AI summary · arXiv: 2611.12686v1 · cs.CV

Differentiable Inverse Rendering Recovers Intrinsic Properties from Intravascular Optical Coherence Tomography

Ajay Manicka, Karim Kadry, Sebastian Ruiz-Lopera, Maryam Alsharqi, Sanjana Arji, Samuel Ruiperez-Campillo, Nestor Uribe-Patarroyo, Brett E. Bouma,
and 2 more authorsFarhad Nezami, Elazer R. Edelman
Original paper Submitted to arXiv on 09 Nov. 2026 27 Pages, 6 Figures, Under Review License: CC BY 4.0 Also in: fr

How to read this summary? Illustrated guide

The essentials

The paper presents a differentiable physics-based method that recovers per-pixel optical attenuation maps from OCT images, validated on phantoms and applied to clinical IV-OCT.

Question

Can intrinsic tissue attenuation be recovered directly from OCT images by inverting a differentiable model of image formation? The authors examine whether this approach works without training data or prior segmentation and transfers to clinical images.

Method

The authors optimized per-pixel attenuation by gradient descent through a differentiable, single-scattering Beer-Lambert OCT forward model, using Adam and image-reconstruction loss. They tested synthetic data, processed 5,120 B-scans from five liquid intralipid phantoms at two focal positions, and applied the approach to 29,269 clinical frames from 128 patients.

Findings

  • Result On 500 synthetic B-scans, optimization recovered the known attenuation maps with 0.15% final MAPE and concordance R² of 0.9997.
    See the passage in the paper
    Across all 500 phantoms, the recovered tissue $\mu$ achieved final MAPE $=0.15\%$, concordance $R^{2}=0.9997$, MAE $=0.007\pm 0.001$ mm-1, and regression slope $=0.998$.
  • Result Recovered attenuation increased linearly with intralipid concentration across the phantom acquisitions, with R² values of 0.990 and 0.993 at the two focal positions.
    See the passage in the paper
    H and L yielded slopes of $0.663$ and $0.629$ mm-1% with $R^{2}=0.990$ and $R^{2}=0.993$, comparable to the CF reference (slopes of $0.623$ and $0.625$ mm-1%).
  • Result For the tested phantom initialization distributions, tissue-mean attenuation was reproducible across random seeds, although some initialization ranges produced different recovered values.
    See the passage in the paper
    Intra-acquisition coefficients of variation (CoV) ranged from $0.18\%$ to $0.32\%$ across the 10 acquisitions (mean $=0.27\pm 0.04\%$; Figure 4c), indicating that, given a fixed initialization distribution, the optimizer yields reproducible tissue-mean attenuation.
  • Authors’ interpretation The method produced spatially resolved attenuation maps on clinical IV-OCT frames, but the study did not establish the absolute accuracy of clinical attenuation values.
    See the passage in the paper
    No pixel-level ground-truth attenuation exists for in vivo coronary tissue, so this experiment demonstrates that the method transfers unmodified to clinical acquisition geometry rather than validating recovered attenuation values.
  • Result Optimizing backscattering separately from attenuation improved clinical image-rendering PSNR without changing the recovered attenuation structure.
    See the passage in the paper
    This reduces residuals across the cohort and raises the mean PSNR to $22.63\pm 0.74$ dB (Dice $=0.954\pm 0.009$; Figure 5, bottom row) without altering the recovered attenuation structure.

