Cross-modal image generation with uncertainty quantification from echocardiogram to MRI.

Area of Science:

• Cardiovascular Imaging
• Artificial Intelligence in Medicine
• Medical Image Synthesis

Background:

• Transthoracic Echocardiography (TTE) offers real-time cardiac visualization but has limitations.
• Cardiac Magnetic Resonance (CMR) provides detailed structural assessment but is time-consuming and costly.
• Bridging the gap between TTE and CMR is crucial for efficient cardiovascular diagnostics.

Purpose of the Study:

• To develop and evaluate a cross-modal generative model for synthesizing CMR-like images from TTE data.
• To leverage deep learning techniques to improve the quality and anatomical consistency of synthesized cardiac images.
• To explore the potential of AI-driven image synthesis in enhancing cardiovascular diagnostic capabilities.

Main Methods:

• A novel generative model architecture combining a UNet backbone with a vision transformer was proposed.
• The UNet component was used for feature extraction from TTE images.
• The vision transformer component was employed for global attention to enhance image synthesis quality.

Main Results:

• The proposed model successfully synthesized realistic and anatomically consistent CMR-like images from TTE.
• Quantitative and qualitative evaluations confirmed the high quality of the generated images.
• The synthesized images showed strong potential for improving diagnostic accuracy.

Conclusions:

• Cross-modal generative modeling, particularly with UNet and vision transformer integration, is effective for synthesizing CMR images from TTE.
• This approach has the potential to overcome limitations of individual imaging modalities.
• The technique offers a promising avenue for enhancing clinical decision-making in cardiovascular diagnostics.