A multimodal deep learning model for cardiac resynchronisation therapy response prediction.

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Cardiac resynchronisation therapy (CRT) is crucial for heart failure management.
• Accurate prediction of CRT response is essential for patient selection and treatment optimization.
• Current prediction methods often lack sufficient accuracy and rely on single imaging modalities.

Purpose of the Study:

• To develop and evaluate a novel multimodal deep learning framework for predicting CRT response.
• To integrate data from 2D echocardiography and cardiac magnetic resonance (CMR) for improved prediction.
• To assess the performance of the multimodal approach against single-modality baselines.

Main Methods:

• Utilized the 'nnU-Net' model for heart segmentation from both 2D echocardiography and CMR data.
• Developed a multimodal deep learning classifier combining latent spaces from segmentation models.
• Evaluated the framework on a cohort of 50 CRT patients with paired imaging data.

Main Results:

• The multimodal classifier achieved a statistically significant improvement in accuracy over the 2D echocardiography-only baseline.
• Achieved 77.38% accuracy, with 83.33% sensitivity and 71.43% specificity in CRT response prediction.
• Demonstrated that the framework can predict CRT response using only 2D echocardiography at test time, leveraging learned CMR-echocardiography relationships.

Conclusions:

• The proposed multimodal deep learning framework represents a significant advancement in CRT response prediction.
• Combining echocardiography and CMR data enhances prediction accuracy, reaching state-of-the-art performance.
• This approach offers a promising tool for personalized CRT by improving treatment response prediction.