pith. sign in

arxiv: 2604.26548 · v1 · submitted 2026-04-29 · 🧮 math.NA · cs.NA

Projections for handling uncertainties and enabling domain truncation in diffuse optical tomography

Aada Hakula , Pauliina Hirvi , Nuutti Hyv\"onen , Altti J\"a\"askel\"ainen , Ville Kolehmainen This is my paper

Pith reviewed 2026-05-07 11:23 UTC · model grok-4.3

classification 🧮 math.NA cs.NA
keywords diffuse optical tomographyprojection methodmodeling errorsJacobian matrixBayesian inversiondomain truncationnuisance parametersoptical parameters
0
0 comments X

The pith

Projection onto the orthogonal complement of nuisance Jacobians mitigates modeling errors from domain truncation, coupling changes, and tissue parameter misspecification in diffuse optical tomography.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper introduces a projection technique that removes the influence of secondary unknowns from the primary linearized forward model in diffuse optical tomography. It constructs the projection from the range or leading singular vectors of nuisance Jacobians that capture domain truncation, optode coupling variations, and differences in baseline tissue optical properties. The projected system is then solved via standard Bayesian inversion with Gaussian priors. The approach is demonstrated on Monte Carlo simulated data for neonatal brain activity imaging, where it enables the use of truncated computational domains. If the projection fully isolates the primary absorption changes, reconstructions become robust to these common sources of modeling error without requiring exact knowledge of the nuisance parameters.

Core claim

The paper claims that the linearized map from absorption coefficient changes to optode measurements can be projected onto the orthogonal complement of the span of the leading left singular vectors of appropriately weighted nuisance Jacobians (or their difference for parameter misspecification), thereby removing the component of the data that is sensitive to domain truncation, coupling coefficients, and baseline tissue parameters, after which Bayesian reconstruction proceeds on the reduced system.

What carries the argument

The orthogonal projection matrix formed from the range or first left singular vectors of the nuisance Jacobian (weighted by prior information), which annihilates the measurement components attributable to the auxiliary unknowns.

If this is right

  • Smaller computational domains can be used without introducing significant reconstruction artifacts from artificial boundaries.
  • Unknown variations in optode coupling coefficients no longer bias the recovered absorption images.
  • Misspecifications in the baseline optical parameters of tissue types produce reduced errors in the final images.
  • The same Bayesian inversion framework applies directly to the projected equation with unchanged Gaussian prior and noise models.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The technique could extend to other linear or linearized inverse problems in imaging where auxiliary parameters are modeled by separate Jacobians.
  • By allowing truncated domains, it reduces the computational cost of repeated forward solves during reconstruction.
  • If the nuisance Jacobians are precomputed offline, the projection can be applied rapidly in near-real-time settings.

Load-bearing premise

The effects of the nuisance parameters are sufficiently well approximated by a low-rank linear span of the nuisance Jacobian's singular vectors, and the first-order linearization around the background remains valid after projection.

What would settle it

Apply the method to measurements simulated on a full-domain model but reconstructed with a truncated-domain primary Jacobian; if the recovered absorption maps still exhibit large artifacts traceable to the truncation, the projection does not fully remove the nuisance effects.

Figures

Figures reproduced from arXiv: 2604.26548 by Aada Hakula, Altti J\"a\"askel\"ainen, Nuutti Hyv\"onen, Pauliina Hirvi, Ville Kolehmainen.

Figure 1
Figure 1. Figure 1: a) The 15 sources (red crosses) and 21 detectors (blue rings) on the studied view at source ↗
Figure 2
Figure 2. Figure 2: Two left columns: cross-sections of target 1. Two right columns: the corre view at source ↗
Figure 3
Figure 3. Figure 3: Two left columns: cross-sections of target 2. Two right columns: the corre view at source ↗
Figure 4
Figure 4. Figure 4: Case 1 with target 1. Two left columns: reconstruction cross-sections without view at source ↗
Figure 5
Figure 5. Figure 5: Case 1 with target 2. Two left columns: reconstruction cross-sections without view at source ↗
Figure 6
Figure 6. Figure 6: Case 2 with target 1 and the region-of-non-interest (RONI) being the view at source ↗
Figure 7
Figure 7. Figure 7: Case 2 with target 2 and the RONI being the S&S. Two left columns: recon view at source ↗
Figure 8
Figure 8. Figure 8: Case 3 with target 1 (two left columns) and target 2 (two right columns). Top view at source ↗
Figure 9
Figure 9. Figure 9: Case 4 with target 1. Two left columns: reconstruction cross-sections without view at source ↗
Figure 10
Figure 10. Figure 10: Case 4 with target 2. Two left columns: reconstruction cross-sections without view at source ↗
read the original abstract

