EuSoMII Virtual Annual Meeting 2020 Book of Abstracts
Jakub Suchánek, A. Rix, Aman Mehan, Chris Doran, Anwar R. Padhani, Christof Kastner, Tristan Barrett, Evis Sala, Oliver Díaz, G. Guidi, Federica Zanca, A Olthof, Hein Stallmann, Mohammad Hosein, Rezazade Mehrizi, Erik Ranschaert, Giovanni Rinaldi, Camila González, Anirban Mukhopadhyay, A. V. Petraikin, Zhanna Belaya, V. A. Gombolevsky, Mikhail Belyaev, К. А. Сергунова, Z. R. Artyukova, Anastasia Kiseleva, В. В. Кондратенко, Maxim Pisov, Е. С. Ахмад, Д. С. Семенов, Nikita Kudryavtsev, Lyudmila Nizovstova, Anton V. Vladzymyrskyy, Sergey MorozovInsights2021 · DOI
A new artificial intelligence (AI)-based system designed to support clinical interpretation of prostate MRI shows promising performance compared to traditional radiology studies. This multi-stage system investigates whether AI can improve the specificity of patient selection for biopsy, identify biopsy targets, and assist in prostate gland segmentation. Evaluated on public datasets like PROSTATEx and PROMISE12, the system achieved a sensitivity of 93% and a specificity of 76% for ruling out clinically significant prostate cancer (csPCa). These results suggest that AI can significantly reduce unnecessary biopsies while maintaining high diagnostic accuracy.
Introduction: Addressing the Specificity Gap in Prostate MRI
The current diagnostic pathway for prostate cancer relies heavily on Magnetic Resonance Imaging (MRI) to determine which patients require a biopsy. While radiologists demonstrate high sensitivity in detecting potential malignancies, they often face challenges with low specificity—leading to "false positives" where benign areas are flagged as suspicious. This results in many patients undergoing invasive and often unnecessary biopsies.
Furthermore, manual tasks such as segmenting the prostate gland for PSA density calculations are time-consuming and subject to inter-observer variability. There is a critical need for automated tools that can reliably rule out cancer, identify biopsy targets, and perform routine segmentation tasks.
The EuSoMII 2020 meeting focused on digital transformation in radiology.
The research presented at the EuSoMII Virtual Annual Meeting 2020 by Suchanek et al. details a multi-stage AI-based system designed for end-to-end prostate MRI workflow. Rather than a "black-box" approach, the system decomposes the diagnostic process into discrete steps:
Automated Segmentation: Outlining the prostate gland to support PSA density evaluation and fusion biopsy.
Detection and Classification: Identifying lesions and classifying them as clinically significant prostate cancer (csPCa), defined as Gleason score ≥3+4.
The system was trained and validated using major public benchmarks: the NCI-ISBI 2013 Challenge, PROMISE12 (for segmentation), and PROSTATEx (for lesion detection). Performance was evaluated on held-out test sets using both biparametric MRI (bpMRI) and standard multiparametric MRI (mpMRI).
Key Results and Performance Metrics
The AI demonstrated robust performance across three primary diagnostic tasks:
Patient Selection (Rule-out test): On the combined PROSTATEx validation and test sets (80 patients), the system achieved 93% sensitivity and 76% specificity (AUC 0.92). This specificity is notably higher than the 18-68% range typically reported for radiologists in major studies.
Target Identification: For lesion-level detection, the system showed 94% sensitivity and 71% specificity (AUC 0.89).
Automated Segmentation: The system achieved an average Dice score of 92% on the PROMISE12 held-out test set, matching the state-of-the-art for automated gland contouring.
Discussion and Clinical Implications
The most significant finding is the AI's ability to maintain high sensitivity while doubling the specificity seen in some clinical environments. By accurately identifying "true negatives," this tool could help a substantial number of patients avoid unnecessary biopsies without missing clinically significant disease.
Additionally, the system performed similarly well on bpMRI as it did on mpMRI. This suggests that AI-supported bpMRI could offer a faster, more cost-effective screening method that avoids the risks associated with gadolinium contrast agents.
