Fiona Maclean1, Claudia Petrucco2, Meridith Peratikos, Amanda De Souza2
1 Douglass Hanley Moir, Franklin.ai, Sydney, NSW, Australia
2 Franklin.ai, Sydney, NSW, Australia
Abstract for ECP 2025, Vienna, Austria
Background & Objectives
Both ISUP and GUPS in 2019 recognised the importance of reporting the presence of intraductal carcinoma (IDC-P) and cribriform architecture in prostate core biopsies (PCB). The presence of these patterns is correlated with a higher risk of biochemical recurrence. However, the agreement between pathologists for both of these entities is suboptimal. Recent consensus meetings held by ISUP and GUPS have attempted to further refine the diagnostic criteria for IDC-P, highlighting the importance of identification This study evaluates our AI model’s ability to assist pathologists in detecting these features, enhancing consistency.
Methods
We performed a clinical validation study to evaluate our AI model’s ability to augment pathologists’ performance identifying the presence of IDC-P and cribriform architecture on PCB specimens using ground truth data as the benchmark. Classification performance was measured using Area Under the Curve (AUC). Reader agreement on classification of clinical findings was assessed using the Intraclass Correlation Coefficient (ICC). A two-way random effects model measured absolute agreement.
Results
AI-assisted pathologists demonstrated superior diagnostic performance. For Acinar Gleason pattern 4 carcinoma with cribriform architecture, AI-assisted pathologists achieved an AUC of 0.870 compared to 0.866 unassisted. For IDC-P, AI assistance resulted in an AUC of 0.840 versus 0.785 unassisted. Interobserver agreement also improved with AI assistance. The ICC for cribriform architecture agreement improved from 0.704 to 0.730 while IDC-P increased from 0.566 to 0.634. These findings highlight both enhanced diagnostic accuracy and increased agreement among pathologists when utilising AI assistance.
Conclusion
We demonstrate that AI-assisted pathologists achieve superior diagnostic accuracy and consistency identifying both IDC-P and cribriform architectures. Patients with IDC-P are ineligible for active surveillance (AS), and the presence of cribriform growth would dissuade some urologists from AS. Thus, increased accuracy and agreement in diagnosing these architectures may impact treatment pathways.