This paper presents a projection-based technique to mitigate the impact of modeling errors related to domain truncation, changes in the optode coupling coefficients, and misspecified optical parameters of different tissue types in diffuse optical tomography. The approach considers the primary Jacobian matrix of the forward map in the image reconstruction scheme, linking the primary unknown, i.e., the per-voxel absorption coefficient changes in the region of interest, to the optode measurements, as well as the nuisance Jacobians that do the same for the auxiliary unknown parameters of secondary interest. To mitigate mismodeled coupling coefficients or domain truncation, the method projects the linearized forward model defined by the primary Jacobian onto the orthogonal complement of the range of a nuisance Jacobian, or onto the orthogonal complement of the span of a number of first left singular vectors for the nuisance Jacobian that has been weighted to account for prior information on the measurement setup. In the case of a misspecified baseline optical parameter for some tissue type, the nullspace of the utilized orthogonal projection is defined to be the span of first left singular vectors for a (weighted) difference of two Jacobian matrices evaluated at two different levels for the considered tissue-wise optical parameter. The reconstruction is formed by applying Bayesian inversion with Gaussian prior and noise models to the projected linearized equation. We evaluate the method on simulated brain activity data obtained via Monte Carlo simulations of the radiative transfer equation in a voxelized head anatomy for a neonate with combined gestational and chronological age of 41.7 weeks.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The paper presents a projection-based technique to mitigate modeling errors in diffuse optical tomography arising from domain truncation, optode coupling coefficient changes, and misspecified tissue optical parameters. The primary Jacobian linking absorption changes in the region of interest to measurements is projected onto the orthogonal complement of the range (or leading left singular vectors) of a nuisance Jacobian, or the difference of Jacobians for parameter misspecification; Bayesian inversion with Gaussian priors is then applied to the projected linear model. The approach is evaluated via Monte Carlo radiative transfer simulations on a voxelized neonatal head geometry.

Significance. If the subspace projection reliably reduces bias from the modeled nuisances, the method would support computationally cheaper truncated domains and more robust DOT reconstructions under realistic uncertainties. The Monte Carlo validation on a realistic neonate head anatomy provides a concrete test bed and is a positive aspect of the work.

major comments (3)
  1. [§3 (projection definition for domain truncation)] The central mitigation claim for domain truncation rests on the assumption that truncation-induced errors are adequately spanned by the leading singular vectors of the nuisance Jacobian computed on the truncated mesh. However, truncation effects are global and nonlinear, so the first-order linearization and subspace coverage may not hold; the Monte Carlo results should include quantitative comparisons (e.g., reconstruction RMSE or bias with vs. without projection) to verify error reduction.
  2. [§4 (Monte Carlo evaluation)] The evaluation uses a single neonate head geometry and one fixed truncation level. This does not directly test whether the spanned-subspace assumption generalizes across varying truncation severities or anatomies, weakening support for the claim that the method enables reliable domain truncation.
  3. [§3.3 (misspecified optical parameters)] For the tissue-parameter misspecification case, the nullspace is defined via the difference of two Jacobians at different parameter levels. It is unclear how sensitive the retained singular vectors are to the choice of the two levels or the weighting matrix; an ablation on these free parameters would be needed to confirm robustness.
minor comments (2)
  1. [Abstract and §3] Clarify whether the projection is onto the orthogonal complement of the full range or only the leading singular vectors, and provide the explicit form of the weighting matrix used for the nuisance Jacobian.
  2. [§3.4 (Bayesian inversion)] The abstract states that the reconstruction uses 'Bayesian inversion with Gaussian prior and noise models' on the projected equation; the manuscript should specify whether the projection is applied before or after incorporating the noise covariance, as this affects the posterior.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their thorough review and valuable feedback on our manuscript. We address each of the major comments in detail below, providing clarifications and indicating revisions where appropriate to strengthen the presentation of our work.

read point-by-point responses

  1. Referee: [§3 (projection definition for domain truncation)] The central mitigation claim for domain truncation rests on the assumption that truncation-induced errors are adequately spanned by the leading singular vectors of the nuisance Jacobian computed on the truncated mesh. However, truncation effects are global and nonlinear, so the first-order linearization and subspace coverage may not hold; the Monte Carlo results should include quantitative comparisons (e.g., reconstruction RMSE or bias with vs. without projection) to verify error reduction.

    Authors: The use of first-order linearization via the Jacobian is a standard and well-established practice in diffuse optical tomography for handling small perturbations, as the forward model is typically linearized around a baseline. While truncation effects can have global and nonlinear aspects, the nuisance Jacobian is constructed to capture the dominant linear effects of the domain truncation on the measurements. In our Monte Carlo evaluation, we present reconstructions both with and without the projection, demonstrating visually reduced artifacts and improved localization in the region of interest. To provide more rigorous verification as suggested, we have added quantitative metrics including reconstruction RMSE and bias comparisons in the revised manuscript. revision: partial

  2. Referee: [§4 (Monte Carlo evaluation)] The evaluation uses a single neonate head geometry and one fixed truncation level. This does not directly test whether the spanned-subspace assumption generalizes across varying truncation severities or anatomies, weakening support for the claim that the method enables reliable domain truncation.

