ECR 2026: Poster Presentation
Brittaney Everitt , Valerio Fortunati, Jiye Kim, Jim Tol, Abigail Ko, Guus Kolpa, Felix Leufkens, Bryan Haslam
Abstract
Purpose
While prostate MRI can improve prostate cancer (PCa) detection and enable more accurate MRI-guided fusion-biopsy, these gains require radiologist expertise. Shortages of expert radiologists and inter-reader variability can lead to inconsistent diagnoses and suboptimal biopsy targeting. AI could improve radiologists’ interpretations. We evaluated an integrated AI web-based platform providing fully automated lesion segmentation and lesion-level risk classification, assessing effects on PCa detection, localisation and inter-reader consistency.
Materials or Background
A blinded, fully crossed multi-reader multi-case study evaluated 12 board-certified radiologists interpreting 250 prostate MRI exams (109 cancer negative, 141 Gleason score ≥6). Each exam was read with and without AI assistance, separated by a four-week washout period. Readers identified and segmented suspicious lesions and recorded PI-RADS (3-5) and level-of-suspicion (LoS) scores (0-100). Exam-level AULROC was computed from lesion-level LoS scores, requiring correct localisation of at least one lesion for cancer cases. Lesion sensitivity was the proportion of malignant lesions correctly identified and localized. Lesion segmentation performance relative to ground truth and inter-reader agreement were evaluated using the Dice similarity coefficient (DSC).
Results
AI assistance increased reader PCa detection, with mean AULROC increasing by 15% from 0.614 unaided to 0.705 aided (ΔAUC=0.092, 95%CI: 0.055-0.128). On average, readers detected 27% more cancer lesions with AI (0.480 to 0.610). Segmentation agreement improved with AI relative to ground truth (median DSC 0.508 [IQR 0.346-0.630] vs 0.549 [IQR:0.439-0.639]) and across readers (median DSC 0.465 [IQR:0.242-0.628] vs 1.0 [IQR:0.866-1.0]), reducing reader variability by 65%.
Conclusion
By providing fully automated lesion segmentations and lesion-level risk assessment, the integrated AI platform enabled more consistent and accurate interpretation, highlighting its potential to standardise and improve prostate MRI assessment and support accurate fusion-biopsy planning.
Disclaimers:
Disclaimers: DeepHealth Prostate Suite comprises multiple applications including Quantib Prostate, DeepHealth Prostate AI and DeepHealth Viewer. Quantib Prostate is Manufactured by Quantib BV and distributed by DeepHealth, Inc, in the US. DeepHealth Viewer is manufactured by eRAD, Inc. and distributed by DeepHealth, Inc. DeepHealth Prostate AI is FDA 510(k) pending Any claims made about Prostate Suite may reference claims associated with its individual components. Not all products and functionalities are commercially available in all countries. For clearance and commercial availability in your geography of functionalities listed and compatibility with other systems, please contact your account manager.
Any claims made about Prostate Suite may reference claims associated with its individual components.
Not all products and functionalities are commercially available in all countries. For clearance and commercial availability in your geography of functionalities listed and compatibility with other systems, please contact your account manager.