Limitations

  • The clinical data lacked scanner-specific calibration and post-processing metadata, so recovered values represent relative rather than absolute attenuation.
    See the passage in the paper
    Because scanner-specific calibration metadata was unavailable for this acquisition, we adopted measured confocal parameters reported for the C7-XR / Dragonfly catheter family ($z_{R}=0.59$ mm, $z_{f}=1.06$ mm) Gargesha et al. (2015), together with an axial pixel size of $0.01932$ mm/pixel estimated from the ILUMIEN OPTIS A-scan range. Combined with the undocumented post-processing transformations applied to the log-compressed clinical images, the recovered attenuation values capture relative tissue contrast rather than absolute quantities.
  • Clinical per-pixel attenuation could not be validated against ground truth because no in vivo reference standard was available.
    See the passage in the paper
    No pixel-level ground-truth attenuation exists for in vivo coronary tissue, so this experiment demonstrates that the method transfers unmodified to clinical acquisition geometry rather than validating recovered attenuation values.
  • The authors describe the forward model as simplified and note that its single-scattering assumption may not suffice when multiple scattering dominates.
    See the passage in the paper
    Lastly, the single-scattering Beer-Lambert forward model may become insufficient at higher optical depths and in cases where multiple scattering dominates, although preliminary comparisons with the Extended Huygens-Fresnel model ($g=0.7$, a value previously used for tissue in OCT simulations Hartinger et al. (2014)) yielded nearly identical recovered attenuation at our phantom concentrations (mean difference $<1\%$; Supplementary Note 5).

Why it matters

The authors argue that differentiable inverse rendering offers a modular, physics-based route to quantitative tissue characterization from routine OCT, without training data or segmentation. They propose that per-pixel attenuation maps could support plaque characterization and retrospective analysis, while noting that absolute calibration is needed for population-level comparisons.

Written by AI from the paper. Passages are checked word for word against the text we analysed; the rest may contain errors.

AI-generated keywords: Intravascular OCT optical attenuation coefficient differentiable inverse rendering Beer–Lambert model quantitative imaging

Read further

  • Comprehensive Summary
  • Layman's Summary
  • Blog article
  • Original abstract
The paper investigates whether intrinsic tissue properties can be recovered directly from medical images by modeling and inverting the physics that produced them, rather than applying post-processing corrections or training a model on examples. The authors study this question using intravascular optical coherence tomography (IV-OCT) and the optical attenuation coefficient (OAC), which measures how quickly light diminishes as it travels through tissue. Their goal is to recover OAC at every pixel in a single OCT cross-section without training data or prior segmentation. The method uses a differentiable Beer–Lambert forward model to predict OCT intensity from a candidate attenuation map and specified imaging parameters. It accounts for depth-dependent confocal effects, system gain, and a noise floor. Automatic differentiation calculates gradients of the image-reconstruction error with respect to pixel-level attenuation values, and the Adam optimizer updates those values to bring the simulated image closer to the measured one. In some phantom experiments, total-variation regularization encourages spatial smoothness. For the clinical application, the authors first estimate attenuation while holding backscattering fixed. They then freeze the attenuation map and optimize a separate per-pixel backscattering parameter to improve image rendering. Because the forward model is replaceable, the optimization procedure could potentially be adapted to other physical models or optical properties. The authors first evaluated numerical correctness using synthetic data. Across 50,000 pixels, the differentiable forward model closely matched an analytical Beer–Lambert calculation, with an R-squared of 1.0000, a mean absolute error of 5.0 × 10^-6 dB, and a maximum error of 6.1 × 10^-5 dB. In inverse tests on 500 synthetic B-scans, the recovered attenuation maps approached the known maps, reaching a mean absolute percentage error of 0.15% and an R-squared of 0.9997 after 30,000 iterations. A comparison of optimizers showed that gradient-based Adam and L-BFGS-B retained high accuracy as the number of coupled parameters increased to 5,000, whereas gradient-free methods deteriorated substantially. At the largest problem size, the reported errors for Random Search and CMA-ES were 30.32% and 100.91%, respectively. Physical-phantom experiments involved 5,120 B-scan inversions across five liquid intralipid concentrations, each measured at two focal positions. Recovered attenuation increased linearly with concentration, with R-squared values of 0.990 and 0.993 for the two positions. The concentration-response slopes were 0.663 and 0.629 mm^-1 per percentage point, close to curve-fitting reference slopes of 0.623 and 0.625. However, recovered values were 0.083–0.176 mm^-1 lower than the reference values. Across repeated runs with different random initializations, tissue-mean attenuation had a coefficient of variation of 0.18–0.32%. Recovery also tracked declining signal-to-noise ratio: it changed relatively little over the first 4–5 dB of added noise, then worsened more sharply. The method was applied to 29,269 clinical frames from 128 patients across four trial cohorts. The attenuation maps followed visible vessel structures. However, the clinical acquisition lacked scanner-specific calibration and processing metadata, and no in-vivo pixel-level reference standard was available. The clinical results therefore demonstrate transfer and recovery of relative contrast, not the accuracy of absolute attenuation values. Adding the backscattering parameter increased mean rendering PSNR from 14.83 to 22.63 dB, while tissue-mask Dice overlap changed from 0.957 to 0.954.

