Automated Analysis Using a Bayesian Functional Mixed-Effects Model With Gaussian Process Responses for Wavelet Spectra of Spatiotemporal Colonic Manometry Signals.

Lukasz Wiklendt,Mark S Scott,Simon J H Brookes,Phil G Dinning,Marcello Costa

doi:10.3389/fphys.2020.605066

Lukasz Wiklendt, Mark S Scott + Show 3 more

Open Access

https://doi.org/10.3389/fphys.2020.605066

Copy DOI

Abstract

Manual analysis of human high-resolution colonic manometry data is time consuming, non-standardized and subject to laboratory bias. In this article we present a technique for spectral analysis and statistical inference of quasiperiodic spatiotemporal signals recorded during colonic manometry procedures. Spectral analysis is achieved by computing the continuous wavelet transform and cross-wavelet transform of these signals. Statistical inference is achieved by modeling the resulting time-averaged amplitudes in the frequency and frequency-phase domains as Gaussian processes over a regular grid, under the influence of categorical and numerical predictors specified by the experimental design as a functional mixed-effects model. Parameters of the model are inferred with Hamiltonian Monte Carlo. Using this method, we re-analyzed our previously published colonic manometry data, comparing healthy controls and patients with slow transit constipation. The output from our automated method, supports and adds to our previous manual analysis. To obtain these results took less than two days. In comparison the manual analysis took 5 weeks. The proposed mixed-effects model approach described here can also be used to gain an appreciation of cyclical activity in individual subjects during control periods and in response to any form of intervention.

Highlights

Colonic manometry is a procedure involving the placement of a flexible catheter incorporating pressure sensors into the colon to record contractile activity
The images contain the manometric traces constructed as PMaps (Figures 2A,E), the wavelet power spectrum of pressure waves at each moment (Figures 2B,F), the global wavelet power spectrum showing the dominant frequencies for the period (Figures 2C,G) and the global wavelet power cross-spectrum showing the dominant frequencies and their directions of propagation (Figures 2D,H)
With our novel, automated technique, we have shown at after a meal the retrograde cyclic activity between 2 and 8 cpm is of significantly greater power than antegrade cyclic activity at the same frequency [see sections “Healthy Adults vs. Patients With Slow Transit Constipation; Descending Colon (Figure 5)” and “Healthy Adults vs. Patients With Slow Transit Constipation; Sigmoid Colon (Figure 6)”]

Summary

INTRODUCTION

Colonic manometry is a procedure involving the placement of a flexible catheter incorporating pressure sensors into the colon to record contractile activity. The images contain information on propagation direction and speed of propagation and the statistical comparisons to determine if any stimulus effects differ between subject groups We have applied this analytical method to data that we had previously analyzed manually in healthy adults (Dinning et al, 2014) and patients diagnosed with slow transit constipation (Dinning et al, 2015). Depending on the experimental design, an optional offset term oη is included in equation (3.3) which may be either set to the mean of all y as a way of centring the data, inferred to include a measure of variability in the centering, or given a different value per observation if some measure of exposure needs to be incorporated that would not otherwise fit as its own predictor in X or Z. Colonic pressures were recorded for a further 2 h

RESULTS

DISCUSSION

ETHICS STATEMENT