Diffuse Optical Tomography Dataset

Overview

This repository houses a simulated dataset designed for Frequency-domain Diffuse Optical Tomography (FD-DOT) experiments. Each example comprises a target volume representing 3D absorption and reduced scattering properties randomized within a biologically realistic range for breast tissue. Additionally, it includes amplitude and phase components of corresponding frequency-domain reflectance measurements simulated using a high-density grid of source/detector pairs. The dataset encompasses raw data, preprocessed data, mesh information, and supplementary metadata. A detailed description of the dataset structure and contents is provided below.

File Structure

Mesh

• data/mesh.mat: The mesh used to generate the data in Matlab using the NIRFAST package.

Main Data

• data/simulated_linescans: The main dataset.

Dataset Information

• data/simulated_linescans/dataset_info.json: Information about the dataset in JSON format.

Measurement List

• data/simulated_linescans/measurement_list.csv: A table containing information on source and detector positions for each SD pair used to generate the data. Each row corresponds to one value in the raw data measurements.

Raw Data

• data/simulated_linescans/raw/1.mat: Raw data for example 1.
  • Each /mat data file contains the following fields:
    • amplitude_clean: Amplitude measurements for each source/detector pair.
    • amplitude_noisy: Amplitude_clean plus added noise based on a system-derived amplitude-dependent noise model.
    • phase_clean: Phase measurements for each source/detector pair.
    • phase_noisy: Phase_clean plus added noise based on a system-derived amplitude-dependent noise model.
    • target: Ground truth optical properties used to simulate the data. Dimensions represent [x position, y position, z position (depth), optical property (1=mua, 2=mus’)]
    • roi_mask: Binary mask indicating the presence of anomalies in the target volume. Dimensions represent [x position, y position, z position (depth)]
    • info: Example-specific information, including the background optical properties, and spatial & contrast details of each anomaly.

Preprocessed Data

• data/simulated_linescans/prepro/prepro_info.json: Information about the preprocessing procedure used to generate this data.
• data/simulated_linescans/prepro/train
  • X.npy: Preprocessed measurement data in the train split in .numpy format. Dimensions represent [number of examples, number of measurements]
  • Y.npy: Preprocessed target volume data in the train split in .numpy format. Dimensions represent [number of examples, x position, y position, z position (depth), optical property (1=mua, 2=mus’)]
  • W.npy: Preprocessed region of interest masks in the train split in .numpy format. Dimensions represent [number of examples, x position, y position, z position (depth)]
• data/simulated_linescans/prepro/val:
  • Same structure for the validation split.

Test Disk Data

• data/simulated_linescans_testdisks: Test dataset containing manually designed volumes to test parameter separation and depth sensitivity.
  • Same structure as data/simulated_linescans but also contains:
    • data/simulated_linescans_testdisks/example_list.csv: Depth and optical property information for each test disk example.

This dataset was produced as part of a project supported by the National Institute of Biomedical Imaging and Bioengineering (NIBIB) of the National Institutes of Health (NIH) Award Number R01EB029595.
Feel free to contact Robin Dale at rbd079@student.bham.ac.uk for any inquiries or additional information.