pith. sign in

arxiv: 2605.22503 · v1 · pith:ZWMX3CPUnew · submitted 2026-05-21 · ⚛️ physics.optics

Scattering correction for infrared spectra of biological cells using computational infrared microspectroscopy and deep learning

Sergio G. Rodrigo , Ilia L. Rasskazov , Luis Martin-Moreno , Martin Schnell This is my paper

Pith reviewed 2026-05-22 03:45 UTC · model grok-4.3

classification ⚛️ physics.optics
keywords infrared microspectroscopyscattering correctiondeep learningFDTD simulationbiological cellsellipsoidal modelsabsorption spectraHeLa cells
0
0 comments X

The pith

A deep learning model trained on FDTD simulations of ellipsoidal cells recovers true absorption spectra and 3D dimensions from scattered IR measurements.

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

The paper develops a scattering correction method for infrared microspectroscopy of single biological cells by combining finite-difference time-domain simulations with deep learning. It generates a library of synthetic spectra from three-dimensional ellipsoid models that represent both spherical suspended cells and flattened adhered cells. A neural network then inverts the scattering-distorted absorbance spectra to extract the underlying absorption spectra that reflect chemical content. The same model also retrieves the cell's three-dimensional size parameters directly from the measured data. This matters because scattering has long prevented reliable chemical analysis of individual cells, and removing that barrier would let infrared spectroscopy reveal molecular details without spherical approximations that fail for real samples.

Core claim

The authors demonstrate that a deep learning network trained on a synthetic library of IR spectra, generated via the finite-difference time-domain method applied to three-dimensional ellipsoidal cell phantoms, can invert measured absorbance spectra to recover the true absorption spectra of cervical cancer cells and simultaneously determine the cells' three-dimensional dimensions.

What carries the argument

Finite-difference time-domain computational microspectroscopy that generates realistic training spectra from 3D ellipsoid cell models, paired with a deep learning model for fast spectral inversion.

If this is right

  • True chemical absorption spectra become extractable from IR absorbance data for both spherical and flattened cells without relying on Mie theory.
  • Cell geometry information that was previously lost to scattering can now be recovered as an additional output of the measurement.
  • The FDTD approach can be extended to generate training data for cells with more complex shapes once the deep learning model is retrained.
  • Scattering effects, normally treated as artifacts, become a usable source of morphological information.

Where Pith is reading between the lines

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

  • The framework could support chemical analysis of cells in their native adhered or suspended states rather than requiring spherical approximations.
  • If the model generalizes well, it might reduce the need for physical sample preparation steps that alter cell chemistry.
  • Similar simulation-plus-inversion pipelines could address scattering in other label-free spectroscopic methods used on biological samples.

Load-bearing premise

Real biological cells behave sufficiently like three-dimensional ellipsoids in the simulated training data so that the resulting deep learning model will work on actual experimental infrared spectra.

What would settle it

Applying the trained model to laboratory IR spectra of HeLa cells whose true absorption spectra and 3D shapes have been independently measured by another technique such as Raman spectroscopy or atomic force microscopy would confirm or refute the recovery of correct spectra and dimensions.

Figures

Figures reproduced from arXiv: 2605.22503 by Ilia L. Rasskazov, Luis Martin-Moreno, Martin Schnell, Sergio G. Rodrigo.

Figure 3
Figure 3. Figure 3: (a) Deep-learning model that accepts the IR absorbance spectrum, 𝐴(𝜈), from FDTD as input and predicts the true absorption spectrum, 𝒜true ED (𝜈), as output. (b) Subset of the 3D ellipsoid training data, showing the Absorbance spectra, 𝐴(𝜈), from FDTD for different volumetric aspect ratios, 𝑟𝑥 𝑅 = 𝑟𝑦 𝑅 = 0.8. .2.1, for a constant ellipsoid volume, 𝑉 = 4 3 𝜋(5 μm) 3 . (c) The corresponding true absorption s… view at source ↗
Figure 4
Figure 4. Figure 4: (a) Deep-learning model that accepts the model parameter, 𝐩 = (𝑟𝑥 , 𝑟𝑦, 𝑟𝑧 , 𝑘), as input and predicts either the absorbance spectrum, 𝐴 𝐩 (𝜈), or the true absorption spectrum, 𝒜true 𝐩 (𝜈), as output. (b) Extraction of ellipsoid size by inversion of DL model in (a) using the absorbance spectra, 𝐴(𝜈), of cell nuclei from [PITH_FULL_IMAGE:figures/full_fig_p014_4.png] view at source ↗
read the original abstract

IR microspectroscopy of single biological cells is challenged by strong light scattering, which produces baseline effects and peak distortions in the IR spectra and hinders the direct extraction of chemical information. Current methods for scattering correction typically rely on Mie theory and are accurate only under the assumption that the cell can be approximated by a sphere. Here, we present a framework for the scattering correction of IR absorbance spectra that is based on 3D ellipsoid models and provides efficient scattering correction for both suspended (spherical) and adhered (flattened) cells. Our approach combines deep learning approaches with computational IR microspectroscopy based on the finite-difference time-domain (FDTD) method. The FDTD method generates a synthetic library of realistic training spectra, while the deep learning model enables fast spectral inversion. We demonstrate scattering correction in silico using numerical cell phantoms of cervical cancer cells (HeLa) and show that the true absorption spectra can be inferred from IR absorbance spectra. We further show that the 3D cell dimensions can be recovered from the IR absorbance spectra, highlighting that the inherent light scattering could be exploited to realize the full analytical potential of IR spectroscopy. We anticipate that deep learning-based scattering corrections can be readily extended to increasingly complex sample geometries owing to the flexibility of the FDTD method to model arbitrary geometries.