Was this synthesis useful?

Score: 0

Why do we need votes?

Votes are used to determine whether we need to re-run our summarizing tools. If the count reaches -10, our tools can be restarted.

Figures and plots

Ask about a figure by number or use its question button to request visual analysis. Up to two figures and four image panels per question. Values read from plots are approximate; check the original data.

AI assistant instructions?

Tools for this paper

Explain it differently 1 free rewrite left today
Save to a collection Group papers in collections to compare them, chat with them and export them. Continue with Google Create a free account Log in

Advanced tools Premium

For researchers who follow a field closely: citation alerts, journal-club slides, version comparison and Zotero export. Discover Premium

Zotero Zotero guide
Citation alerts Citation alerts guide
Who works on it?

Among the 23 papers around this one, the name that comes back most: Ya-Ning Wang (2 papers)

1 more person signs several of them. The full map adds their institutions and approaches, from the papers of the last years.

Continue with Google Create a free account: your first map is free
Weekly letter Presented in "Artificial intelligence this week" of 16 November 2026. Artificial intelligence this week: What impressive scores leave out · Receive one by email for free

Papers around this one

See them on a map

Ranked by how close their title and abstract are to this paper’s. Select some to read them together.

Closest papers

  1. Very close Another approach to the problem Cited by this paper
    Depth-dependent attenuation and backscattering characterization of optical coherence tomography by stationary iterative method
    Ya-Ning Wang, Shu-Wen Wei, J. U. Kang · 2023 · Journal of Biomedical Optics · cited 8 times · Not summarized yet

    The stationary iterative method jointly estimates depth-wise attenuation and backscattering, whereas this paper recovers per-pixel attenuation through differentiable forward-model inversion.

    Abstract

    Abstract. Significance Extracting optical properties of tissue [e.g., the attenuation coefficient (μ) and the backscattering fraction] from the optical coherence tomography (OCT) images is a valuable tool for parametric imaging and related diagnostic applications. Previous attenuation estimation models depend on the assumption of the uniformity of the backscattering fraction (R) within layers or whole samples, which does not accurately represent real-world conditions. Aim Our aim is to develop a robust and accurate model that calculates depth-wise values of attenuation and backscattering fractions simultaneously from OCT signals. Furthermore, we aim to develop an attenuation compensation mod…

  2. Very close Another approach to the problem Cited by this paper
    Tissue characterization with depth-resolved attenuation coefficient and backscatter term in intravascular optical coherence tomography images
    Shengnan Liu, Y. Sotomi, J. Eggermont et al. · 2017 · Journal of Biomedical Optics · cited 57 times · Not summarized yet

    The depth-resolved model estimates attenuation and a backscatter-related term per pixel in IVOCT, while this paper optimizes attenuation through a differentiable model.

