The gastrointestinal electrical mapping suite (GEMS): software for analyzing and visualizing high-resolution (multi-electrode) recordings in spatiotemporal detail

GEMS is implemented using MATLAB® R2009a (The MathWorks Inc., Natick, MA, USA), a licensed program which combines sophisticated mathematical operations and GUI compatibility. The benefits associated with using MATLAB include ease of use, in-built libraries of mathematical and graphical routines, and the ability to run on several operating systems (MS-Windows, Linux, and Mac OS X). GEMS can be run as a stand alone application that does not require the user to own a Matlab license.

Data analysis in GEMS is divided into three stages: pre-processing, processing and post-processing as shown in Figure1. In the pre-processing stage, the ‘raw recorded data’ input is converted to a file that is visualized and filtered in MATLAB. Channel selection controls allow the user to discard electrodes that contain no reliable recorded data (e.g., due to poor contact of the electrode with the GI tract or a technical fault). The output of the pre-processing stage is the ‘filtered data’, which becomes the input for the processing stage. Using the inbuilt algorithms, activation times can be automatically detected and the marked events can be grouped or ‘clustered’ into a series of wavefronts. The output of the processing stage is ‘marked clustered events’, which becomes the input for the final post-processing stage. In this stage, pseudo-colored contour maps can be produced to show the propagation and distribution of activation times, amplitudes and velocity fields. In addition, propagation movies can be produced to allow animated visualization of the spread of electrical activity.

At each stage, the user can interact with the program, perform processing steps, and tune multiple parameters to their needs via a user-friendly GUI. Further explanations of each stage of the software implementation are now provided in detail.

Upon launching GEMS, the user is prompted to provide a file containing the recorded raw electrical data, or a previously saved file. GEMS is currently configured to work with the ActiveTwo System (BioSemi, Amsterdam), which generates data files in a 24 bit version of the European Data Format (.bdf). However, GEMS could also be readily configured to work with a number of other acquisition systems producing a file that can be imported into MATLAB, as per the needs of the user community. Signal files in simple text format can be imported.

The distribution and inter-electrode spacing of the electrodes in the recording array must be specified in order to generate maps of the propagation of the electrical activity along and around the GI tract. This information is stored in an appropriately formatted file (‘the electrode configuration file’) for input, or standard templates can be generated from within GEMS. During the pre- and processing stages, the user can view and orient the electrode configuration to match the experimental positioning of their array on the GI tract to aid visualization of the processed data.

GEMS uses a large set of parameters for both the GUI and the back-end algorithms. These parameters cover a substantial range of functions, including filtering methods, the “Falling-Edge, Variable Threshold” (FEVT) and “Region Growing using Polynomial Surface Stabilization” (REGROUPS), algorithm tuning options (discussed below), activation time map design (e.g., isochronal intervals, plotting of an electrode grid, contour smoothing and use of interpolation), and propagation animation settings (e.g., start and end times, wavefront colors, supplementary text).

Technical knowledge of the algorithms used by GEMS (Table1) is helpful in order to tune the parameters for optimal software performance and graphical outputs. For example, modifying parameters for event detection and clustering (see below), according to the profiles of individual data sets, can enable more accurate results in response to particular types of noise or dysrhythmic events [13, 14]. However, a default set of parameters that is suitable for typical gastric applications has been predefined in the software, meaning that the software should also be readily useable by researchers not familiar with data or signal processing principles.

A ‘parameters GUI’ is incorporated into GEMS, which groups all of the parameters systematically according to their functions. This GUI allows the user to alter the parameters, save them, and also load alternative previously defined parameter sets. A brief description of each parameter is displayed next to each option.

The core back-end analysis functions within GEMS are comprised of slow wave analysis algorithms that span the analysis process from slow wave event detection to graphical visualization. Brief descriptions of these algorithms are provided in the following sections, and they are listed in Table1 together with their validation references. GEMS is readily extensible such that new analysis algorithms developed by the user community can be readily incorporated.

Filters can be applied to the gastric bioelectrical signals to eliminate baseline wander (such as body movement artifacts) and high frequency noise (such as power line interference) [21, 22]. A range of filter types are available, ranging from low pass, high pass, band pass, and averaging filters. The filters that are currently implemented in GEMS are Butterworth, Wavelet, Savitzky Golay and Moving Median filters [23‐25]. Recent work suggests that the moving median and Savitzky-Golay filters are among the most appropriate filters for many mapping purposes [21]. Users can adjust the filter parameters as required.

