Alertness assessment during resting state fMRI studies is crucial. However, EEG is not easy to use in such case. Recent studies have showed that some vegetative variables easier to record in MRI room may predict alertness.

We have tested a cardio-respiratory coupling index, MRI compatible, in order to know if it can be use during resting state fMRI studies.

In 16 young normal subjects, a polysomnographically recorded nap was performed twice at least two weeks apart outside a MRI room. In one hand, sleep stages were scored based on EEG, EOG and EMG derivations. On the other hand, ECG, plethysmographic and respiratory signals were analyzed, mainly with specifically developped Matlab routines, in order to compute heart rate variability indexes, pulse transit time and cardiorespiratory coupling index based on both breathing cycles and tachogram.

Compare to wake stage, subjects in N2 showed a longer heart beat interval (1.04±0.17 vs. 1.10±0.16, P<0.01) associated with a higher variability (RMSSD: 59.4±30.3 vs. 72.7+- 31.7, P<0.05) and a less tidal volume (704.7±327.4 vs. 439.3±278.2, P<0.001). Paradoxically, cardiorespiratory coupling index was highly significantly increased (0.35±0.13 vs. 0.53±0.15, P<0.001).

Results show that cardiorespiratory coupling assessment would be a useful alternative to evaluate alertness when polysomnography recordings are hard to achieve. Ongoing analysis will be the comparisons of the performance of machine learning predictions based on the previous features of the signals and automatic feature extractions achieved using recurrent and convolutional deep learning networks.