Introduction
Cribriform morphology within Gleason pattern 4 prostate cancer represents an aggressive feature associated with adverse clinical outcomes, making its accurate identification critical for risk stratification and patient management. The current study evaluated the performance of a clinical AI algorithm (Ibex Prostate) in detecting cribriform patterns in hematoxylin and eosin stained core needle biopsy samples.
Materials and Methods
A multi-center, standalone study was conducted using whole slide images, collected from three institutions in France and Israel. The ground truth was established through the consensus of two independent pathologists, with an addition of a third pathologist in cases of disagreement. The primary study endpoint was defined as area under the curve (AUC) with a performance goal of 0.9, while the secondary endpoint was sensitivity and specificity with thresholds of 80%.
Results
Study cohort included 335 slides, with cribriform patterns present in 143 specimens and 192 without. The algorithm achieved AUC of 0.95 (95% CI: 0.92–0.97), exceeding the primary performance goal. Additionally, the algorithm demonstrated a sensitivity of 81.12% (95% CI: 73.08%–89.16%) and a specificity of 92.71% (95% CI: 88.71%–96.71%), achieving secondary endpoints. Performance was consistent across different sites, ages, and Gleason scores.
Conclusions
The study successfully validated the AI algorithm in detecting cribriform morphology, meeting all primary and secondary endpoints. The high diagnostic accuracy demonstrates the algorithm’s ability to support pathologists in identifying aggressive patterns of prostate cancer, offering an important tool for computer-aided diagnosis in pathology practice.