As in cardiac mapping, a fundamental step in GI mapping is the detection of the biphasic or triphasic slow wave depolarization events, which approximate the second derivative of the transmembrane potential, and correspond to the arrival of the depolarization wavefront at the region sensed by the electrode [13]. Determining these activation times is the first essential step in the mapping process, and must be achieved to generate isochronal, velocity, amplitude and time-interval maps from the HR data.

The FEVT (Falling-Edge, Variable Threshold) algorithm automates detection of slow wave activation times in GEMS. The FEVT algorithm identifies relatively high-energy, high-frequency, downward deflecting components in the pre-processed recordings. FEVT is described in detail with its validation in [13]. When the signal under analysis crosses the time-varying threshold, a slow wave event is marked with a red point. The FEVT algorithm increases data processing speed by ~100x compared to manual marking, while maintaining high sensitivity (~90%) and low false-negative and false-positive rates (~10%), even when the recorded signal-to-noise ratio is relatively low [13, 26].

The second essential step in the mapping process is to ‘cluster’ the many individually marked activation times into groups that represent independent slow wave cycles (i.e., such that all data points in a clustered group relate to the same propagating wavefront). The REGROUPS (Region Growing using Polynomial Surface Stabilization) method is used to cluster activation times in GEMS (described in detail with its validation in [14]).

The REGROUPS algorithm uses a recursive search technique in combination with a continuously updated 2^nd order spatiotemporal filter. The marked activation times are searched and the algorithm predicts when activation times should occur at adjacent electrodes. Once the activation times are chosen, they are grouped into ‘cycles’ or wavefronts. REGROUPS has been shown to properly group activation times from normal and abnormal propagation patterns, including when data are of relatively patchy quality [14]. However, manual correction may be required in complex cases (see below).

Producing contour maps of activation times (isochronal) is a highly effective means of displaying a large volume of data in an intuitive graphical display that can be readily interpolated and understood [27]. In an isochronal map, contour lines are drawn at or between electrode points where depolarization occurred at the same time, and the successive contour lines (and intervening pseudo-colored ‘isochronal bands’) quantify the propagation sequence.

A two-step ‘Spatial Interpolation and Visualization’ (SIV) scheme is implemented in GEMS to aid visualization of slow wave patterns to account for patchy data quality when necessary [14]. The user is presented with the option to interpolate between data values. In brief, if a blank site (no marked activation times) is surrounded by a threshold number of marked sites, then the blank site’s activation time is interpolated. The second stage repeats a similar process, this time using marked sites interpolated during the first stage, as well as the initially marked sites to fill in the missing borders. This scheme was implemented due to its practicality and simplicity. Full details of the SIV scheme and its validation are given in [14].

The velocity and amplitude of the propagating wavefronts can be estimated/computed using GEMS. These computations provide valuable spatial and statistical data on regional variations in the speed, direction and extracellular amplitude of slow wave propagation from cycle to cycle, which are important variables for understanding slow wave excitation in health and disease. In the normal human stomach, for example, HR mapping has recently shown that normal slow wave excitation is characterized by marked velocity and amplitude transitions occuring between the normal pacemaker region and the corpus, and within the gastric antrum [8]. Abnormally high velocity and amplitude activity has also recently been found to routinely accompany gastric dysrhythmias, and the velocity and amplitude data provided by GEMS can, therefore, assist in identifying and understanding the onset, location and pattern of abnormal events [28, 29].

The default method for velocity field calculation in GEMS is a finite difference approach that incorporates interpolation and Gaussian filter smoothing functions [15]. According to a recent validation study, this velocity calculation method is superior to alternative approaches that have been used previously in the GI field, such as simple finite differences [1, 5], and polynomial-based methods [11, 30], because it has a low average error and aids accurate visualization of velocity fields [15]. The above two alternative approaches are also available for use in GEMS, however, if desired by the user.

The amplitude of the gastric signals is obtained by taking a 1.5 s window of the signal for analysis, based on the point of interest (detected by FEVT). In the section of the signal considered for analysis, the amplitude of the gastric signal is estimated using a peak-trough detection algorithm using the ‘zero-crossing’ of the first and second order signal derivative [21, 31]. This approach is superior to simple peak-trough detection algorithms, as commonly used previously in the GI HR mapping field [5, 20], because it better accounts for competing noise and slow wave fractionation, thereby improving accuracy [21].

