Deep-learning Evaluation for Enhanced Prognostics - PSMA PET¶

News¶

Welcome to the DEEP-PSMA challenge.

26 August 2025 - Final Leaderboard Now Available - Final Testing Leaderboard All of the submissions to the testing phase were of a very high standard. Congratulations to team Protactinium however the performance among all of the top three methods was very close. It's been great to see the clever refinements developed for the task and thanks to everybody who has taken part. We look forward to seeing you on 23 September at the MICCAI workshop.

20 August 2025 - Public Submissions Now Closed - Great work to everybody and thank you all for your participation in the Challenge. The submission pages and live leaderboard will no longer update while we finalise the results for the final testing phase. The submission page may be open periodically while we resolve a few technical points for specific users but we'll no longer be updating with new algorithms.

5 August 2025 - Final Testing Phase Open through 19 August - Submissions now open for final leaderboard phase. Each participant/team can make up to 3 submissions with the final working algorithm scored for the leaderboard. For this phase, a brief algorithm description (2-3 pages) and link to the source code are required for eligibility. A reminder, please only use training data from the DEEP-PSMA challenge and review some of the tips which may provide simple performance improvements related to the threshold method used for manually contouring the ground truth cases.

18 July 2025 - Preliminary Validation Phase Open for benchmarking on previously unseen data (10 cases, live leaderboard). We will open submissions for the final validation phase beginning the first week of August with final algorithm submission to close on Tuesday, 19 August. Reminder that for the final phase, you will need to include a brief technical description of the algorithm and link to the source code for consideration in the final leaderboard and challenge awards.

8 July 2025 - Debug Phase Open for testing algorithm functionality. See How to submit page for additional details if not familiar with the Grand Challenge platform.

1 July 2025 - Example Baseline Algorithm and Dockerfile for adapting as submission template are now available here

28 April 2025 - Training data is now available for those who have registered and agreed to the access terms. If you have previously registered on Grand-Challenge please check your message inbox.

We are pleased to announce that we have been selected as one of the MICCAI 2025 registered challenges. Workshop to be held on 23 September.

Please take a moment to familiarise with the Dataset and Dataset Format pages and feel free to reach out with any questions.

Introduction¶

The discovery of the molecular target, prostate-specific membrane antigen (PSMA), has enabled new techniques to both image and target metastatic castration-resistant prostate cancer (mCRPC). Three FDA approved tracers for Positron Emission Tomography (PET) imaging - 68Ga-PSMA-11, 18F-DCFPyL and 18F-rhPSMA-7 - now enable more accurate detection of metastatic spread throughout the body and also selection of patients for radioligand treatment.

Treatment of mCRPC with the radiopharmaceutical 177Lu-PSMA-617 (LuPSMA) presents opportunities to improve quality-of-life and life expectancy in men who have progressed after receiving chemotherapy or hormone therapies. The likelihood of response to LuPMSA treatment is related to the ability to concentrate therapeutic radiation into prostate cancer cells; a factor related to the expression of PSMA receptors in target tissues which can be quantified by PSMA PET imaging. Quantitative imaging biomarkers from PSMA and FDG imaging have been shown to be predictive of response to 177Lu-PSMA therapy.

We host the DEEP-PSMA (Deep-learning Evaluation for Enhanced Prognostics - Prostate Specific Membrane Antigen) challenge to benchmark artificial intelligence models for marking up the extent of disease on PSMA PET/CT and FDG PET/CT in order to automate scoring quantitative biomarkers relevant for targeted radionuclide therapy.

DEEP-PSMA will be hosted as part of the MICCAI 2025 conference and is supported by the The Prostate Theranostics and Imaging Centre of Excellence (ProsTIC) at the Peter MacCallum Cancer Centre, Australia and through the Prostate Cancer Foundation (PCF).

Task¶

Automatic whole-body tumor lesion segmentation in both PSMA PET/CT and FDG PET/CT imaging on a database of 100 patients (training database) managed at a single centre:

• Accurate total-body tumour burden segmentation

• Avoidance of false positive regions (liver*, kidneys, brain, bladder, salivary glands)

• Potential utilisation of paired image data (PSMA PET/CT with FDG PET/CT data)

Preliminary validation includes 10 cases for fine-tuning algorithms using live leaderboard.

Final Testing performed on 40 (held-out final test set) all originating from the same clinic as the training database.

Publications involving manual annotation of total tumour volume for determination of predictive biomarkers in mCRPC can be found in Ferdinandus et al. and Buteau et al..