Related Experiment Video
Updated: Dec 1, 2025
Creation of Patient-Specific Silicone Cardiac Models with Applications in Pre-surgical Plans and Hands-on Training
Published on: February 10, 2022
Simulator-generated training datasets as an alternative to using patient data for machine learning: An example in
Christos G Xanthis1, Dimitrios Filos2, Kostas Haris2
1Laboratory of Computing, Medical Informatics and Biomedical-Imaging Technologies, School of Medicine, Faculty of Health Sciences, Aristotle University of Thessaloniki, Greece; Department of Clinical Physiology, Clinical Sciences, Lund University and Lund University Hospital, Lund, Sweden.
Advanced MR simulations create artificial datasets for machine learning, overcoming data limitations in medical imaging. This cost-effective approach trains neural networks without real MRI scans or personnel.
Area of Science:
- Medical Imaging
- Machine Learning
- Computational Anatomy
Background:
- Supervised machine learning for MRI analysis requires extensive, representative, and well-labeled training data, which is often scarce or unavailable.
- Limitations include data scarcity, small dataset size, lack of representativeness, and weak labels, hindering model development.
Purpose of the Study:
- To develop a cost-effective method for generating artificial MRI training datasets using advanced simulations.
- To overcome limitations associated with real-world MRI data acquisition for supervised learning.
Main Methods:
- Utilized the 4D-XCAT model and coreMRI simulation platform to generate artificial short-axis MR images.
- Trained a neural network for left ventricular (LV) endocardium and epicardium delineation using simulated and real MRI data.
Main Results:
- Achieved high performance metrics (94% endocardium, 90% DICE epicardium) on real MRI data.
- Incorporating 10% real MRI data into the artificial dataset boosted performance to 97% DICE.
- Simulated data covering the entire LV, combined with real data (80%-20% mix), yielded 85% (endocardium) and 88% DICE (epicardium).
Conclusions:
- Advanced MR simulations offer a low-cost solution for creating artificial training datasets.
- This approach bypasses the need for real MRI scanners, patient scanning, and specialized personnel.
- Enables robust supervised learning model development in medical image analysis.
