aiqbench analysis
This study exposes a critical flaw in current OCT-to-OCTA synthesis models, demonstrating that high reconstruction metrics like PSNR and SSIM do not guarantee clinical utility for downstream diagnostic tasks. The research reveals that fine structural details essential for grading diabetic retinopathy are often fabricated rather than reconstructed, causing significant performance drops when these synthetic images are used by frozen segmentation probes. Consequently, the field must shift evaluation standards from visual fidelity to task-specific accuracy before deploying such tools in medical workflows.
In Defense of OCTA: The Reconstruction-Utility Gap in OCT-to-OCTA Synthesis
arXiv:2608.15626v1 Announce Type: new Abstract: Optical coherence tomography angiography (OCTA) images retinal blood flow, giving capillary-perfusion and foveal-avascular-zone biomarkers that grade diabetic-retinopathy ischemia. Because OCTA hardware is less common than structural OCT, recent work synthesizes it from OCT, reporting strong reconstruction (3D PSNR > 31 dB, SSIM > 0.9). We ask not whether the synthetic image looks similar, but whether it supports the measurements OCTA is acquired for. A frozen real-OCTA segmenter, applied as a probe to two synthesizers (XOCT, TransPro), shows downstream Dice falling with structural fineness: large vessels survive (0.862 -> 0.831) while the fine capillary network collapses (0.798 -> 0.635, five times the large-vessel loss; paired Wilcoxon p < 1e-3), TransPro worse throughout. A matched-blur control shows this detail is fabricated, not blurred. Retrained on a private Spectralis dataset, neither synthesizer reproduces the neovascular lesion (qualitative, n=3). Reconstruction fidelity is not clinical utility; we establish downstream-task fidelity as the evaluation OCT-to-OCTA synthesis needs.
Abstract:Optical coherence tomography angiography (OCTA) images retinal blood flow, giving capillary-perfusion and foveal-avascular-zone biomarkers that grade diabetic-retinopathy ischemia. Because OCTA hardware is less common than structural OCT, recent work synthesizes it from OCT, reporting strong reconstruction (3D PSNR > 31 dB, SSIM > 0.9). We ask not whether the synthetic image looks similar, but whether it supports the measurements OCTA is acquired for. A frozen real-OCTA segmenter, applied as a probe to two synthesizers (XOCT, TransPro), shows downstream Dice falling with structural fineness: large vessels survive (0.862 -> 0.831) while the fine capillary network collapses (0.798 -> 0.635, five times the large-vessel loss; paired Wilcoxon p < 1e-3), TransPro worse throughout. A matched-blur control shows this detail is fabricated, not blurred. Retrained on a private Spectralis dataset, neither synthesizer reproduces the neovascular lesion (qualitative, n=3). Reconstruction fidelity is not clinical utility; we establish downstream-task fidelity as the evaluation OCT-to-OCTA synthesis needs.
10 pages, 2 figures, 3 tables. Accepted at OMIA 2026 (13th Ophthalmic Medical Image Analysis Workshop, MICCAI 2026). This is the pre-peer-review submitted version; the camera-ready revises the characterization of the capillary failure
Machine Learning (cs.LG)
arXiv:2608.15626 [cs.LG]
(or arXiv:2608.15626v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2608.15626 arXiv-issued DOI via DataCite (pending registration)
Full text under the publisher's open licence · 202 words
