A Large and Diverse Ground Truth Dataset of Simulated Realistic Cardiac Ultrasound Recordings

To address the scarcity of cardiac ultrasound datasets crucial for developing and validating machine learning-based ultrasound data processing algorithms, we have created a ground truth dataset comprising 1296 synthetic cardiac ultrasound recordings. These recordings feature realistic speckle texture derived from clinical scans and left ventricular deformation (strain patterns) generated using the CircAdapt heart model. A detailed description of the dataset creation process is provided in the parent publication titled "Large-scale simulation of realistic cardiac ultrasound data with clinical appearance: methodology and open-access database".
In this dataset, each simulated recording includes ground truth left ventricular motion, beamformed radio frequency signals, scatter maps (comprising ultrasound scattering sites and the respective amplitudes), all for the entire cardiac cycle.
Specifically, in 204 out of the 1296 recordings, the left ventricular motion is imposed using regional longitudinal and radial strain patterns for two segments of the left ventricle: the left ventricular free wall (LVFW) and the septal wall (SPW). In the remaining 1092 recordings, left ventricular motion is imposed using both global longitudinal and radial strain patterns, as well as the average of the LVFW and SPW strain patterns.
This dataset serves as a valuable resource (e.g., a potential augmentation tool) for the development and validation of new machine and deep learning-based ultrasound data processing algorithms. This work is a part of MARie Curie Intelligent UltraSound (MARCIUS) project which is funded by the European Union’s Horizon 2020 research and innovation programme under the Marie Sklodowska Curie grant agreement number 860745.