    Authors: We agree that broader testing across multiple geometries and truncation levels would further support generalization. However, the presented evaluation on a realistic, voxelized neonatal head geometry with a practical truncation level serves as a concrete demonstration in a clinically relevant setting, where domain truncation is particularly beneficial for computational efficiency. The method itself is formulated generally and does not depend on the specific anatomy. We have added a discussion section addressing the applicability to other cases and the potential for the subspace projection to extend to varying truncation severities, while acknowledging that comprehensive multi-geometry studies are an important direction for future work. This does not weaken the support for the claim in the context of the evaluated scenario. revision: partial

  3. Referee: [§3.3 (misspecified optical parameters)] For the tissue-parameter misspecification case, the nullspace is defined via the difference of two Jacobians at different parameter levels. It is unclear how sensitive the retained singular vectors are to the choice of the two levels or the weighting matrix; an ablation on these free parameters would be needed to confirm robustness.

    Authors: The two parameter levels are selected to represent a plausible range of misspecification (e.g., ±20% variation in baseline absorption or scattering for the tissue type), consistent with typical uncertainties in DOT. The weighting matrix is derived from the measurement noise model and prior information on the setup. To address the sensitivity concern, we have performed an additional ablation study in the revised manuscript, varying the levels within the expected uncertainty range and testing different diagonal weightings. The results show that the leading singular vectors are stable, with minimal impact on the retained nullspace and reconstruction accuracy, thereby confirming the robustness of the approach. revision: yes

Circularity Check

0 steps flagged

No circularity: projection defined directly from independent Jacobian computations

full rationale

The derivation constructs the primary and nuisance Jacobians from the linearized forward model of the radiative transfer equation, then defines the projection operator explicitly as the orthogonal complement to the range (or leading left singular vectors) of the nuisance Jacobian or its weighted difference. This is a direct algebraic construction with no reduction to a fitted scalar, no self-referential definition of the target quantity, and no load-bearing self-citation invoked to justify uniqueness or the form of the projection. Bayesian inversion is applied to the resulting projected system using standard Gaussian priors. The Monte Carlo validation on simulated data provides an external check rather than a tautological confirmation. No step reduces by construction to its own inputs.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The method rests on standard assumptions of linearization around a baseline and Gaussian statistics for Bayesian inversion; free parameters are limited to the number of retained singular vectors and the weighting matrix for the nuisance Jacobian.

free parameters (2)
  • number of leading singular vectors retained
    Chosen to account for prior information on the measurement setup when defining the projection nullspace
  • weighting matrix for nuisance Jacobian
    Incorporates prior information on the measurement setup before SVD truncation
axioms (2)
  • domain assumption Linearized forward model via Jacobian is sufficient for reconstruction
    Invoked when the primary and nuisance Jacobians are formed from the forward map
  • domain assumption Gaussian prior and noise models for Bayesian inversion
    Used to form the reconstruction from the projected linearized equation

pith-pipeline@v0.9.0 · 5588 in / 1428 out tokens · 66790 ms · 2026-05-07T11:23:55.478498+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

4 extracted references · 4 canonical work pages

  1. [1]

    https://mcx.space

    [1]Monte Carlo eXtreme (MCX) web page. https://mcx.space. [2]S. R. Arridge,Optical tomography in medical imaging, Inverse problems, 15 (1999), pp. R41– R93, https://doi.org/10.1088/0266-5611/15/2/022. [3]S. R. Arridge, J. P. Kaipio, V. Kolehmainen, M. Schweiger, E. Somersalo, T. Tar- vainen, and M. V auhkonen,Approximation errors and model reduction with ...

    work page doi:10.1088/0266-5611/15/2/022 1999

  2. [2]

    Heiskala, S

    [17]J. Heiskala, S. R. Arridge, V. P. Kolehmainen, T. Tarvainen, and J. P. Kaipio,Approx- imation error method can reduce artifacts due to scalp blood flow in optical brain activa- tion imaging, J. Biomed. Opt., 17 (2012), p. 096012, https://doi.org/10.1117/1.JBO.17.9. 096012. [18]J. Heiskala, P. Hiltunen, and I. Nissil ¨a,Significance of background optic...

    work page doi:10.1117/1.jbo.17.9 2012

  3. [3]

    6629, p 66290T

    Vol. 6629, p 66290T. [20]J. Heiskala, M. Pollari, M. Mets ¨aranta, P. E. Grant, and I. Nissil ¨a,Probabilistic atlas can improve reconstruction from optical imaging of the neonatal brain, Opt. Express, 17 (2009), pp. 14977–14992, https://doi.org/10.1364/OE.17.014977. [21]E. M. C. Hillman, A. Devor, M. B. Bouchard, A. K. Dunn, G. W. Krauss, J. Skoch, B. J....

    work page doi:10.1364/oe.17.014977 2009

  4. [4]

    Kangasniemi, M

    [28]J. Kangasniemi, M. Mozumder, A. Pulkkinen, and T. Tarvainen,Stochastic Gauss-Newton method for estimating absorption and scattering in optical tomography with the Monte Carlo method for light transport, Biomed. Opt. Express, 15 (2024), pp. 4925–4942, https: //doi.org/10.1364/BOE.528666. [29]V. Kolehmainen, M. Schweiger, I. Nissil ¨a, T. Tarvainen, S. ...

    work page doi:10.1364/boe.528666 2024