arxiv: 2604.19051 · v1 · submitted 2026-04-21 · ⚛️ physics.ins-det · astro-ph.IM· hep-ex

Recognition: unknown

Three-dimensional recoil-electron reconstruction using combined optical imaging and waveform readout for electron-tracking Compton cameras

Tomonori Ikeda , Tatsuya Sawano , Naomi Tsuji , Yoshitaka Mizumura

Authors on Pith no claims yet

Pith reviewed 2026-05-10 01:47 UTC · model grok-4.3

classification ⚛️ physics.ins-det astro-ph.IMhep-ex

keywords recoil electron reconstructionelectron tracking Compton cameraoptical imagingwaveform readoutdeep learningtime projection chambergamma ray imagingangular resolution

0 comments

The pith

Combining 2D optical images and 1D waveform signals with deep learning reconstructs three-dimensional recoil-electron directions at 44-degree angular resolution.

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

The paper aims to establish a practical method for recovering the three-dimensional direction of recoil electrons in electron-tracking Compton cameras without the data burden of full 3D readout systems. It does this by combining a high-resolution two-dimensional optical image of the track projection, a one-dimensional waveform signal for the longitudinal component, and a deep-learning model trained on simulated data for an argon-based gas time projection chamber. This matters for gamma-ray imaging because accurate recoil-electron direction reconstruction sharpens the point-spread function of the camera, and full 3D systems are impractical at large scales due to data volume. The method delivers an angular resolution of approximately 44 degrees in the 40-50 keV range, a factor of 1.3 better than prior strip-readout techniques, plus improved starting-point resolution from 5 to 50 keV. A reader would care if this hybrid approach scales to real detectors and enables higher-performance Compton cameras in applications where size and data handling are constraints.

Core claim

The central claim is that complementary information from a high-resolution two-dimensional optical image and a one-dimensional waveform signal can be combined through a deep-learning model trained on Geant4 and MAGBOLTZ simulations to reconstruct the three-dimensional recoil-electron direction in an argon-based gas time projection chamber. This yields an angular resolution of approximately 44 degrees for the 40-50 keV range, representing an improvement by a factor of about 1.3 over the previous strip-readout approach, while also enhancing the starting-point resolution of the electron track across the 5-50 keV energy range. The work shows that this partial readout strategy recovers the full 3

What carries the argument

The fusion of transverse optical track images and longitudinal waveform signals processed by a deep neural network to infer the three-dimensional recoil-electron direction.

Load-bearing premise

The Geant4 and MAGBOLTZ simulations accurately capture the real optical and waveform signals in the argon-based gas time projection chamber, and the deep-learning model trained only on those simulations generalizes to actual experimental data without major loss of performance.

What would settle it

Collecting real experimental data from the electron-tracking Compton camera and measuring the angular resolution of the reconstructed recoil-electron directions to check whether it reaches or approaches the simulated value of 44 degrees in the 40-50 keV range.

read the original abstract

Accurate reconstruction of recoil-electron directions is critical for enhancing the point-spread function of electron-tracking Compton cameras (ETCCs) in gamma-ray imaging. Although full three-dimensional (3D) readout systems achieve high-precision reconstruction, they are impractical for large-area detectors because of the enormous data volume. This study proposes and demonstrates a practical alternative for inferring the 3D recoil-electron direction in Compton scattering. This method combines a high-resolution two-dimensional optical image, a one-dimensional waveform signal, and a deep-learning-based method through simulations. The proposed method achieved an angular resolution of approximately $44^\circ$ for the recoil-electron direction in the 40-50 keV range, corresponding to an improvement of a factor of about 1.3 compared with our previous strip-readout approach using pseudo-experimental data generated by Geant4 and MAGBOLTZ simulations for an argon-based gas time projection chamber. In addition, the starting-point resolution of the electron track was improved over the previous method across the 5-50 keV electron energy range. These results demonstrate that complementary information from the transverse image and longitudinal waveform can effectively recover the 3D track topology without requiring full 3D readout. The proposed approach provides a realistic pathway for improving ETCC imaging performance.

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.