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

1 major / 2 minor

Summary. The manuscript presents a framework that combines finite-difference time-domain (FDTD) simulations of 3D ellipsoidal HeLa cell phantoms with deep learning to correct scattering distortions in infrared microspectroscopy. Synthetic absorbance spectra are generated via FDTD, a neural network is trained to invert these to true absorption spectra, and the same model is shown to recover the 3D cell dimensions; all demonstrations are performed in silico on numerical phantoms.

Significance. If the inversion generalizes, the method would extend scattering correction beyond the spherical Mie-theory limit to flattened or adhered cells and would additionally turn scattering into a source of morphological information. The use of FDTD to generate training libraries for arbitrary geometries is a clear technical strength and supports the claim of future extensibility. The in-silico recovery results are internally consistent under the stated modeling assumptions.

major comments (1)
  1. [Abstract/Results] Abstract and Results sections: the in-silico test phantoms are generated with the identical FDTD ellipsoid forward model (uniform refractive index, no organelles, perfect ellipsoidal shape) used to create the training library. Consequently the reported recovery of absorption spectra and cell dimensions demonstrates only that the network has learned the exact inverse operator of its own simulator; it does not test performance under the mismatches (heterogeneous RI, nuclear scattering, membrane irregularities, non-ellipsoidal adhesion geometry) that separate the phantoms from real biological cells. This assumption is load-bearing for any claim of experimental utility.
minor comments (2)
  1. [Methods] Methods: specify the precise neural-network architecture, loss function, and hyper-parameter search procedure so that the training protocol can be reproduced.
  2. [Figures] Figure captions and text: ensure consistent notation for the input absorbance spectra versus the target absorption spectra throughout.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their thoughtful and constructive review of our manuscript. We address the major comment point by point below and describe the revisions we intend to make.

read point-by-point responses

  1. Referee: [Abstract/Results] Abstract and Results sections: the in-silico test phantoms are generated with the identical FDTD ellipsoid forward model (uniform refractive index, no organelles, perfect ellipsoidal shape) used to create the training library. Consequently the reported recovery of absorption spectra and cell dimensions demonstrates only that the network has learned the exact inverse operator of its own simulator; it does not test performance under the mismatches (heterogeneous RI, nuclear scattering, membrane irregularities, non-ellipsoidal adhesion geometry) that separate the phantoms from real biological cells. This assumption is load-bearing for any claim of experimental utility.

    Authors: We agree that the current in-silico tests use phantoms drawn from the identical forward model employed for training, so the reported recoveries primarily confirm that the network has learned the inverse operator under these controlled assumptions. The manuscript already states that demonstrations are performed in silico on numerical phantoms and does not present experimental results. To strengthen the presentation, we will revise the Abstract and Results to explicitly note this limitation, clarify that the work constitutes a proof-of-concept validation of the FDTD-plus-deep-learning framework, and add a forward-looking discussion of planned extensions to heterogeneous refractive-index distributions, organelle scattering, and non-ellipsoidal geometries. We will also moderate any phrasing that could be read as implying immediate experimental utility. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper generates a synthetic training library independently via FDTD simulations of 3D ellipsoidal cell phantoms based on physical wave propagation, then trains a deep learning model to learn the mapping from simulated absorbance spectra (with scattering) back to the known input absorption spectra and geometric parameters. In-silico evaluation uses phantoms from the same forward model distribution, which is a standard controlled validation where ground truth is available by construction of the simulator; this does not equate the claimed recovery result to a fitted parameter or self-referential definition, nor does any load-bearing step reduce to a self-citation chain or ansatz smuggled from prior work. The derivation remains self-contained against the external benchmark of the FDTD physics engine, with the ellipsoid approximation stated explicitly as an assumption rather than derived from the target spectra.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on standard computational electromagnetics assumptions and the modeling choice of ellipsoidal cell geometry; no free parameters or new entities are explicitly introduced in the abstract.

axioms (2)
  • domain assumption The finite-difference time-domain (FDTD) method accurately simulates electromagnetic scattering and absorption for the modeled cell geometries.
    Invoked to generate the synthetic training spectra library.
  • domain assumption 3D ellipsoids provide a sufficient geometric approximation for both suspended (spherical) and adhered (flattened) biological cells.
    This modeling choice enables the framework to handle realistic cell shapes beyond spherical Mie theory.

pith-pipeline@v0.9.0 · 5779 in / 1384 out tokens · 47870 ms · 2026-05-22T03:45:48.132655+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

3 extracted references · 3 canonical work pages

  1. [1]

    Resonant Mie Scattering (RMieS) correction of infrared spectra from highly scattering biological samples,

    P. Bassan, A. Kohler, H. Martens, J. Lee, H. J. Byrne, P. Dumas, E. Gazi, M. Brown, N. Clarke, and P. Gardner, "Resonant Mie Scattering (RMieS) correction of infrared spectra from highly scattering biological samples," Analyst 135, 268–277 (2010)

    work page 2010

  2. [2]

    Theory of Midinfrared Absorption Microspectroscopy: I. Homogeneous Samples,

    B. J. Davis, P. S. Carney, and R. Bhargava, "Theory of Midinfrared Absorption Microspectroscopy: I. Homogeneous Samples," Analytical Chemistry 82, 3474–3486 (2010)

    work page 2010

  3. [3]

    Theory of Mid -infrared Absorption Microspectroscopy: II. Heterogeneous Samples,

    B. J. Davis, P. S. Carney, and R. Bhargava, "Theory of Mid -infrared Absorption Microspectroscopy: II. Heterogeneous Samples," Analytical Chemistry 82, 3487–3499 (2010)

    work page 2010