Conclusion and Future Outlook
The results indicate that this multi-stage AI system has promising specificity, sensitivity, and negative predictive value (NPV) to support the prostate cancer diagnostic pathway. While currently limited to retrospective research use, its development by Lucida Medical points toward future clinical integration. Larger, multi-center prospective trials will be essential to validate these metrics across diverse clinical settings and scanner types.
Disclosure: AW Rix and E Sala co-founded Lucida Medical, the company developing this AI system.
Summary updated
Abstract
A new artificial intelligence (AI)-based system, designed to support clinical interpretation of prostate MRI, shows promising performance compared to radiology studies and CAD/AI literature. Purpose/Objectives: Investigate whether AI could improve specificity of patient selection for biopsy and biopsy target identification, and assist segmentation, in the prostate cancer diagnostic pathway. Methods and materials: A multi-stage AI-based system was developed for MRI analysis, and trained with the NCI-ISBI 2013 Challenge, PROMISE12 and PROSTATEx datasets, split into training, validation and held-out test sets. Clinically significant prostate cancer (csPCa) was defined as Gleason3+4 disease. Accuracy metrics were computed on validation and held-out test sets, and compared with literature on prostate MRI studies and CAD/AI models. Results: To support patient selection for biopsy as a rule-out test, sensitivity identifying patients with csPCa was 93% (95% CI 82-100%), specificity 76% (64-87%), NPV 95% (88-100%), and AUC 0.92 (0.84-0.98), evaluated with biparametric MRI (bpMRI) data from the combined PROSTATEx validation and held-out test sets (prevalence 35%, 80 patients). Performance was higher on the held-out test set (40 patients). Similar AI/CAD publications report 93% sensitivity using held-out/blinded data at specificity between 6%-42%. In major studies, radiologists' sensitivity per-patient was 88-93%, specificity 18-68%, and NPV 76-97%, with Likert/PI-RADS 3 defined as positive. Note that methodological and dataset differences and test set size limit comparisons. For identifying biopsy targets, the AI system detected csPCa lesions with per-lesion sensitivity 94% (85-100%), specificity 71% (61-89%), NPV 97% (93-100%), and AUC 0.89 (0.83-0.95), in the same combined PROSTATEx development validation/test set (128 lesions, 80 patients). Performance was higher on the held-out test set. For prostate gland segmentation to support analysis, PSA density evaluation, and fusion biopsy, the system showed 92% average Dice score, against radiologist ground-truth segmentations, on held-out test cases from the PROMISE12 dataset (10 patients). This is comparable to the stateof-the-art. The AI system performed similarly in the above tasks when evaluated using multiparametric MRI (mpMRI) data. Conclusion: The results suggest the AI system has promising specificity, sensitivity and NPV to help csPCa-free patients avoid biopsy, to support biopsy targeting, and to provide high-quality automated segmentations. Disclosure: AW Rix and E Sala co-founded Lucida Medical, the company developing this AI system. While regulatory approvals are in progress, the system is not currently available and is described here in retrospective research use.
BibTeX
@inproceedings{suchanek2021eusomii,
author = {Jakub Suchánek and A. Rix and Aman Mehan and Chris Doran and Anwar R. Padhani and Christof Kastner and Tristan Barrett and Evis Sala and Oliver Díaz and G. Guidi and Federica Zanca and A Olthof and Hein Stallmann and Mohammad Hosein and Rezazade Mehrizi and Erik Ranschaert and Giovanni Rinaldi and Camila González and Anirban Mukhopadhyay and A. V. Petraikin and Zhanna Belaya and V. A. Gombolevsky and Mikhail Belyaev and К. А. Сергунова and Z. R. Artyukova and Anastasia Kiseleva and В. В. Кондратенко and Maxim Pisov and Е. С. Ахмад and Д. С. Семенов and Nikita Kudryavtsev and Lyudmila Nizovstova and Anton V. Vladzymyrskyy and Sergey Morozov},
title = {{EuSoMII Virtual Annual Meeting 2020 Book of Abstracts}},
booktitle = {Insights},
year = {2021},
doi = {10.1186/s13244-021-00975-x},
}