Benchmarking deep learning-based reconstruction methods for photoacoustic computed tomography with clinically relevant synthetic datasets.
4 affiliations
- 1University of Illinois Urbana-Champaign, Department of Bioengineering, Urbana, Illinois, United States.
- 2Washington University in St. Louis, Mallinckrodt Institute of Radiology, St. Louis, Missouri, United States.
- 3University of Illinois Urbana-Champaign, Department of Electrical and Computer Engineering, Urbana, Illinois, United States.
- 4The University of Texas at Austin, Department of Biomedical Engineering, Austin, Texas, United States.
Abstract
Significance
Deep learning (DL)-based image reconstruction methods for photoacoustic computed tomography (PACT) have developed rapidly in recent years. However, most existing methods have not employed standardized datasets, and their evaluations rely on traditional image quality (IQ) metrics that may lack clinical relevance. The absence of a standardized framework for clinically meaningful IQ assessment hinders fair comparison and raises concerns about the reproducibility and reliability of reported advancements in PACT.Aim
A benchmarking framework is proposed that provides open-source, anatomically plausible synthetic datasets and evaluation strategies for DL-based acoustic inversion methods in PACT.Approach
The datasets each include over 10,000 two-dimensional (2D) stochastic breast objects with clinically relevant lesions and paired measurements at varying modeling complexity. The evaluation strategies incorporate both traditional and task-based IQ measures to assess fidelity and clinical utility. A preliminary benchmarking study is conducted to demonstrate the framework's utility by comparing DL-based and physics-based reconstruction methods.Results
The benchmarking study demonstrated that the proposed framework enabled comprehensive, quantitative comparisons of reconstruction performance and revealed important limitations in certain DL-based methods. Although they performed well according to traditional IQ measures, they often failed to accurately recover lesions. This highlights the inadequacy of traditional metrics and motivates the need for task-based assessments.Conclusions
The proposed benchmarking framework enables systematic comparisons of DL-based acoustic inversion methods for 2D PACT. By integrating clinically relevant synthetic datasets with rigorous evaluation protocols, it enables reproducible, objective assessments and facilitates method development and system optimization in PACT.
Related
- FBS-DiffNet: a deep unfolding network with diffusion prior for joint reconstruction of optical absorption and sound speed in photoacoustic tomography
- Untrained CNN and coefficient-space regularization for joint reconstruction in photoacoustic tomography
- Dynamic quantitative photoacoustic tomography under sparse sampling via residual-driven Kalman filter assisted reconstruction.
- Integrating visual state space models with frequency selection for artifact reduction in photoacoustic tomography
- A multimodal adjoint-state formulation of Quantitative Photoacoustic Tomography with external acoustic sources
BibTeX
@article{chen2026benchmarking,
title = {{Benchmarking deep learning-based reconstruction methods for photoacoustic computed tomography with clinically relevant synthetic datasets.}},
author = {Panpan Chen and Seonyeong Park and Gangwon Jeong and Refik Mert Cam and Umberto Villa and Mark A Anastasio},
journal = {Journal of biomedical optics},
year = {2026},
month = oct,
doi = {10.1117/1.jbo.32.s1.s14903},
url = {https://europepmc.org/article/MED/42876942},
note = {CC BY},
}
RIS
TY - JOUR
TI - Benchmarking deep learning-based reconstruction methods for photoacoustic computed tomography with clinically relevant synthetic datasets.
AU - Panpan Chen
AU - Seonyeong Park
AU - Gangwon Jeong
AU - Refik Mert Cam
AU - Umberto Villa
AU - Mark A Anastasio
PY - 2026
DA - 2026/10/14
JO - Journal of biomedical optics
DO - 10.1117/1.jbo.32.s1.s14903
UR - https://europepmc.org/article/MED/42876942
AB - Significance Deep learning (DL)-based image reconstruction methods for photoacoustic computed tomography (PACT) have developed rapidly in recent years. However, most existing methods have not employed standardized datasets, and their evaluations rely on traditional image quality (IQ) metrics that may lack clinical relevance. The absence of a standardized framework for clinically meaningful IQ assessment hinders fair comparison and raises concerns about the reproducibility and reliability of reported advancements in PACT.Aim A benchmarking framework is proposed that provides open-source, anatomically plausible synthetic datasets and evaluation strategies for DL-based acoustic inversion methods in PACT.Approach The datasets each include over 10,000 two-dimensional (2D) stochastic breast objects with clinically relevant lesions and paired measurements at varying modeling complexity. The evaluation strategies incorporate both traditional and task-based IQ measures to assess fidelity and clinical utility. A preliminary benchmarking study is conducted to demonstrate the framework's utility by comparing DL-based and physics-based reconstruction methods.Results The benchmarking study demonstrated that the proposed framework enabled comprehensive, quantitative comparisons of reconstruction performance and revealed important limitations in certain DL-based methods. Although they performed well according to traditional IQ measures, they often failed to accurately recover lesions. This highlights the inadequacy of traditional metrics and motivates the need for task-based assessments.Conclusions The proposed benchmarking framework enables systematic comparisons of DL-based acoustic inversion methods for 2D PACT. By integrating clinically relevant synthetic datasets with rigorous evaluation protocols, it enables reproducible, objective assessments and facilitates method development and system optimization in PACT.
SN - 1083-3668
SN - 1560-2281
ER -