Abstract
Most synthetic optical coherence tomography (OCT) meth ods work in image space. SynthOCT 2026 instead asks for a digital phan tom, represented as a distribution of optical scatterers, that reproduces a target real B-scan after processing by a fixed physics-based Virtual Scan ner. We use train-free coherent inversion performed separately for each scan. A differentiable implementation of the published Fourier-domain OCTmodel fits a complex reflectivity field, decodes it into approximately 2.95 × 105 amplitude-capped sub-scatterers, and refines their depths and energies with scorer-aligned losses. The objective covers the structural image, optical attenuation coefficient (OAC), speckle contrast (SC), re fined speckle contrast (RSC), MS-SSIM, and LPIPS. The submitted pre liminary set achieved an official global score of 0.9947, median MS-SSIM of 0.9971, and median LPIPS of 0.0054, tying for first on the prelim inary leaderboard. Optimisation uses only the differentiable surrogate, while the fixed scanner is reserved for verification and organiser scoring. All recorded Linux/GPU runs met the 600-s limit. A capped GTX 1650 audit took 445.73 s and used 1.733 GiB peak allocated VRAM. Direct Windows/RTX 3060 execution has not been performed.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/SynthOCT_003.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/SynthOCT_003_supp.pdf
Link to Open Review
Open Review Page: https://openreview.net/forum?id=UI2w0Ad58S
BibTex
@InProceedings{ModUra_Coherent_MICCAISAT2026,
author = { Modi, Ura AND Raval, Mehul S.},
title = { { Coherent Differentiable Inversion for High-Fidelity Digital Phantom Generation in Optical Coherence Tomography } },
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}
}