Ultasound Elastography Simulation Database for Deep Learning

The Dataset contains 2400 simulated phantom of size 40*32 mm with the center frequency of 5 MHz. There are 24 different models with hard inclusions. 100 images are simulated for each phantom with 10 different average strain from 0 to 4.5 % and 10 different random scatterers. The Ground Truth contains axial and lateral displacements with respect to the first image (pre-compression). The motion is linear so that different displacements pairs can be calculated. Also, there are simulated images with a center frequency of 10 MHz for 19 models.

There are one or two hard inclusions in each model with Young modulus in the range 45-60 kPa. The tissue Young modulus is around 20 Kpa and Poisson Ratio is 0.49. FEM is done by ABAQUS software and publicly available FIELD II is used to simulate images. There is no intensity difference between the tissue and the inclusions. The parameters for simulation are given in the readme file.

If you use this database, please cite the following paper :

A. K. Z. Tehrani and H. Rivaz, “Displacement Estimation in Ultrasound Elastography using Pyramidal Convolutional Neural Network,” in IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control.

Latest Updates:

  • Ground truth displacement accuracy has been improved.
  • Models with a center frequency of 10 MHz are added.