    Abstract

    Abstract. An important application of intravascular optical coherence tomography (IVOCT) for atherosclerotic tissue analysis is using it to estimate attenuation and backscatter coefficients. This work aims at exploring the potential of the attenuation coefficient, a proposed backscatter term, and image intensities in distinguishing different atherosclerotic tissue types with a robust implementation of depth-resolved (DR) approach. Therefore, the DR model is introduced to estimate the attenuation coefficient and further extended to estimate the backscatter-related term in IVOCT images, such that values can be estimated per pixel without predefining any delineation for the estimation. In order…

  3. Very close Another approach to the problem Cited by this paper
    End-to-end reconstruction of OCT optical properties and speckle-reduced structural intensity via physics-based learning
    Jing-Lun Yu, Ya-Ning Wang, Wenhan Guo et al. · 2026 · BiOS · cited 2 times · Not summarized yet

    The end-to-end network reconstructs optical-property maps and speckle-reduced OCT intensity from simulated training data, unlike this paper’s training-free gradient-based inversion.

    Abstract

    Inverse scattering in optical coherence tomography (OCT) seeks to recover both structural images and intrinsic tissue optical properties, including refractive index, scattering coefficient, and anisotropy. This inverse problem is challenging due to attenuation, speckle noise, and strong coupling among parameters. We propose a regularized end-to-end deep learning framework that jointly reconstructs optical parameter maps and speckle-reduced OCT structural intensity for layer visualization. Trained with Monte Carlo–simulated ground truth, our network incorporates a physics-based OCT forward model that generates predicted signals from the estimated parameters, providing physics-consistent super…

  4. Very close Broader context
    Improving Automated Tissue Characterization in Optical Coherence Tomography by Melding Attenuation Compensation with Deep Learning
    Ya-Nan Niu, Max L. Olender, David Marlevi et al. · 2021 · 2021 IEEE EMBS International Conference on Biomedical and Health Informatics (BHI) · cited 2 times · Not summarized yet

    The deep-learning classification method combines attenuation-compensated OCT images with learned tissue characterization, rather than recovering attenuation as its primary output.

    Abstract

    The added benefit of intravascular imaging, such as optical coherence tomography (OCT), is that lumenography is augmented by lesion micro-morphology. For intravascular image analysis, deep learning methods have achieved satisfactory performance in automatic characterization of micro-morphology and tissue distribution, but have still been restricted by inherent OCT signal attenuation. Leveraging domain knowledge, we demonstrate the benefits of melding deep learning with existing attenuation compensation approaches to achieve more accurate full atherosclerotic tissue classification in OCT images. By combining acquired OCT images with attenuation-compensated and contrast-enhanced equivalents, o…

  5. Close Another approach to the problem
    QOCT-Net: A Physics-Informed Neural Network for Intravascular Optical Coherence Tomography Attenuation Imaging
    Zheng Sun, Shuyan Wang, Ying-Sa Hou et al. · 2023 · IEEE journal of biomedical and health informatics · cited 4 times · Not summarized yet

    QOCT-Net uses a trained physics-informed network and a multiple-scattering model to infer IVOCT attenuation, unlike this paper’s direct optimization through a single-scattering model.

    Abstract

    Intravascular optical coherence tomography (IVOCT) provides high-resolution, depth-resolved images of coronary arterial microstructure by acquiring backscattered light. Quantitative attenuation imaging is important for accurate characterization of tissue components and identification of vulnerable plaques. In this work, we proposed a deep learning method for IVOCT attenuation imaging based on the multiple scattering model of light transport. A physics-informed deep network named Quantitative OCT Network (QOCT-Net) was designed to recover pixel-level optical attenuation coefficients directly from standard IVOCT B-scan images. The network was trained and tested on simulation and in vivo datase…

  6. Close Same method, another problem
    Differentiable Imaging Meets Adaptive Neural Dropout: An Advancing Method for Transparent Object Tomography
    Delong Yang, Shaohui Zhang, Jia-Song Sun et al. · 2025 · Not summarized yet

    The ADNN uses differentiable, physics-constrained imaging to reconstruct refractive-index distributions in optical diffraction tomography rather than attenuation in OCT.

