OpenBreastUS Dataset

Overview

OpenBreastUS is a large-scale wave equation dataset consisting of 8,000 breast phantoms and 16,384,000 wavefields. The phantoms are immersed in a water medium to replicate real experimental conditions.

Dataset Structure

To represent the distribution of diverse human breast types, the dataset is divided into four types.

• Heterogeneous (HET): 2,000 phantoms.
• Fibroglandular (FIB): 3,000 phantoms.
• All fatty (FAT): 2,000 phantoms.
• Extremely dense (EXD): 1,000 phantoms.

How to Access the Dataset

1. Download Sample Data: Users can download a 500 kHz frequency-band sample dataset directly from Huggingface to validate data format and workflow compatibility.
2. Download Speed Data: Download breast_train_speed.mat and breast_test_speed.mat from the OpenBreastUS Repository and Split them into individual files using split_data.m.
3. Generate Full Dataset: Install the provided data generation software from OpenWavesInstaller.zip, to generate wavefield data for all frequencies (0.30 MHz to 0.65 MHz). The software supports customizable simulation parameters to replicate real experimental conditions. You can compare the first 4 entries in your generated test dataset with the provided sample dataset to verify the consistency.

OpenWaves.exe User Guide

1. Setup Requirements

Input Data: the split speed data (the output of split_data.m.).

Software: MATLAB Runtime (if compiled) or MATLAB ≥2020b.

Hardware: Multi-core CPU (≥16 cores recommended).

3. Execution Steps

Launch GUI:

Configure:

​Data Type: Select organ (breast).

​Output Directory: Save path for generated data.

​Train/Test Speed Directories: Paths to split speed data.

​CPU Cores: Optimize for your system (default: 40).

​Simulation Frequency: Frequency （100kHz） (default: 5.0).

Data Generation:

​Training Data: Simulates 256 transducer sources per phantom, saves as train_X.mat.

​Testing Data: Generates test_X.mat with same parameters.

​Log File: Progress/errors recorded in log.txt (output directory).