Desk Editor's Note private letter to a colleague

The hybrid 2D-optical plus 1D-waveform plus DL method gives a 1.3x simulated angular-resolution gain for ETCC recoil tracks over the authors' prior strip approach, but the entire result rests on unvalidated Geant4/MAGBOLTZ pseudo-data.

read the letter

The paper's main takeaway is straightforward: in argon TPC simulations, feeding a high-resolution 2D optical image and a 1D waveform into a deep-learning model recovers usable 3D recoil-electron direction information, reaching roughly 44° angular resolution at 40-50 keV and improving the track starting-point resolution across 5-50 keV. That is about 1.3 times better than the authors' own earlier strip-readout baseline run in the same simulation framework. The motivation is practical—full 3D readout produces too much data for large-area detectors—so the hybrid is positioned as a lower-data-volume alternative that still captures both transverse and longitudinal track information. The work is clearly an incremental engineering step on top of the group's previous publications rather than a broad conceptual leap. The simulations are described in enough detail to understand the setup, and the reported gains are consistent within that setup. The central weakness is that no real detector data appear anywhere. The abstract and stress-test note both indicate the performance numbers come exclusively from pseudo-experimental events generated by Geant4 and MAGBOLTZ; there is no quantitative check of how well the simulated scintillation, drift, diffusion, or optical signals match actual measurements. Without that, it is impossible to know whether the 1.3x improvement survives when the model is applied to hardware. Training details for the deep-learning component are also thin in the provided summary. This paper is aimed at the small community working on electron-tracking Compton cameras for gamma-ray astronomy and medical imaging. Detector physicists who already follow the authors' strip-readout papers will find the hybrid idea worth reading. It deserves peer review because the method is described coherently, the simulation comparison is internally consistent, and the engineering problem it addresses is real. Referees can reasonably ask for simulation-to-data validation and more on the network architecture and training procedure.

Referee Report

3 major / 2 minor

Summary. The manuscript proposes a hybrid readout approach for electron-tracking Compton cameras that combines high-resolution 2D optical imaging of scintillation light with 1D waveform signals from a gas TPC, processed via deep learning, to infer 3D recoil-electron track directions without full 3D readout. Using Geant4 and MAGBOLTZ simulations of an argon-based detector, the method reports an angular resolution of ~44° in the 40-50 keV range (1.3× better than the authors' prior strip-readout simulation) and improved starting-point resolution across 5-50 keV.

Significance. If the simulation fidelity is confirmed, the work demonstrates a practical, data-volume-efficient route to recovering usable 3D track topology from complementary transverse and longitudinal signals. This could meaningfully improve the point-spread function of large-area ETCCs for gamma-ray imaging applications while avoiding the hardware complexity of pixelated 3D readouts.

major comments (3)

[Results and Simulation Validation sections] Results and Simulation Validation sections: No quantitative comparison is presented between the simulated optical images, scintillation yields, diffusion parameters, or waveform shapes and any corresponding experimental measurements from the argon TPC. This leaves the central assumption—that Geant4/MAGBOLTZ pseudo-data faithfully reproduce real detector response—unverified, directly affecting the credibility of the 44° resolution and 1.3× improvement claims.
[Methods] Deep-learning model description (Methods): The manuscript provides insufficient detail on network architecture, training/validation/test splits, hyperparameter choices, regularization, and uncertainty quantification (e.g., standard errors or bootstrap estimates on the reported angular resolution). Without these, it is impossible to evaluate whether the quoted performance is robust or sensitive to simulation variations.
[Results] Comparison to prior strip-readout result: Both the new hybrid method and the baseline are evaluated exclusively within the same Geant4/MAGBOLTZ simulation framework. The factor-of-1.3 improvement is therefore internal to the model assumptions and does not yet constitute evidence of superiority under real experimental conditions.

minor comments (2)

[Abstract] The abstract and introduction should explicitly state that all results are simulation-only and clarify the absence of real-data validation to set reader expectations.
[Figures] Figure captions and axis labels should include units and simulation parameters (e.g., gas pressure, drift field) for reproducibility.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed review. The comments raise important points about simulation fidelity, methodological transparency, and the scope of the performance comparison. We address each major comment below, with clarifications and proposed revisions where appropriate.

read point-by-point responses

Referee: [Results and Simulation Validation sections] Results and Simulation Validation sections: No quantitative comparison is presented between the simulated optical images, scintillation yields, diffusion parameters, or waveform shapes and any corresponding experimental measurements from the argon TPC. This leaves the central assumption—that Geant4/MAGBOLTZ pseudo-data faithfully reproduce real detector response—unverified, directly affecting the credibility of the 44° resolution and 1.3× improvement claims.

