qPACT Assessment via Virtual Imaging Studies

Code release accompanying the paper
Application of a Virtual Imaging Framework for Investigating a Deep Learning–Based Reconstruction Method
for 3D Quantitative Photoacoustic Computed Tomography (qPACT)

---

📌 Overview

This repository contains the source code used to conduct virtual imaging studies (VIS) for the evaluation of a deep learning–based 3D quantitative photoacoustic computed tomography (qPACT) reconstruction method.

The code supports the full experimental pipeline, including:

• preparation of network inputs via acoustic time-reversal reconstruction,
• deep learning–based joint segmentation and quantitative estimation,
• and task-specific assessment of segmentation accuracy and quantitative oxygen saturation (sO₂) metrics.

The implementation was developed to support the simulation-based studies reported in the accompanying paper and is intended to facilitate reproducibility, transparency, and reuse for related qPACT research.

---

🧠 Key Contributions Enabled by This Repository

• A virtual imaging study for evaluating qPACT reconstruction under controlled yet realistic conditions
• End-to-end integration of:
  • numerical acoustic reconstruction,
  • learning-based inference,
  • and quantitative post-hoc assessment
• Study-specific evaluation protocols for:
  • vessel and tumor segmentation,
  • vessel-wise sO₂ estimation accuracy,
  • confusion-matrix–based performance metrics

---

📁 Repository Structure

qPACT_assessment_via_VI_studies/
│
├── dataloader/
│   └── PyTorch dataset definitions and data-loading utilities
│
├── model/
│   └── Deep learning architectures used for joint segmentation and regression
│
├── scripts/
│   └── Training and inference scripts for the learning-based reconstruction model
│
├── acoustic_reconstruction_time_reversal/
│   ├── time_reversal_recon2.m
│   └── README.md
│   └── MATLAB code to generate time-reversal reconstructions
│       used as network inputs (Study 2)
│
├── assessment/
│   ├── separate_vessels_and_tumors.m
│   ├── compute_confusion_matrix.m
│   ├── compute_average_vessel_so2.m
│   └── README.md
│   └── MATLAB scripts for quantitative and segmentation-based evaluation
│
└── README.md

---

🗂 Data Availability

The numerical breast phantom datasets used with this code are publicly available via the Illinois Data Bank:

• Title: Anatomy-Matched Multi-Variant Stochastic Optoacoustic Numerical Breast Phantoms and Simulated OAT Measurement Data
• DOI: https://doi.org/10.13012/B2IDB-8164905_V1

This repository does not redistribute the dataset. Users should download the data directly from the official repository and configure local paths as needed for the provided scripts.

---

🔬 Acoustic Reconstruction (Time Reversal)

The folder acoustic_reconstruction_time_reversal/ contains MATLAB code used to generate 3D time-reversal reconstructions using the k-Wave toolbox.

These reconstructions serve as inputs to the deep learning model in Study 2 of the paper.

Main features:

• Multi-wavelength time-reversal reconstruction
• Optional noise injection and frequency-domain filtering
• Consistent reconstruction protocol across all virtual subjects

See the folder-level README for detailed usage instructions and assumptions.

---

📊 Assessment and Evaluation

The assessment/ folder contains MATLAB scripts used to compute evaluation metrics reported in the paper.

Included assessments:

• Vessel / tumor separation
• Confusion-matrix–based segmentation metrics
• Average vessel sO₂ estimation

Each script includes a clearly marked USER CONFIG section to specify:

• input/output paths,
• label conventions,
• and simple evaluation options,

while preserving the original numerical logic used in the study.

---

🚀 Typical Workflow

A typical evaluation pipeline proceeds as follows:

1. Generate acoustic reconstructions
   → acoustic_reconstruction_time_reversal/time_reversal_recon2.m

2. Train or run inference with the deep learning model
   → scripts/

3. Post-hoc quantitative assessment
   → assessment/

This modular structure allows individual components (e.g., reconstruction or assessment) to be reused independently.

---

⚙️ Requirements

Depending on which parts of the pipeline are used:

MATLAB

• MATLAB (R2020a or newer recommended)
• k-Wave Toolbox (https://www.k-wave.org/)

Python

• Python ≥ 3.8
• PyTorch
• NumPy, SciPy, h5py
• Additional dependencies as noted in training scripts

---

📄 Citation

If you use this code, please cite the accompanying paper:

@article{CAM2025100792,
title = {Application of a virtual imaging framework for investigating a deep learning-based reconstruction method for 3D quantitative photoacoustic computed tomography},
journal = {Photoacoustics},
pages = {100792},
year = {2025},
issn = {2213-5979},
doi = {https://doi.org/10.1016/j.pacs.2025.100792},
url = {https://www.sciencedirect.com/science/article/pii/S2213597925001156},
author = {Refik Mert Cam and Seonyeong Park and Umberto Villa and Mark A. Anastasio},
keywords = {Quantitative photoacoustic computed tomography, Numerical breast phantoms, Breast imaging, Virtual imaging studies}
}

---

📜 License

The license for this repository will be specified upon publication of the paper.

---