Approach: Using simultaneously recorded cardiac (e.g., EKG, pulse-ox) and BOLD signals (i.e., fMRI) we developed a machine learning pipeline to predict continuous fluctuations in heart period (inter-beat intervals) over time in individual participants.

We did this using high powered, within-subject samples collected at different scanners and at different magnetic fields (3T and 7T).

Our primary goal was to see if we could get reliable prediction of heart period at the single subject level.

2/6

Because of the hemodynamic lag (it takes ~6-8 seconds between neural firing and the fMRI BOLD response peaking), we adopted a temporal lag analysis. This afforded us to distinguish associations due to artifacts (e.g., pulsatile waves, global blood oxygenation changes), which happen within 0-2 seconds, and drive signals from cortex (happening around 8 seconds).

3/6

Results: We find that you can reliably predict instantaneous heart rate *at the single subject level*, using the whole-brain lag shifted BOLD signal. We also show that regions of the so-called visceral control circuits, contribute heavily to this prediction.

4/6

We go on to show that models trained on one participant can, to a weaker degree, reliably predict heart rate in another participant. This is true both within samples collected at the same scanner and from data collected on different scanners.

5/6

Conclusion: It is possible to reliably predict instantaneous heart rate from global brain signals (after some cleaning). These prediction models generalize across people.

This work teases at the possibility of building a measure of the neural control of cardiac function that can be used to develop a biomarker of cardiovascular disease that is currently missing in the medical field. Stay tuned for more on that.

6/6