Poster L26 - Individualized Muscle Sympathetic Nerve Activity Denoising Using Valsalva Maneuver Response

Introduction: Muscle sympathetic neural activity (MSNA) is used to study the pathophysiology of autonomic disorders like baroreflex failure, orthostatic hypotension, or postural orthostatic tachycardia syndrome (POTS).

MSNA can be quantitated by the rate of action potential firing – the so-called spike rate. However, extracting action potentials from background noise in raw neural recordings is challenging. In animal studies, noise is plainly measured from postmortem neural signals, which is impossible in human studies[1]. Wavelet denoising with modified hard thresholding has been optimized and generalized by our group[2,3]. The limitation of this approach is that it does not adjust for background noise specific to individual recordings. Recording noise can depend on the equipment, environment, and subject. Here we propose an individual adaptive wavelet denoising for each subject’s recordings.

The Valsalva Maneuver – a maximal exhalation against resistance – is a common test to verify the adequacy of MSNA recordings. Sympathetic activity is activated by a drop in blood pressure during Valsalva straining and depressed by the blood pressure overshoot after releasing. We propose to use this physiological baroreflex MSNA suppression for individualized noise estimation of the denoising algorithm. We hypothesize that the individualized denoising method estimates higher MSNA responses than the generalized method.

Materials and

Methods: We retrospectively studied 17 healthy and 19 POTS female subjects who underwent autonomic function test (AFT) in an approved clinical trial (NCT04050410) at Vanderbilt Autonomic Dysfunction Center. Every subject performed a 30 mmHg Valsalva maneuver for 15 seconds. MSNA (microneurography at the peroneal nerve) was sampled at 20 kHz during Valsalva maneuver. Continuous blood pressure, heart rate, and strain pressure were also recorded simultaneously.

Data were processed on customized software written in Matlab (Physiowave©, A. Diedrich, Vanderbilt University Medical Center, TN, USA). MSNA were divided to resting supine, straining (Valsalva phase 2, VM2) and blood pressure overshoot (Valsalva phase 4, VM4)[4]. We denoised MSNA using stationary wavelet transformed with mother wavelet ‘Symlet7’ and 5 decomposition levels. Thresholds were computed from 4 times the standard deviation of VM4 MSNA for each level 3, 4, and 5. Only level 3 to 5 wavelet coefficients larger than the computed level-specific threshold were kept by hard thresholding. Inverse wavelet transformation was used to construct denoised MSNA signal. Action potential spikes on the denoised signal were counted by simple peak detection.

The MSNA spike rate was determined during each Valsalva phase for each subject. Further, we calculated the increase of spike rate in phase 2 from baseline. We compared the spike rates and phase 2 increases detected by the proposed individualized method and the two-stage kurtosis stationary wavelet transformation[2] using paired T-tests with p< 0.05 considered significant. Statistical analyses were done on Prism® software.

Results, Conclusions, and Discussions: Results
Compared to the kurtosis method, the individualized signal processing method showed significantly higher MSNA spike detection including Valsalva phase 2 spike rates of 14±6 vs. 22±8 spike/s (mean ± SD; p=0.0033) and 15±8 vs. 22±11 spike/s (p=0.0319) in healthy and POTS participants, respectively (Figure 1).
The MSNA response, measured as the spike rate difference between VM2 and baseline, was also higher in the individualized than the kurtosis method, including increases of +6±3 vs. +13±7 spike/s (p= < 0.0001) and +6±4 vs.+14±9 spike/s (p=0.0003) in healthy and POTS, respectively (Figure 2).
Discussions
We demonstrated greater spike detection for a method that takes advantage of maximal baroreflex-mediated suppression of MSNA via blood pressure overshoot during Valsalva phase 4 to compute individualized wavelet detail coefficient noise thresholds. The lower MSNA detection of the generally-optimized kurtosis method suggests it may overestimate background noise leading to an excessively high noise threshold.
The greater spike rate for the individualized method could result from either true or false detections. However, the higher spike rate increases from baseline to VM2 for the individualized method (Figure 2) better reflects the expected physiological response during straining and supports the assumption that increases were mostly due to improved true detections. The agreement between findings in healthy and POTS patients demonstrates the method’s flexibility.
Figure 1 reveals that the individualized method may substantially change POTS MSNA detection during VM2 from 7 to 39 spike/s, which may impact the differential diagnosis of POTS subtypes e.g., between neuropathic POTS and hyperadrenergic POTS. This addresses the significance of our new method when analyzing individual data, instead of the whole group.
Conclusions
We propose a new MSNA signal processing method by estimating noise for each wavelet decomposition level using individual MSNA recordings from Valsalva phase 4. This new individualized method detected higher MSNA spike rates than the current two-stage kurtosis method. The proposed method is applicable for both healthy and POTS subjects.

Acknowledgements (Optional): André Diedrich and Surat Kulapatana were supported partly by NIH grant R01 HL142583.