Postprocessing algorithm for automated analysis of pelvic intraoperative neuromonitoring signals

1 Introduction

Pelvic intraoperative neuromonitoring (pIONM^®) has been introduced in rectal cancer surgery as a tool to electrophysiologically map the autonomic nerves in order to preserve urinary, sexual and anorectal function [1]. Ongoing technical and methodological developments enabled further application of pIONM^® during laparoscopic [2], [3] and robotic-assisted low anterior resection [4].

The current pIONM^® method is based on intracorporeal electric neurostimulation under needle electromyography (EMG) of internal anal sphincter (IAS) and manometry of urinary bladder. Within the BMBF-funded research project named “autoPIN”, new requirements arising from a more minimal invasive pIONM^® are met by developing and evaluating a new generation of systems and accessories. This project aims for developments of non-invasive IAS-EMG recordings and extracorporeal neurostimulation approach.

A first evaluation of the novel systems and accessories was performed in an experimental setting. A standardized and reproducible approach produced numerous data sets which needed to be analyzed. In order to ensure systematic evaluation of this data, we established an automated postprocessing routine, which combined different analysis methods and provided adequate representation of the results. This contribution describes the developed routines for automatic analysis of data obtained from EMG recordings of IAS and manometry of urinary bladder.

2 Material and methods

Four male pigs underwent low anterior rectal resection. Animal care and experimental procedures were approved by the local ethics committee (approval code G 14-1-061). Anesthesia was carried out with intravenous thiopental sodium (Trapanal^®, Nycomed, Germany) and Piritramid (Dipidolor^®, Janssen-Cilag, Germany). The perioperative management has been described by Kauff et al. [5].

For EMG recording a bipolar needle electrode (Art.-No. 530227, inomed Medizintechnik GmbH, Emmendingen, Germany) was inserted into the IAS. Additional intra-anal surface EMG of IAS was realized with a newly developed electrode design. This electrode was equipped with two conductive silicone electrode contacts which were arranged circularly for recording [5].

Manometry of the urinary bladder was carried out via a transurethral catheter which was connected to a pressure sensor (Art.-Nr. 520332, inomed Medizintechnik GmbH, Emmendingen, Germany).

Intracorporeal stimulation of pelvic autonomic nerves (inferior hypogastric plexus and pelvic splanchnic nerves) was performed with a handheld bipolar stimulation probe (Art.-No. 522610, inomed Medizintechnik GmbH, Emmendingen, Germany). Monophasic cathodic pulse stimulation was performed in the range of 0.5–6 mA with 200 μs pulse width and a frequency of 30–40 Hz. Further extracorporeal stimulation was carried out via stainless steel surface electrode pads with a diameter of 1 cm (Fraunhofer IBMT, St. Ingbert, Germany). Ten electrodes were placed over the sacral foramina. Stimulation was carried out via a carbon anode (AAM GmbH, Triberg, Germany) which was placed in four positions. Stimulation parameters were 25 mA, 30 Hz and 200 μs pulse width with cathodic rectangular shape. The trigger output of the stimulator was additionally recorded parallel to EMG and bladder manometry acquisition in order to determine the exact stimulation intervals for later signal analysis. For data acquisition, a neuromonitoring device with the NeuroExplorer software (ISIS Xpress and NeuroExplorer Version 5.1, both inomed Medizintechnik GmbH, Emmendingen, Germany) was used. The device allowed to record and display up to 16 channels simultaneously with individual filter and measurement range settings. Additionally, the unfiltered raw data was stored with 20 kHz sampling frequency.

Offline data analysis was performed postoperatively using MATLAB (Version R2009b, The MathWorks, Inc., Natick, United States).

3 Results

The automatic analysis routine combined the following steps:

• Converting the recorded data from the inomed NeuroExplorer format (.emg-files) to MATLAB

• Identifying the stimulation intervals based on trigger information from the stimulator

• Plotting single data sets of needle and non-invasive IAS-EMG in the following form:

  • Raw data (Figure 2A)

  • Filtered data (software bandpass between 5 and 20 Hz) (Figure 2B)

  • Fast Fourier transformations (FFT) of 5-second-windows of resting state (Figure 3A) and meanwhile stimulation (Figure 3B)

• Plotting single data sets of bladder manometry

• Combining the plotted data sets in a PowerPoint presentation for reference (Figure 1)

• Performing quantitative analysis of each data set of needle and non-invasive IAS-EMG and bladder manometry

Figure 1:

Graphical flow chart of the analysis of experimental data.