Authors: We acknowledge that the manuscript does not include direct quantitative comparisons with experimental measurements from our argon TPC. The simulations rely on standard Geant4 particle transport and MAGBOLTZ electron drift/diffusion models, with parameters (scintillation yield, diffusion coefficients, and gas properties) taken from established literature on argon TPCs. We will revise the Simulation Validation section to cite specific prior experimental validations of these models and to explicitly discuss the assumptions and possible limitations of the pseudo-data. A full experimental benchmark is beyond the scope of this simulation study but is planned for future work. revision: partial
Referee: [Methods] Deep-learning model description (Methods): The manuscript provides insufficient detail on network architecture, training/validation/test splits, hyperparameter choices, regularization, and uncertainty quantification (e.g., standard errors or bootstrap estimates on the reported angular resolution). Without these, it is impossible to evaluate whether the quoted performance is robust or sensitive to simulation variations.

Authors: We agree that greater detail on the deep-learning implementation is required for reproducibility and robustness assessment. In the revised manuscript we will expand the Methods section to specify the network architecture (layer types, dimensions, and activation functions), the exact training/validation/test split ratios, all hyperparameter values (learning rate, batch size, optimizer, epochs), regularization techniques (e.g., dropout rates), and uncertainty quantification (standard deviation across multiple independent training runs with different random seeds). revision: yes
Referee: [Results] Comparison to prior strip-readout result: Both the new hybrid method and the baseline are evaluated exclusively within the same Geant4/MAGBOLTZ simulation framework. The factor-of-1.3 improvement is therefore internal to the model assumptions and does not yet constitute evidence of superiority under real experimental conditions.

Authors: The comparison is deliberately performed inside the identical simulation framework so that any performance difference can be attributed solely to the reconstruction method rather than to differences in detector modeling. This controlled setting provides a fair, quantitative demonstration that the hybrid optical-plus-waveform approach recovers more directional information than the strip-readout baseline under the same physics assumptions. We will add clarifying text in the Results and Discussion sections to emphasize that the reported factor of 1.3 is a relative improvement within the validated simulation and that experimental confirmation remains necessary for absolute performance claims. revision: no

Circularity Check

0 steps flagged

No significant circularity; performance metrics derived directly from simulation evaluation

full rationale

The paper's central result—an angular resolution of ~44° and 1.3× improvement—is obtained by training and testing a deep-learning model on Geant4/MAGBOLTZ-generated pseudo-data for the new hybrid readout, then comparing against the authors' prior strip-readout method run on the identical simulation framework. This constitutes a controlled, within-model benchmark rather than a reduction of the claim to its own inputs by definition or by re-fitting. No equation or procedure redefines the target metric in terms of itself, no uniqueness theorem is invoked, and the self-citation to prior work serves only as a reference baseline whose implementation is independently re-executed here. Simulation fidelity to hardware is an external-validity assumption, not an internal circularity in the derivation chain.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central performance claims rest on the accuracy of Geant4/MAGBOLTZ simulations for generating training and test data; no free parameters are explicitly fitted in the abstract, but the deep-learning model itself contains many implicit parameters.

axioms (1)

domain assumption Geant4 and MAGBOLTZ simulations faithfully reproduce the optical and waveform signals in an argon-based gas TPC
Pseudo-experimental data for training and evaluation are generated exclusively from these simulations

pith-pipeline@v0.9.0 · 5544 in / 1291 out tokens · 40400 ms · 2026-05-10T01:47:50.018485+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

30 extracted references

[1]

Siegert et al., Astron

T. Siegert et al., Astron. Astrophysics, 586, A84 (2016)

2016
[2]

C. A. Kierans et al., Astrophys. J., 895(1), 44 (2020)

2020
[3]

Siegert et al., Astrophys

T. Siegert et al., Astrophys. J., 897(1), 45 (2020)

2020
[4]

Diehl, Rep

R. Diehl, Rep. Prog. in Phys., 76(2), 026301 (2013)

2013
[5]

Chiaberge et al., Mon

M. Chiaberge et al., Mon. Notices R. Astron. Soc., 324(4), L33–L37 (2001)

2001
[6]

