Dr Fiona Maclean 1,2
1 Faculty of Medicine, Macquarie University, Macquarie Park, NSW, Sydney, Australia
2 Franklin.ai, Sydney, NSW, Australia
Oral presentation (link) at 36th European Congress of Pathology, 7 - 11 September 2024, Florence, Italy
Prostate cancer poses a significant healthcare burden, with a global projected increase to 2.9 million new cases annually by 2040. Concurrently, there's a chronic worldwide pathologist shortage. This study validates an Artificial Intelligence (AI) model as a diagnostic support tool.
A Multi-Reader Multi-Case (MRMC) study, including 30 pathologists was conducted using an enriched test set of over 1,700 H+E stained whole slide images (WSI) representing individual patients. Pathologists interpreted ground truthed test set images, rating their diagnostic confidence in 45 clinical findings from a predefined ontology. A statistical analysis was performed in comparison to the AI model interpretation.
The classification performance of the AI model achieved statistical non-inferiority compared to unassisted pathologists for all 45 findings. Furthermore, statistical superiority was shown for 98 % of needle core biopsy (NCB) and 91% of transurethral resection of prostate (TURP) findings. The AI model successfully identified malignancy (AUC 0.97) compared to pathologists (AUC 0.92). In NCB the model identified acinar adenocarcinoma Gleason patterns 3, 4 and 5 with AUCs of 0.96, 0.98, 0.97 respectively, compared to pathologists with AUCs of 0.85, 0.93, 0.81. Other clinical findings where the model outperformed pathologists include cribriform architecture, adenocarcinoma with ductal features, and prognostic factors including intraductal carcinoma, extraprostatic extension, lymphovascular invasion and perineural invasion.
This study validates the performance of the AI model in detection of all included prostate findings on H+E stained WSIs and supports its use as a decision support tool for pathologists in detection of prostate adenocarcinoma and associated findings. Given the projected surge in prostate cancer over the coming years, and the fundamental role pathology has in diagnosis and determination of treatment, this model has potential for great utility.
DOI: https://doi.org/10.1007/s00428-024-03880-y
The results presented in this abstract are based on an early version of our AI model therefore outcomes reported here are subject to change as the model undergoes further development and validation.
Interested in learning more? Get in touch