    Abstract

    Label-free tomographic microscopy offers a compelling means to visualize three-dimensional (3D) refractive index (RI) distributions from two-dimensional (2D) intensity measurements. However, limited forward-model accuracy and the ill-posed nature of the inverse problem hamper artifact-free reconstructions. Meanwhile, artificial neural networks excel at modeling nonlinearities. Here, we employ a Differentiable Imaging framework that represents the 3D sample as a multi-layer neural network embedding physical constraints of light propagation. Building on this formulation, we propose a physics-guided Adaptive Dropout Neural Network (ADNN) for optical diffraction tomography (ODT), focusing on net…

  7. Close Groundwork for this paper Cited by this paper
    Depth-resolved model-based reconstruction of attenuation coefficients in optical coherence tomography.
    Koen A. Vermeer, J. Mo, J. Weda et al. · 2013 · Biomedical Optics Express · cited 353 times · Not summarized yet

    The single-scattering model reconstructs pixel-level OCT attenuation from depth profiles and phantom measurements, providing an earlier model-based approach to the property recovered here.

    Abstract

    We present a method, based on a single scattering model, to calculate the attenuation coefficient of each pixel in optical coherence tomography (OCT) depth profiles. Numerical simulations were used to determine the model's response to different depths and attenuation coefficients. Experiments were performed on uniform and layered phantoms with varying attenuation coefficients. They were measured by a 1300 nm OCT system and their attenuation coefficients were evaluated by our proposed method and by fitting the OCT slope as the gold standard. Both methods showed largely consistent results for the uniform phantoms. On the layered phantom, only our proposed method accurately estimated the attenu…

  8. Close Same question Cited by this paper
    Parameter estimation of atherosclerotic tissue optical properties from three-dimensional intravascular optical coherence tomography
    M. Gargesha, Ronny Shalev, D. Prabhu et al. · 2015 · Journal of Medical Imaging · cited 26 times · Not summarized yet

    The study estimates attenuation and backscattered intensity of atherosclerotic tissue from clinical IVOCT pullbacks, addressing quantitative tissue optical-property recovery.

    Abstract

    Abstract. We developed robust, three-dimensional methods, as opposed to traditional A-line analysis, for estimating the optical properties of calcified, fibrotic, and lipid atherosclerotic plaques from in vivo coronary artery intravascular optical coherence tomography clinical pullbacks. We estimated attenuation μt and backscattered intensity I0 from small volumes of interest annotated by experts in 35 pullbacks. Some results were as follows: noise reduction filtering was desirable, parallel line (PL) methods outperformed individual line methods, root mean square error was the best goodness-of-fit, and α-trimmed PL (α-T-PL) was the best overall method. Estimates of μt were calcified (3.84±0.…

Prior work it builds on

  1. Close Broader context
    Noise and bias in optical coherence tomography intensity signal decorrelation
    N. Uribe-Patarroyo, A. Post, Sebastián Ruiz-Lopera et al. · 2020 · OSA Continuum · cited 24 times · Not summarized yet

    The study analyzes noise and bias in OCT intensity decorrelation, a different OCT measurement problem from recovering tissue attenuation.

    Abstract

    Functional optical coherence tomography (OCT) imaging based on the decorrelation of the intensity signal has been used extensively in angiography and is finding use in flowmetry and therapy monitoring. In this work, we present a rigorous analysis of the autocorrelation function, introduce the concepts of contrast bias, statistical bias and variability, and identify the optimal definition of the second-order autocorrelation function (ACF) g(2) to improve its estimation from limited data. We benchmark different averaging strategies in reducing statistical bias and variability. We also developed an analytical correction for the noise contributions to the decorrelation of the ACF in OCT that ext…

  2. Close Review
    Review of methods and applications of attenuation coefficient measurements with optical coherence tomography
    Shuang Chang, Audrey K. Bowden · 2019 · Journal of Biomedical Optics · cited 79 times · Not summarized yet

    The review surveys physical models, estimation methods, and clinical applications for extracting OCT attenuation coefficients.

