Deep Learning-Driven ECG Synthesis from Beat Pilot Tone Signals for Physiological Monitoring and Image Reconstruction

Motivation: The electrocardiogram (ECG) is vital in MRI for cardiac gating and physiological monitoring. However, conventional ECG requires adhesive electrodes, which may compromise patient comfort and imaging experience. Goal(s): Achieve contactless ECG monitoring during MR scans. Approach: We captured cardiac mechanical motion using the beat pilot tone method and developed a deep learning network to achieve end-to-end mapping to ECG. Results: In our best-case results, contactless ECG measurements achieve trigger timing accuracy with a 19.6±8.6 ms error and 0.0059 Root-Mean-Square-Error compared to ground truth ECG. Physiological signals from our method and Siemens Physio Suite were each used for XD-GRASP reconstruction, producing comparable image quality. Impact: This work explores the potential for contactless ECG monitoring during MRI, improving patient comfort and workflow. Compared to BPT-derived cardiac signals, the synthesized ECG better suits clinicians' interpretation habits, potentially enhancing physiological monitoring. Additionally, it offers sharper signals for triggering.