Tanimori et al., Sci

T. Tanimori et al., Sci. Rep., 7(1), 41511 (2017)

2017
[7]

Tanimori et al., Astrophys

T. Tanimori et al., Astrophys. J., 810(1), 28 (2015). 22

2015
[8]

Soﬀitta, IXPE the Imaging X-ray Polarimetry Explorer, In UV, X-Ray, and Gamma-Ray Space Instrumentation for Astronomy XX , volume 10397, page 103970I

P. Soﬀitta, IXPE the Imaging X-ray Polarimetry Explorer, In UV, X-Ray, and Gamma-Ray Space Instrumentation for Astronomy XX , volume 10397, page 103970I. SPIE (2017)

2017
[9]

S. L. O’Dell et al., The Imaging X-Ray Polarimetry Explorer (IXPE): technical overview II, In UV, X-Ray, and Gamma-Ray Space Instrumentation for Astronomy XXI , volume 11118, page 111180V. SPIE (2019)

2019
[10]

Bellazzini et al., Nucl

R. Bellazzini et al., Nucl. Instrum. Methods Phys. Res. A, 510(1-2), 176–184 (2003)

2003
[11]

Di Marco et al., Astron

A. Di Marco et al., Astron. J., 163(4), 170 (2022)

2022
[12]

Kitaguchi et al., Nucl

T. Kitaguchi et al., Nucl. Instrum. Methods Phys. Res. A, 880, 188–193 (2018)

2018
[13]

Li et al., Nucl

T. Li et al., Nucl. Instrum. Methods Phys. Res. A, 858, 62–68 (2017)

2017
[14]

Adams et al., Phys

C. Adams et al., Phys. Rev. D, 99, 092001 (2019)

2019
[15]

Dominé et al., Phys

L. Dominé et al., Phys. Rev. D, 104, 032004 (2021)

2021
[16]

Takao Kitaguchi et al., Nucl. Instrum. Methods Phys. Res. A, 942, 162389 (2019)

2019
[17]

A. L. Peirson et al., Nucl. Instrum. Methods Phys. Res. A, 986, 164740 (2021)

2021
[18]

Ochi et al., Nucl

A. Ochi et al., Nucl. Instrum. Methods Phys. Res. A, 471(1), 264–267, Imaging 2000 (2001)

2000
[19]

Takada et al., Nucl

A. Takada et al., Nucl. Instrum. Methods Phys. Res. A, 573(1), 195–199 (2007)

2007
[20]

Ikeda et al., Prog

T. Ikeda et al., Prog. Theor. Exp. Phys., 2021(8), 083F01 (2021)

2021
[21]

Ghrear et al., Machine Learning: Science and Technology, 5(3), 035009 (2024)

M. Ghrear et al., Machine Learning: Science and Technology, 5(3), 035009 (2024)

2024
[22]

Sauli, Nucl

F. Sauli, Nucl. Instrum. Methods Phys. Res. A, 805, 2–24, Special Issue in memory of Glenn F. Knoll (2016)

2016
[23]

Fujiwara et al., J

T. Fujiwara et al., J. Instrum., 9(11), P11007 (2014)

2014
[24]

N. S. Phan et al., Astropart. Phys., 84, 82–96 (2016)

2016
[25]

H. M. Araújo et al., Astropart. Phys., 151, 102853 (2023)

2023
[26]

F. M. Brunbauer et al., IEEE Transactions on Nuclear Science, 65(3), 913–918 (2018)

2018
[27]

Agostinelli et al., Nucl

S. Agostinelli et al., Nucl. Instrum. Methods Phys. Res. A, 506(3), 250–303 (2003)

2003
[28]

S. F. Biagi, Nucl. Instrum. Methods Phys. Res. A, 283(3), 716–722 (1989)

1989
[29]

Kaboth et al., Nucl

A. Kaboth et al., Nucl. Instrum. Methods Phys. Res. A, 592(1), 63–72 (2008)

2008
[30]

Carion et al., End-to-end object detection with transformers, In Computer Vision – ECCV 2020 , pages 213–229, Cham (2020)

N. Carion et al., End-to-end object detection with transformers, In Computer Vision – ECCV 2020 , pages 213–229, Cham (2020). Springer International Publishing. 23

2020