Abstract
This challenge was motivated by the need to explore various methodologies for generating spatial distributions of optical scatterers that accurately mimic the microstructure of real biological tissue based on in vivo Optical Coherence Tomography (OCT) scans. We emphasize that the objective of this challenge was not to directly synthesize an OCT image, but rather to reconstruct the underlying physical microstructure. When this microstructure is processed by a physics-based Virtual Scanner, which faithfully replicates the propagation and scattering of light, the resulting synthetic scan must converge with the real target scan across a set of similarity metrics. A key feature of the SynthOCT challenge is our reliance not only on the structural similarity of the images but also on physics-based parametric maps: Optical Attenuation Coefficient (OAC) maps and Speckle Contrast (SC) maps (both classical and refined). During the preparation phase, we evaluated a wide range of pairwise similarity metrics and selected Multi-Scale Structural Similarity (MS-SSIM) and Learned Perceptual Image Patch Similarity (LPIPS) to evaluate all four scan representations (Struct, OAC, SC and RSC). MS-SSIM serves as a multi-scale structural fidelity metric, while LPIPS acts as a perceptual similarity metric. The challenge utilized an open-access dataset of in vivo skin OCT scans. Crucially, a portion of this dataset was strictly held out and hidden by the organizers for the final testing phase to prevent any data leakage into the participants’ algorithms. The submission and evaluation process consisted of two stages. In the first stage, participants uploaded their generated digital phantoms (scatterer distributions) for the training and validation sets, which were automatically verified and quantitatively assessed. In the second stage, the top-performing teams submitted their complete source codes. The benchmarking team then executed these codes on the held-out test set to generate the test set phantoms and compute the metrics for the final leaderboard. The challenge proposal underwent a rigorous peer-review process governed by the MICCAI Challenge Committee. We note that successfully reconstructing the governing scatterer distribution is a significant step toward the realization of non-invasive, OCT-based virtual histology. We acknowledge that creating biologically consistent virtual histology will ultimately require defining specific constraints and heuristics regarding the properties, combinations, and distributions of scatterers. However, within the scope of this challenge, the inverse problem was successfully solved: participants inverted the OCT scanning process to reconstruct governing scatterer distributions that, upon forward scanning, yielded synthetic scans practically indistinguishable from the original targets. Initially, 12 teams registered for the challenge, and six teams submitted their final solutions. During the first stage, a clear performance gap emerged: the top-3 teams achieved median performance metrics exceeding 98\%, while the remaining teams plateaued at 70\% or lower. The results were made public, and the top-3 teams submitted their codes for the second phase of testing alongside full papers detailing their algorithms. These papers were peer-reviewed via the OpenReview platform, with reviews made publicly available to enhance the reproducibility of the presented approaches. This opening article serves as an introduction to the challenge, describing the task, the physics-based metrics, the Virtual Scanner, and briefly discussing the participants’ results. Detailed descriptions of each of the top-3 approaches are presented in their respective papers within this volume.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/SynthOCT_000.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: Not Available
BibTex
@InProceedings{MatLev_SynthOCT_MICCAISAT2026,
author = { Matveev, Lev AND Remedios, Samuel W. AND Zam, Azhar AND Hassan, Taimur AND Mikhailenko, Daniil AND Nikoshin, Denis AND Ryabkov, Maxim AND Petrova, Ksenia AND Kulagina, Maria AND Sovetsky, Alexander AND Kulagin, Pavel AND Matveyev, Alexander AND Gelikonov, Grigory AND Zaitsev, Vladimir},
title = { { SynthOCT Challenge: Introduction and Brief Overview } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17258},
month = {pending},
page = {pending}
}