    Abstract

    Abstract. The optical attenuation coefficient (AC), an important tissue parameter that measures how quickly incident light is attenuated when passing through a medium, has been shown to enable quantitative analysis of tissue properties from optical coherence tomography (OCT) signals. Successful extraction of this parameter would facilitate tissue differentiation and enhance the diagnostic value of OCT. In this review, we discuss the physical and mathematical basis of AC extraction from OCT data, including current approaches used in modeling light scattering in tissue and in AC estimation. We also report on demonstrated clinical applications of the AC, such as for atherosclerotic tissue chara…

  3. Close Broader context
    Consensus standards for acquisition, measurement, and reporting of intravascular optical coherence tomography studies: a report from the International Working Group for Intravascular Optical Coherence Tomography Standardization and Validation.
    G. Tearney, E. Regar, T. Akasaka et al. · 2012 · Journal of the American College of Cardiology · cited 1,896 times · Not summarized yet

    The consensus report standardizes IVOCT acquisition, measurement, and reporting, providing clinical measurement context rather than an attenuation-recovery method.

    Abstract

    OBJECTIVES The purpose of this document is to make the output of the International Working Group for Intravascular Optical Coherence Tomography (IWG-IVOCT) Standardization and Validation available to medical and scientific communities, through a peer-reviewed publication, in the interest of improving the diagnosis and treatment of patients with atherosclerosis, including coronary artery disease. BACKGROUND Intravascular optical coherence tomography (IVOCT) is a catheter-based modality that acquires images at a resolution of ~10 μm, enabling visualization of blood vessel wall microstructure in vivo at an unprecedented level of detail. IVOCT devices are now commercially available worldwide, th…

  4. Close* Broader context
    Differentiable Imaging: Progress, Challenges, and Outlook
    Ni Chen, David J. Brady, Edmund Y. Lam · 2025 · Advanced Devices &amp; Instrumentation · cited 21 times · Not summarized yet

    The overview discusses progress and challenges in differentiable imaging, the broader framework used for this paper’s OCT inversion.

  5. Close Same method, another problem
    Intraoperative 2D/3D Image Registration via Differentiable X-ray Rendering
    Vivek Gopalakrishnan, Neel Dey, Polina Golland · 2023 · Computer Vision and Pattern Recognition · cited 58 times · Not summarized yet

    DiffPose uses differentiable physics-based X-ray rendering for self-supervised 2D/3D registration, applying differentiable inverse rendering to a different imaging task.

    Abstract

    Surgical decisions are informed by aligning rapid portable 2D intraoperative images (e.g. X-rays) to a high-fidelity 3D preoperative reference scan (e.g. CT). However, 2D/3D registration can often fail in practice: conventional optimization methods are prohibitively slow and suscepti-ble to local minima, while neural networks trained on small datasets fail on new patients or require impractical land-mark supervision. We present DiffPose, a self-supervised approach that leverages patient-specific simulation and differentiable physics-based rendering to achieve accurate 2D/3D registration without relying on manually labeled data. Preoperatively, a CNN is trained to regress the pose of a random…

How are these papers found?

We compare the title and abstract of this paper with those of every paper summarized on this site, and with the papers Semantic Scholar recommends, those it cites and those citing it (for a much-cited paper, its most recent citations and, from OpenAlex, the most cited ones). Among the prior work and the follow-ups, the papers that use each other in depth, a method it uses and the much-cited follow-ups come a little higher. The levels (very close, close, related) come from the similarity of the texts, not from a judgment of quality; * marks a paper without an abstract, compared by its title. The sentence under a paper is written by an AI from the two abstracts: check it in the papers. A paper marked not relevant by three readers leaves the list.

Updated on 16 November 2026.

The illustrated guide to the papers around a paper

Disclaimer: The AI-based summarization tool and virtual assistant provided on this website may not always provide accurate and complete summaries or responses. We encourage you to carefully review and evaluate the generated content to ensure its quality and relevance to your needs.