The magnitude of the amplitudes and velocities are represented with colors at associated electrode points, and the directions of the velocity field are displayed as arrows overlaid at the electrode points on the maps [15, 21]. Examples of amplitude and velocity map generation and output are provided in the following section.

GEMS also incorporates algorithms for calculating the frequency of wavefronts. To achieve this step, GEMS calculates the intervals between activation times recorded by each electrode in the array between consecutive cycles [7, 8]. These outputs can be viewed as ‘time-interval maps’, which show the spatial pattern of the time intervals between consecutive cycles, together with the mean and standard deviation values. These features enable ‘bradygastric’ (reduced slow wave frequency) and ‘tachygastric’ (rapid slow wave frequency) dysrhythmic events to be detected, for correlation with the propagation patterns revealed by the isochronal and velocity field maps.

Once the raw data file is read into GEMS, the user is presented with the pre-processing stage display as shown in Figure2. In this stage, the recorded data can be viewed and filtered using a GUI. The user can view the recorded data from each electrode individually, or from multiple electrodes at a time as a stacked plot. The duration of the viewed data can be altered to show a narrow time window of events in finer detail, or a wider time window in coarser detail.

The main purpose of the pre-processing stage is to perform signal processing tasks, such as removal of baseline drift and noise. A range of filter options are currently available to the user to fulfill these tasks (detailed above), the effects of which are displayed live in the pre-processing stage. The user can also mark electrodes that demonstrate a poor signal-to-noise ratio, or no signal, such that they will not be considered for activation time detection. These channels can otherwise induce false positives into the FEVT algorithm results, leading to inaccuracies in the mapping outcomes [13]. New methods are currently being developed to automatically detect channels with poor signal quality, allowing them to be deleted prior to further data processing [32].

The data prepared in the pre-processing stage is passed into the processing stage, incorporating the channel numbers, time window, baseline removal, filtering methods and other parameter settings specified by the user. Slow wave activation time marking occurs automatically via the FEVT method [13]. The default FEVT parameters that are implemented have been optimized for the processing of gastric signals; a user wishing to tune these parameters further can access them conveniently through the parameter selection window. Parameters for FEVT processing of small intestine slow wave signals are also currently being identified [33].

The automatically marked data are viewed within a new window, where they are manually reviewed for further processing. The false positive and negative rates of FEVT depend on the quality of the input raw data, filtering methods and the choice of tuning parameters [13]. Considering the large quantity of recorded data, it is essential that the user can review the marks and manually correct these irregularities in the easiest and most efficient way possible.

One important feature of the GUI is, therefore, providing the user with the flexibility to visually inspect the signals, select single or multiple electrode/s, and modify the marked events with a mouse click. This method emulates the approach that is used in SmoothMap and several cardiac mapping systems [18]. In this approach, adjacent electrode channels can be viewed as a stacked column of electrograms, to visually evaluate mark quality and the ‘time lag’ between activation times in adjacent channels that is a hallmark of propagating activation sequences. Figure3(a) shows the GEMS ‘channel selection’ window, comprising an example electrode configuration matrix displayed in the form of buttons. The user can select electrodes to view, by row, column, or free choice, and the selected electrode signals are then displayed in a stacked column in a separate figure for editing, as per the example in Figure3(b). Using the processing buttons on the left hand side of the GUI, shown in Figure3(b) the user can delete or add activation time marks from the displayed electrodes using a cursor target.

Each electrogram in the processing window is scaled to its amplitude range, which is displayed next to the signal. The electrograms can also be displayed according to a common amplitude scale, and exported as a high-quality image of a size specified by the user.

Once the activation time marks are prepared, they can be automatically clustered into common wavefronts (‘cycles’) using the REGROUPS algorithm outlined above [14]. An alternative or complimentary manual clustering option is also available, which is useful for correcting abnormalities that can occur when REGROUPS is applied to data of poorer quality, of highly atypical or abnormal pattern that REGROUPS cannot adequately handle, or when the user wishes to view an alternate clustering option.