To interpret the huge amount of data that the algorithm had produced, the plots required automatic annotation. In order to include certain parameters of interest (e.g. stimulation parameters) in the plots, a standardized Excel file had to be prepared. The table sheet contained the information regarding the whole set of performed recordings, the used stimulation parameters (current amplitude, frequency, used system, stimulation site), intraoperative observations regarding the IAS-EMG and bladder manometry, as well as further, additional remarks. A corresponding paper version of the Excel sheet served as a template to protocol the procedure of each study. The routine allowed the analysis of 10 stimulation intervals in one single recording file.

As Kauff et al. described in [6], stimulation of the autonomic pelvic nerves results in modulation of the IAS resting activity in terms of a shift of frequency components towards higher frequencies (5 – 20 Hz). As shown in Figure 2B, filtering in the time domain was focused on these frequencies. Also in FFT special attention was payed to this frequency range (Figure 3).

Figure 2:

Graphical representation of the IAS EMG and bladder manometry. A: Raw Signal. B: Bandpass filtered signal between 5 and 20 Hz. (Note the different scales of plot A and B), C: Signal of bladder manometry.

Figure 3:

Graphical representation of frequency analysis of the IAS EMG. A: In the resting state, B: During stimulation.

The quantitative analysis was based on the calculation of features from the signals in the time and frequency domain in order to compare these features between the resting state and during stimulation (see Figures 2 and 3). This enabled the statistical analysis of the data to identify the correlation between the electrical stimulation and the neuromodulatory response, which could be identified by an amplitude increase in the filtered time domain signal or an increased power between 5 and 20 Hz in the frequency domain. For each data set, an amplitude feature from the filtered signal and a frequency feature from the FFT of the IAS-EMG of both recording modalities were calculated for the resting state and each stimulation interval. These values were stored in the Excel table, where they could easily be correlated with the applied stimulation parameters. For bladder manometry, a mean pressure in the resting state and the maximal pressure in a stimulation interval were calculated and included in the Excel sheet.

The animal studies resulted in a total of 236 data sets (mean of 59 data sets per animal experiment). All in all, 1359 stimulation intervals were recorded and had to be analyzed. In average 5.8 stimulations were performed in one data set. This resulted in a total counting of 3826 plots of graphical representations of IAS-EMG in the time and frequency domain, as well as of bladder pressure. Regarding the quantitative analysis, the four animal experiments resulted in a total of 7798 calculated values for the described features of the IAS-EMG and 1595 values for the feature of bladder manometry.

4 Discussion

The major aim of this work was the development of a postprocessing algorithm which facilitates and automates the analysis of a high number of recordings obtained from the pIONM^®.

Since our approach of a more minimal invasive pIONM^® is completely new, the optimal parameters for this kind of stimulation and recording have not yet been described.

This is why in this project we needed to generate huge amounts of data, which then had to be analyzed in order to extract these parameters (like electrode placement, stimulation current, frequency, pulse with, etc.). Thus, an automatic data analysis is a crucial tool, which helps determining the usability of non-invasive EMG recordings and extracorporeal stimulation for pIONM^®.

Like every automated algorithm, this method relies on a certain set of internal parameters and their quality, which have to be respected during recording. The number of recording channels can be varied. However, in order to be able to use the algorithm without any manual changes, the channel configuration summarized in Table 1 needs to be maintained.

Another standard condition, which has to be prepared, is the Excel table from which the algorithm uses information from and also writes results to. This table has to be manually filled, since notes are usually taken by hand during the experiment. This table also limits the maximal number of stimulation intervals to 10. This restriction has to be taken into account during the experiment.

Since the quantitative analysis of the algorithm relies on comparing features from the resting state and during stimulation, the standardized procedure of each stimulation routine needs to ensure the recording of a sufficiently long time window of the resting signal. Care needs to be taken so that during this period the signal remains undisturbed by any movement or electrical artifacts.

The results of the algorithm are displayed as a graphical visualization of the data sets in different representations (raw signal, filtered signal and FFT). The quantitative evaluation currently includes the calculation of two different features for a time window of the resting state and during stimulation. An automatic interpretation, that is the classification of the signal as a reaction to the stimulation or no reaction, is currently not part of the signal analysis routine. Defining thresholds for the single calculated features, which indicate a reaction of the IAS to the performed stimulation, is nevertheless the next upcoming challenge, which should be approached by large-scale statistical analysis of the obtained data. At one point this could further facilitate the analysis of huge data sets obtained in experimental and clinical studies by including an automatic interpretation of the neuromodulatory response. In a further outlook, this work could also be the basis for facilitating the intraoperative online representation of stimulation results and supporting the surgeon’s decision by automated analysis and interpretation of the recorded signals.