Multi-stage AI analysis system to support prostate cancer diagnostic imaging

No author information is available for this submission.




The authors are in the process of completing this submission.

Please try again later.


An Artificial intelligence (AI) system was developed to support interpretation of pre-biopsy prostate multiparametric MRI (mpMRI), aiming to improve patient selection for biopsy, biopsy target identification, and productivity of segmentation and reporting, in the prostate cancer diagnostic pathway.

For segmentation, the system achieved 92% average Dice score for prostate gland segmentation on held-out test cases from the PROMISE12 dataset (10 patients).

For biopsy assessment, the system identified patients with Gleason ≥3+4 clinically significant prostate cancer (csPCA) with sensitivity 93% (95% CI 82-100%), specificity 76% (64-87%), NPV 95% (88-100%), and AUC 0.92 (0.84-0.98), using biparametric MRI (bpMRI) data from the combined PROSTATEx development validation and test sets (80 patients). Performance on the held-out PROSTATEx test set (40 patients) was higher. Radiologists in major studies achieved 93% per-patient sensitivity at specificity from 18-73%. Equivalent sensitivity is reported for comparable AI/CAD systems at specificity from 6%-42%.

For biopsy targeting, the system identified lesions containing csPCa in the PROSTATEx blinded test set (208 lesions, 140 patients) with AUC 0.84/0.85 with bpMRI/mpMRI data respectively.

The AI system shows promising performance compared to radiologists and the literature. Further development and evaluation with larger, multi-centre datasets is now planned to support regulatory approval of the system.


Ask a Question

Get involved to find out more about this Presentation.

All Comments

Create an Account or Log In to participate in the discussion

Loading discussions