Anthropomorphic phantoms with tissue-mimicking spectra

This repository contains code for the paper "Anthropomorphic tissue-mimicking phantoms for oximetry validation in multispectral optical imaging". As Python package, it provides functionalities for photoacoustic (PA) imaging and hyperspectral imaging (HSI). The package includes the following functionalities:

• Reproducibility for all results and figures in the paper including:
  • Analysis of absorption spectra of oil-based dyes
  • Analysis of PA and HSI data
  • Simulation of PA data as quality assurance
  • Spectral linear unmixing for optical absorption spectra
• Combination of absorption spectra to create tissue-mimicking spectra

Project Organization

├── LICENSE
├── README.md                           <- The top-level README for developers using this project.
├── pyproject.toml                      <- PEP 518 configuration file with requirements for the project.
└── ap                                  <- Source code for use in this project.
    ├── __init__.py                     <- Makes ap a Python module
    ├── reproduce_paper.sh              <- Script to reproduce the paper results
    ├── fit_dyes_to_spectrum.py         <- Script to fit dyes to spectrum
    ├── data                            <- Functionality to load ICG and Methylene blue data
    ├── dye_analysis                    <- Functionality to plot, anylyse and optimise dye absorption spectra
    ├── examples                        <- Simple examples to show how to do linear unmixing with an optimizer
    ├── hsi_image_analysis              <- Script to analyse and estimate oxygenation in HSI images
    ├── oxy_estimation                  <- Functionality to estimate and evaluate oxygenation in spectral data
    ├── pa_image_analysis               <- Script to analyse and estimate oxygenation in PA images
    ├── simulations                     <- Scripts for all simulation studies
    ├── utils                           <- Utility functions
    └── visualization                   <- Scripts to create exploratory and results oriented visualizations

---

• Getting started
• Run experiments
• Custom dye optimisation

Getting started

Installation and setup

To install this repository and its dependencies, you need to run the following commands:

1. Make sure that you have your preferred virtual environment activated (we recommend Python 3.10) with one of the following commands:
   • virtualenv ap and then source ap/bin/activate
   • conda create -n ap and then conda activate ap
2. Install the project by running pip install . in the anthropomorphic-phantoms directory.

Download data

The data needed to reproduce the paper results can be found here: https://doi.org/10.5281/zenodo.15102333 The folder called publication_data has the following structure:

---

├── 3D_Models                           <- 3D models of the forearm and the 3D-printable moulds
├── HSI_Data                            <- Hyperspectral data of 5 example forearms
├── Measured_Spectra                    <- Measured spectra of different dye samples (B05 - B43),
│                                          Hb (BS0) and HbO2 (BIR3) proxies with 4 intermediate oxygenation levels
│                                          (B90-B97), where BIR3 would be 100% oxygenated and all forearm background
│                                          properties (BF1-BF9) including the OOD phantom (BF10A-BF10C).
├── PAT_Data                            <- Photoacoustic data of 5 example forearms including simulations.
└── (Paper_Results)                     <- Location to store the results of the paper

---

To reproduce the paper results you will have to define the environment variable indicated in line 3 of the ap/reproduce_paper.sh script.

export BASE_PATH="<path to where you put the downloaded data, e.g. /home/user/publication_data>"

Run experiments

The easiest way of running all experiments in this repo is by using the reproduce_paper.sh script. For example, open a terminal in the ap directory and run:

bash reproduce_paper.sh

or

./reproduce_paper.sh

This script will run all experiments in the correct order and save the results in the publication_data/Paper_Results folder.

The experiments can also be run separately by running the corresponding .py scripts indicated in reproduce_paper.sh.

Run reconstructions

We provide the raw time series data of the PA images in the publication_data/PAT_Data folder with the filenames Scan_X_time_series.hdf5. In order to run the reconstructions, you can use the run_reconstruction.py script in the ap/pa_image_analysis directory. Keep in mind to adjust the base_path variable in the script to the correct path where the data is stored. The reconstructed images will be saved in the Paper_Results/PAT_Reconstructions folder which will be created.

Run simulations

Similarly, the simulations can be run with the air_bubble_simulation.py and speed_of_sound_comparison-py scripts in the ap/simulations/pat directory. They can also be run by setting RUN_SIMULATION="True" in reproduce_paper.sh. The results will be saved in the Paper_Results/PAT_Simulations folder. Make sure that you have simpa installed and all the requirements that it needs. Please refer to the simpa GitHub page for all instructions.

Custom dye optimisation

In this paper, dyes and their concentrations were selected and optimised to mimic the absorption spectra of Hemoglobin (Hb) and Oxyhemoglobin (HbO2). The fit_dyes_to_spectrum.py script in the ap directory can be used to fit dyes to any other spectrum, too. The script will use an optimizer to find the best combination of dyes and their concentrations. It will use all spectra in the publication_data/Measured_Spectra folder and print the results to the console.

how to use it yourself:

• Target Spectrum: Change the value of target_spectrum_name or pass your own target spectrum array.

• Optimization Parameters: Adjust n_iter to change the number of optimization iterations or max_concentration to alter the allowable range of dye concentrations.

• Wavelength Range: Modify unmixing_wavelengths if your spectra are defined over a different set of wavelengths. Ensure that all spectra (target and measured) are interpolated to the same wavelength grid.