usct-breast-phantom: A python library to anatomically and physiologically realistic numerical breast phantoms for USCT virtual imaging studies

This software project aims at generating stochastic numerical breast phantoms with anatomically realistic tissue structures and physiologically realistic acoustic properties for use in ultrasound computed tomography (USCT) virtual imaging studies.

This software builds on top of tools from the Virtual Imaging Clinical Trial for Regulatory Evaluation (VICTRE) project. These tools are used to generate high-resolution stochastic tissue maps of breast anatomy (breast phantoms) as well as spiculated tumor numerical phantoms.

The output of the software consists of three-dimensional high-resolution maps of the breast acoustic properties, including speed-of-sound, density, and acoustic attenuation. To reduce computational cost, the user can also generate acoustic properties maps for a specific 2D slice or thin 3D slab (parallel to the frontal plane) rather then the full 3D phantom.

If you use this software for your research please cite: Fu Li, Umberto Villa, Seonyeong Park, Mark Anastasio. Three-dimensional stochastic numerical breast phantoms for enabling virtual imaging trials of ultrasound computed tomography IEEE Trans Ultrasonics, Ferroelectrics, and Frequency Control, under review, 2021. preprint

Dependencies

numpy, scipy, h5py, hdf5storage

Usage

1. Generation of tissue structure data (tissue label maps)

An example of a VICTRE phantom (raw format) is available from the Harvard dataverse. The example data for testing our code can be downloaded using the following command:

wget https://dataverse.harvard.edu/api/access/datafile/4950808 -O ./data/Phantom_set/p_324402160.mhd
wget https://dataverse.harvard.edu/api/access/datafile/4950809 -O ./data/Phantom_set/p_324402160.raw.gz

The downloaded data should be store in the ./data/Phantom\_set folder.

Alternatively, users can create their own tissue label phantoms by the use of the VICTRE software.

1. VICTRE BreastPhantom: a c++ opensource software for the generation of anthropomorphic breast phantoms
2. VICTRE BreastMass: a c++ opensource software for the generation of three-dimensional breast lesions

VICTRE configuration file templates for different breast types can be found in the folder ./data/Phantom\_cfg.

Two files are required for the execution of the next step:

• p_{phantom_id}.raw.gz: A compressed 3D anatomical data.

• p_{phantom_id}.mhd: A header file for this phantom that contains the physical location of the voxel with index [0, 0, 0], the total number of voxels in the x, y, and z, and the voxel size.

2. Assignment of acoustic properties

In this step

1. We read the 3d tissue labels map
2. We remove labels corresponding to tissues that are invisible in USCT imaging
3. We assign stochastic acoustic properties maps

Before executing this step, make sure all paths for data loading are set correctly.

To generate a full 3D NBP

python3 run_assign_properties.py -phantom_id <the digit identifier of the breast phantom> \
                                 -raw_data_path <the location of the VICTRE NBP> \
                                 -output_path <output path> \
                                 -resolution <voxel size in mm>

To generate a 2D slice from the 3D NBP

python3 run_assign_properties.py -phantom_id <the digit identifier of the breast phantom> \
                                 -raw_data_path <the location of the VICTRE NBP> \
                                 -target_slice <the z location (mm) of 2D-slice or slab center to be extracted> \
                                 -output_path <output path> \
                                 -resolution <voxel size in mm>

To generate a 3D thin slab from the 3D NBP

python3 run_assign_properties.py -phantom_id <the digit identifier of the breast phantom> \
                                 -raw_data_path <the location of the VICTRE NBP> \
                                 -target_slice <the z location (mm) of 2D-slice or slab center to be extracted> \
                                 -thickness <the thinkness of the slab in mm> \
                                 -output_path <output path> \
                                 -resolution <voxel size in mm>

where

• phantom_id is the digit identifier of the breast phantom that is part of the raw data filename.
• raw_data_path is the folder with raw anatomical data and header files
• target_slice is the target slice location (mm) to be extracted from the 3D phantom.
• thickness is the thickness (mm) of a 3D slab (centered at the target slice) extracted from the phantom.
• output_path is the folder for saving the output data.

If parameter target_slice or thickness is not specified, the full 3D phantom will be generated. An example script is given in file ./run_assign_properties.sh

Data formats

Tissue label maps

• *.mhd: the header file includes phantom size
• p_*raw.gz: phantom anatomical data

    Labels = {
        'Water':         0,
        'Fat':           1,
        'Skin':          2,
        'Glandular':    29,
        'Nipple':       33,
        'Muscle':       40,
        'Ligament':     88,
        'TDLU':         95,
        'Duct':        125,
        'Artery':      150,
        'Tumor':       200,
        'Vein':        225,
    }

Of these labels, only Water, Fat, Skin Glandular, Ligament, and Tumor are preserved.

Acoustic properties maps (2D)

speed of sound map:      a mat file with data type float32,unit mm/\mus
density map:             a mat file with data type float32,unit kg/mm^3
attenuation map:         a mat file with data type float32,unit dB/mm/Mhz^y
label map:               a mat file with data type unit8