This research addresses the challenge of detecting cardiac scars in MRI images by using synthetic data generation and domain knowledge integration to overcome limited training data.

• Combines text-based learning with synthetic data to improve performance in Late Gadolinium Enhancement (LGE) MRI analysis
• Utilizes clinical reports as a rich source of medical knowledge to enhance model training
• Employs a CLIP-based approach to bridge medical text and imaging data
• Demonstrates how domain-specific knowledge can compensate for limited annotated datasets

This innovation matters because it shows a path forward for medical AI where expert annotations are scarce but clinical text data is abundant, potentially accelerating development of diagnostic tools for cardiac conditions.

Original Paper: Fake It Till You Make It: Using Synthetic Data and Domain Knowledge for Improved Text-Based Learning for LGE Detection