Resources
Zenodo Dataset
Official challenge data
GitHub Baseline
Starter code & Virtual Scanner
OpenReview
Paper submission (will be available soon)
Challenge Information
Unlike traditional medical image synthesis tasks that focus on visual plausibility (using GANs or Diffusion models), SynthOCT requires participants to solve an inverse physics problem: reconstructing the underlying tissue microstructure.
Participants must generate "Digital Phantoms" — spatial distributions of optical scatterers. These phantoms are processed by our fixed, physics-based Virtual Scanner to generate synthetic OCT B-scans. The quality of synthesis is evaluated not just on visual similarity, but on physical consistency using Optical Attenuation Coefficients (OAC) and Speckle Statistics.
Dr. Lev Matveev (Russian Academy of Sciences)
PhD. Samuel Remedios (Johns Hopkins University)
Prof. Azhar Zam (New York University)
Prof. Taimur Hassan (Abu Dhabi University)
Technical Organizers (Registration, API, Submission and Evaluation):
Evaluation & Benchmark Leads:
Denis Nikoshin (HSE University)
Daniil Mikhailenko (HSE University)
Team Registration: Due June 7 (via registration form).
Preliminary Submission: Due June 21 (upload digital phantoms).
Final Submission: Due July 1 (Code and/or model, or a link to them, via this platform and full paper via OpenReview link).
Initial Decision: Notifications will be sent out by July 20.
Final Decision: The final decision, following the rebuttal phase, is due on August 10.
Code Availability: Please make sure that your complete code is made public in a repository (e.g., GitHub) if your paper is accepted. Acceptance is provisional; your paper will not be published if the code is not publicly available by August 10.
MICCAI 2026 venue update: The conference has been relocated from Abu Dhabi to Strasbourg, France. Participation rules have been updated accordingly.
Presentation & attendance policy: This OCT Challenge is part of the SASHIMI Workshop. For accepted papers, at least one author must present the results. Online participation is allowed if: (1) a team consists of members whose passport/citizenship requires a visa to enter France; (2) for participants (e.g., based in the US) who anticipate visa or re-entry problems when returning to their country of residence. In-person participation is mandatory for participating teams that have members currently living and working within the Schengen zone.
1) Download the Dataset: Please download the official dataset from Zenodo.
2) Access the Baseline: Obtain the baseline codes from our GitHub repository and carefully review the provided instructions there. You will also find the standalone OCT Virtual Scanner there.
3) Register Your Team: Please ensure you register your team on this website before June 7! (Upon registration, you will be granted access to our online portal to upload your digital phantoms for preliminary validation). To avoid any technical issues, please also notify us via email regarding your team's registration at drlevmatveev@gmail.com
4) Develop Your Tissue-Mimicking Digital Phantom Generator: Develop a digital phantom generator that outputs a scatterer distribution suitable for the Virtual Scanner (provided within the baseline and via API) and replicates the real input OCT scan as closely as possible after virtual scanning. Please ensure you test it yourself prior to submission. Please also ensure that your code and model are runnable on a PC with 16GB RAM and an RTX 3060 12GB (Windows 10/11). We highly recommend ensuring compatibility with Python 3.12.12 and PyTorch 2.10.0; please verify in advance that your code runs successfully in this specific environment. Additionally, the maximum runtime for generating one phantom with scatterers based on a real B-scan must not exceed 600 seconds per B-scan. These are the maximum hardware and runtime requirements.
5) Preliminary Submission: Submit before June 21 your first set of digital phantoms for preliminary validation.
6) Final Submission: Submit your full paper and code/model by July 1. Please note that full paper submissions and the peer review process will be conducted mandatorily via OpenReview (the submission link will be provided soon). To determine your final score and ranking, your submitted approach will be executed by the organizers on a hidden, hold-out test dataset.
Development & Design Team (HSE University, Nizhny Novgorod):
Alexander Kruglikov
Artem Ishchanov
Roman Stemasov
Maxim Myagkiy
Nataliya Matveeva (Supervisor)