[Paper Review] OADAT: Experimental and Synthetic Clinical Optoacoustic Data for Standardized Image Processing

Fırat Özdemir, Berkan Lafci+3 more|arXiv (Cornell University)|Jun 17, 2022

Photoacoustic and Ultrasonic Imaging4 citations

TL;DR

This paper introduces OADAT, a standardized dataset of experimental and synthetic optoacoustic (OA) raw signals and reconstructed images across diverse acquisition geometries and parameters, enabling benchmarking of data-driven OA image processing. It provides 44 standardized experiments, trained deep learning models for limited-view reconstruction, undersampling artifact reduction, and anatomical segmentation, and open-source code to accelerate reproducible research in clinical OA imaging.

ABSTRACT

Optoacoustic (OA) imaging is based on excitation of biological tissues with nanosecond-duration laser pulses followed by subsequent detection of ultrasound waves generated via light-absorption-mediated thermoelastic expansion. OA imaging features a powerful combination between rich optical contrast and high resolution in deep tissues. This enabled the exploration of a number of attractive new applications both in clinical and laboratory settings. However, no standardized datasets generated with different types of experimental set-up and associated processing methods are available to facilitate advances in broader applications of OA in clinical settings. This complicates an objective comparison between new and established data processing methods, often leading to qualitative results and arbitrary interpretations of the data. In this paper, we provide both experimental and synthetic OA raw signals and reconstructed image domain datasets rendered with different experimental parameters and tomographic acquisition geometries. We further provide trained neural networks to tackle three important challenges related to OA image processing, namely accurate reconstruction under limited view tomographic conditions, removal of spatial undersampling artifacts and anatomical segmentation for improved image reconstruction. Specifically, we define 44 experiments corresponding to the aforementioned challenges as benchmarks to be used as a reference for the development of more advanced processing methods.

Motivation & Objective

• To address the lack of standardized, open-access datasets for optoacoustic (OA) image processing, which hinders objective comparison of new and established methods.
• To enable benchmarking of data-driven approaches for key challenges in clinical OA imaging: limited-view reconstruction, spatial undersampling artifacts, and anatomical segmentation.
• To support reproducible research by providing both experimental and synthetic OA data with consistent metadata and processing pipelines.
• To accelerate innovation in clinical OA by offering a unified reference framework for evaluating new processing algorithms.

Proposed method

• The authors collected 44 experimental optoacoustic datasets using varying transducer arrays, wavelengths, and acquisition geometries on healthy human volunteers.
• Synthetic optoacoustic data were generated using realistic physical models of light propagation and acoustic wave generation in heterogeneous tissue phantoms.
• Deep learning models, including modUNet, were trained end-to-end on the OADAT dataset to solve three core challenges: limited-view reconstruction, artifact reduction from sparse sampling, and segmentation for improved reconstruction.
• The dataset includes raw pressure signals (100×2030×256) and corresponding reconstructed images (100×256×256) across multiple experimental conditions.
• All models were trained and evaluated using a unified codebase hosted in the oadat-evaluate repository, with support for inference, fine-tuning, and model evaluation.
• The dataset is hosted on ETH Zurich’s Research Collection with a DOI, ensuring long-term persistence and open access under CC-BY-NC license.

Experimental results

Research questions

• RQ1 Can a standardized, open-access dataset of optoacoustic data improve reproducibility and objective benchmarking of image processing methods in clinical OA imaging?
• RQ2 How effective are deep learning models in reconstructing high-quality optoacoustic images from limited-view and undersampled data?
• RQ3 To what extent can anatomical segmentation improve the accuracy and quantitativeness of optoacoustic reconstructions?
• RQ4 Can synthetic data effectively simulate real experimental conditions to support training and validation of OA processing algorithms?
• RQ5 How does the integration of physical modeling and data-driven learning enhance the robustness of optoacoustic image reconstruction?

Key findings

• OADAT provides 44 standardized experiments across diverse acquisition parameters, enabling consistent evaluation of new image processing algorithms.
• The trained modUNet models achieve significant improvements in image quality under limited-view and sparse sampling conditions, reducing artifacts and enhancing resolution.
• Anatomical segmentation via deep learning improves reconstruction accuracy by enabling correct assignment of speed-of-sound and optical fluence values in heterogeneous tissues.
• The synthetic data closely match experimental data in signal characteristics, validating their utility for training and generalization.
• The open-source codebase and pre-trained models allow researchers to reproduce results, fine-tune models, and integrate OADAT into new workflows with minimal setup.
• The dataset is persistently hosted with a DOI and guaranteed 10-year retention, ensuring long-term accessibility for the research community.