Comprehensive framework of GPU-accelerated image reconstruction for photoacoustic computed tomography

The package gapat provides code of the following paper using easy-to-use Python interface with another form and realization of the algorithm.

Comprehensive framework of GPU-accelerated image reconstruction for photoacoustic computed tomography (Link)

Abstract

Significance: Photoacoustic Computed Tomography (PACT) is a promising non-invasive imaging technique for both life science and clinical implementations. To achieve fast imaging speed, modern PACT systems have equipped arrays that have hundreds to thousands of ultrasound transducer (UST) elements, and the element number continues to increase. However, large number of UST elements with parallel data acquisition could generate a massive data size, making it very challenging to realize fast image reconstruction. Although several research groups have developed GPU-accelerated method for PACT, there lacks an explicit and feasible step-by-step description of GPU-based algorithms for various hardware platforms.

Aim: In this study, we propose a comprehensive framework for developing GPU-accelerated PACT image reconstruction (Gpu-Accelerated PhotoAcoustic computed Tomography, GAPAT), helping the research society to grasp this advanced image reconstruction method.

Approach: We leverage widely accessible open-source parallel computing tools, including Python multiprocessing-based parallelism, Taichi Lang for Python, CUDA, and possible other backends. We demonstrate that our framework promotes significant performance of PACT reconstruction, enabling faster analysis and real-time applications. Besides, we also described how to realize parallel computing on various hardware configurations, including multicore CPU, single GPU, and multiple GPUs platform.

Results: Notably, our framework can achieve an effective rate of approximately 871 times when reconstructing extremely large-scale 3D PACT images on a dual-GPU platform compared to a 24-core workstation CPU. Besides this manuscript, we shared example codes in the GitHub.

Conclusions: Our approach allows for easy adoption and adaptation by the research community, fostering implementations of PACT for both life science and medicine.

Keywords: photoacoustic computed tomography, large-scale data size, GPU-accelerated method, Taichi Lang for python, multiple GPU platform.

Installation

The code requires python>=3.8, and pre-installing CUDA on your computer is strongly recommended (no specific version requirement).

We recommend using the conda environment to install the cuda toolkit. The following command installs the cuda toolkit.

conda create -n gapat python=3.9 -y
conda activate gapat
conda install -c conda-forge cudatoolkit=11.8

Then install the package using pip.

pip install gapat

Getting Started

Here is a simple example to show how to use the package. You can also read the following documentation for more details.

import time

import numpy as np
from gapat.algorithms import recon
from gapat.processings import negetive_processing
from gapat.utils import load_dat, calculate_detector_location, save_mat

# Settings
data_path = "./data"
num_channels = 256
num_times = 2048
detector_interval_x = 0.10e-3  # Scanning spacing
detector_interval_y = 0.50e-3  # Linear array spacing
x_range = [-10.0e-3, 110.0e-3]
y_range = [-10.0e-3, 110.0e-3]
z_range = [10.0e-3, 50.0e-3]
res = 0.20e-3
vs = 1500.0
fs = 40.0e6
delay = 0
method = "das"
device = "gpu"
num_devices = 1
device_no = 0

# Load data
signal_backproj = load_dat(data_path, num_channels, num_times)
num_detectors = signal_backproj.shape[0]
detector_location = calculate_detector_location(num_detectors, num_channels, detector_interval_x, detector_interval_y)
detector_normal = np.zeros((num_detectors, 3), dtype=np.float32)
detector_normal[:, 2] = 1.0  # The normal of the detectors points to the volume of planar array

# Reconstruction
start = time.time()
signal_recon = recon(
    signal_backproj,
    detector_location,
    detector_normal,
    x_range,
    y_range,
    z_range,
    res,
    vs,
    fs,
    delay,
    method,
    device,
    num_devices,
    device_no,
)
end = time.time()
print(f"Reconstruction time: {end - start} seconds")

# Save the result
signal_recon = negetive_processing(signal_recon, method="zero", axis=0)
save_mat("result.mat", "signal_recon", signal_recon)

Documentation

gapat.algorithms

gapat.algorithms.recon(signal_backproj, detector_location, detector_normal, x_range, y_range, z_range, res, vs, fs, delay=0, method="das", device="gpu", num_devices=1, device_no=0, block_dim=512)

Reconstruction of photoacoustic computed tomography.

Warning: When using multi-device reconstruction, the function must be called on the main process.

Parameters

Returns

gapat.processings

gapat.processings.bandpass_filter(signal_matrix, fs, band_range, order=2, axis=0)

Bandpass filter the signal matrix.

Parameters

Returns

gapat.processings.negetive_processing(signal_recon, method="zero", axis=0)

Process the negative signal.

Parameters

Returns

gapat.utils

gapat.utils.load_mat(filename)

Load .mat file and return a dictionary with variable names as keys, and loaded matrices as values.

Parameters

Returns

gapat.utils.save_mat(filename, varname, data)

Save data to .mat file with the given variable name.

Parameters

gapat.utils.load_dat(data_path, num_channels, num_times, dtype=np.int16, order="F", zero_set=True)

Load all .dat files in the given directory and return a numpy array.

Parameters

Returns

gapat.utils.calculate_detector_location(num_detectors, num_channels, detector_interval_x, detector_interval_y)

Calculate the location of the detectors.

Parameters

Returns