upgrading from biopsy to radical prostatectomy using an interpretable machine learning model with mpMRI-diagnosed prostatitis: a retrospective study

A single-center study of 503 consecutive men who had a transperineal prostate biopsy followed by laparoscopic radical prostatectomy built a machine learning model to predict whether their cancer grade would be upgraded after surgery. The random forest model scored an AUC of 0.823 and average precision of 0.799 on an independent test set.

Read the study abstract

The International Society of Urological Pathology (ISUP) grade upgrading from biopsy to radical prostatectomy (RP) occurs in approximately 30%-40% of prostate cancer patients and may compromise treatment planning. Although multiparametric MRI (mpMRI) improves lesion characterization, prostatitis-like signal changes can mimic clinically significant cancer and confound radiological interpretation; however, no predictive model has incorporated mpMRI-diagnosed prostatitis as a variable for ISUP upgrading. This singl…

Predicting ISUP upgrading from biopsy to radical prostatectomy using an interpretable machine learning model with mpMRI-diagnosed prostatitis: a retrospective study
"20250808 Intensive care unit bed" by Ismael Olea, CC BY 4.0.

The gain

In a retrospective study of 503 men who had prostate biopsy then surgery, a random forest model predicted which cases would be upgraded in grade with an AUC of 0.823, and MRI-diagnosed prostatitis added predictive value beyond PI-RADS scoring.

In brief

Grade upgrading from biopsy to surgery happens in roughly 30%-40% of prostate cancer patients and can compromise treatment planning. In this cohort it occurred in 195 patients (38.8%). MRI-diagnosed prostatitis was more common in upgraded cases, at 27.2% versus 11.4%.

The model is meant to help identify patients at heightened risk of having their cancer undergraded on biopsy. It was developed and tested within one hospital's records, so its performance in other settings remains unestablished.

Main points

  1. The study enrolled 503 consecutive men treated between May 2021 and October 2024, all biopsied with a 12+1-core transperineal method plus cognitive MRI-targeted cores.
  2. ISUP upgrading occurred in 195 patients, or 38.8%.
  3. MRI-diagnosed prostatitis was found in 27.2% of upgraded cases versus 11.4% of non-upgraded cases (P < 0.001).
  4. Six machine learning classifiers were trained after LASSO feature selection and tested on an independent set from a 7:3 split; random forest scored highest.
  5. The random forest reached ROC-AUC 0.823 and average precision 0.799.
  6. SHAP analysis ranked basal tumor location first (0.145), then PI-RADS ≥4 (0.121), percentage of positive cores (0.100), maximum tumor diameter (0.062) and prostatitis (0.040).

The full story

Researchers enrolled 503 consecutive men who had a 12+1-core transperineal ultrasound-guided systematic biopsy with cognitive MRI-targeted cores, followed by laparoscopic radical prostatectomy between May 2021 and October 2024. Upgrading in ISUP grade between biopsy and surgery affected 195 of them (38.8%).

Features were chosen with LASSO logistic regression and 10-fold cross-validation. Six machine learning classifiers were trained and compared on an independent test set using a 7:3 split, and random forest had the highest weighted composite score, with ROC-AUC 0.823 and average precision 0.799.

SHAP analysis showed the strongest predictors were basal tumor location, PI-RADS ≥4 and percentage of positive cores. MRI-diagnosed prostatitis ranked fifth with a mean SHAP value of 0.040, but the authors say it adds predictive value beyond PI-RADS scoring.

Sources

1 on file
  1. Asian Journal of Andrologydoi:10.4103/aja202643
    Peer-reviewed2026-11-17

Talk it over

Three questions to start on, two more if you have time
The model reached an AUC of 0.823 but was tested only on a 7:3 split of the same 503 men from one hospital. How much would that limit your trust in it elsewhere?

Start with a question

Prostate cancer study of 503 men finds AI model can flag biopsies likely to understate gradehttps://truvace.com/item/predicting-isup-upgrading-from-biopsy-to-radical-prostatectomy-using-an-interpre

  1. The model reached an AUC of 0.823 but was tested only on a 7:3 split of the same 503 men from one hospital. How much would that limit your trust in it elsewhere?

    Answer in the thread
  2. Grade upgrading occurred in 195 patients, or 38.8%. If you were a man facing biopsy then surgery, how would you want a high-risk flag from this model to shape your treatment planning?

    Answer in the thread
  3. Which result would convince you the model is ready for use in clinics: testing at hospitals outside the study, or a comparison with doctors' own judgment of upgrade risk?

    Answer in the thread
If you have more time
  • Basal tumor location, PI-RADS ≥4 and percentage of positive cores ranked above prostatitis in the SHAP analysis. What would you ask the authors about why they still stress prostatitis?

    Answer in the thread
  • All 503 men were treated between May 2021 and October 2024 at one center. What would you look up to learn how biopsy methods differ at other hospitals?

    Answer in the thread

What readers say