Using the processing and display controls the user can review the REGROUPS results in stacked electrograms via the same channel selection method described above as shown in Figure3(b). Clustered cycle numbers and markers are uniquely colored by cycle number to guide visualization of the clustered results. Ungrouped events, termed ‘orphans’, are marked as green squares without numbers. These orphans may represent isolated activities, FEVT false positives, or REGROUPS false negatives, and can be manually distributed into numbered wave clusters if desired by the user.

Once the data are processed to the user’s satisfaction, the user is able to generate data maps, tables, figures and movies to provide visual interpretation of the experimental data and use them for publication or presentation purposes. A detailed description of the different visual options and their underlying algorithms is given in the following sections.

As detailed above, activation time (isochronal) maps are often among the most valuable results derived from multi-electrode mapping studies, conveying information about the pattern, direction, speed and variability of propagation [27]. Activation maps for clustered marked events are generated in GEMS according to the parameters specified by the user (e.g., isochronal spacing, interpolation methods and color range).

GEMS plots the identified activation times in the spatial arrangement determined by the electrode configuration file. Either simple ‘patch plots’ of the activation times or contoured smoothed isochronal maps can be plotted. If contoured plots are used, the time bands between the isochronal bars are pseudo-colored according to a red-blue spectrum, as per the example shown in Figure4(a). The electrode array can also be superimposed on the map as a grid of circles; black circles indicate electrodes with detected marked events, and white circles indicate electrodes with no detected events (including where data were interpolated according to the algorithm described above) [14].

In addition to viewing the activation time maps, users can also view spatial maps of the amplitudes, velocities and time intervals (frequencies) of the detected slow waves comprising each wavefront. These data are generated according to the algorithms described above. For the velocity fields, arrows represent direction of propagation and are overlaid on a pseudo-colored ‘speed map’ Figure4(b). For the amplitude map, a pseudo-color is used to represent the magnitude of the wave at each electrode site Figure4(c). Time-interval maps are displayed as ‘patch plots’.

For all of the different map types, GEMS is designed to allow the user to view three wavefronts at a time as illustrated in Figure4(a), and the user can navigate between the different waves. This feature allows the user to compare a current wavefront with the previous and successive wavefronts. The plotted maps can be saved into high-quality image files for the purpose of analysis, publications and presentation. Examples of results from GEMS can be found in a recent publication, including maps of dysrhythmic slow wave patterns [10].

In addition, the values of the activation times, time intervals, amplitudes and velocities of the selected events can be saved to text files for statistical analysis (i.e., non-interpolated data only). Marked data can also be exported in a text file format suitable for import into SmoothMap.

Movies are used to visualize the propagation of the marked events along the electrode matrix as a function of time, and are a particularly useful aid for understanding or presenting complex data sequences [9, 27]. Activation times are colored in sequence over an array that is arranged in the same manner as the electrode configuration file. The colored pixels are then set to fade before turning off again to simulate visualizations of the wavefront’s ‘refractory tail’. Wavefront sequences can be animated according to the clustered or unclustered times. If clustered times are used, then each successive wavefront can be uniquely colored to improve clarity of visualization. The user chooses the start and end times of the desired recording period, and can set the frame rate and duration of the displayed ‘tailing edge’. The propagation movies can be exported as per the example in Additional file 1. In addition to the saved files, a movie player is generated to allow the user to scroll through the files, pause, fast forward, loop, etc. Examples of movies of dysrhythmic activities generated through GEMS can be found in a recent publication that employed the software [10].

GEMS allows the user to save the filtered data with the marked events (clustered or unclustered). The file will also automatically save the electrode configuration and the default parameters used to obtain the results. Alternatively, the user can also save the electrode configuration and the default parameters into isolated files for the flexibility of using them with other experiments. To reload analyzed data, at start-up the user can specify the analyzed data file and GEMS will launch in the processing mode.

The GEMS Software has undergone both quantitative validation (detailed in Table1), and qualitative validation. Qualitative validation has been recently demonstrated in a published experimental study in a porcine model, in which a range of gastric dysrhythmias were characterized, quantified, and classified, including several new patterns, such as rapid slow wave re-entry in the gastric corpus [10]. Further clinical works are currently progressing [12].

Figure5 demonstrates a specific example of the application of GEMS in the mapping of gastric dysrhythmia, in a porcine model. The GEMS outputs demonstrated in this figure include isochronal, velocity, amplitude, and time-interval maps, statistical parameters, and an animation sequence.