This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR Biomedical Engineering, is properly cited. The complete bibliographic information, a link to the original publication on http://biomedeng.jmir.org/, as well as this copyright and license information must be included.
Electrogastrography is a noninvasive electrophysiological procedure used to measure gastric myoelectrical activity. EGG methods have been used to investigate the mechanisms of the human digestive system and as a clinical tool. Abnormalities in gastric myoelectrical activity have been observed in subjects with diabetes.
The objective of this study was to use the electrogastrograms (EGGs) from healthy individuals and subjects with diabetes to identify potentially informative features for the diagnosis of diabetes using EGG signals.
A total of 30 features were extracted from the EGGs of 30 healthy individuals and 30 subjects with diabetes. Of these, 20 potentially informative features were selected using a genetic algorithm–based feature selection process. The selected features were analyzed for further classification of EGG signals from healthy individuals and subjects with diabetes.
This study demonstrates that there are distinct variations between the EGG signals recorded from healthy individuals and those from subjects with diabetes. Furthermore, the study reveals that the features Maragos fractal dimension and Hausdorff box-counting fractal dimension have a high degree of correlation with the mobility of EGGs from healthy individuals and subjects with diabetes.
Based on the analysis on the extracted features, the selected features are suitable for the design of automated classification systems to identify healthy individuals and subjects with diabetes.
Digestion is the breakdown of food into small water-soluble molecules that can be absorbed by the intestinal epithelium [
Electrogastrography is a noninvasive technique used to measure and record the gastric myoelectrical activity associated with the process of digestion [
Diabetic gastropathy is defined as a spectrum of neuromuscular abnormalities of the stomach. In diabetic gastropathy, the normal average EGG signal (3 cpm) is disrupted by bradygastrias, tachygastrias, and other mixed dysrhythmias [
The frequency spectra of healthy and diabetic EGG signals often show an exponential increase of power toward the very low frequency range (<1 cpm) [
Feature extraction is a technique used to extract useful information that is hidden in biosignals. The selection of the appropriate feature is important, as it leads to precise analysis and high classification accuracy [
The objective of this work was to extract features from EGG signals from healthy individuals and subjects with diabetes to select useful and highly informative features for the diagnosis of diabetes. Additionally, we aimed to evaluate the correlation between the selected features and the process of digestion in both groups of individuals.
A total of 30 healthy individuals and 30 subjects with diabetes participated in this study. Participants ranged in age from 20 to 50 years.
The ethical clearance (HR/2017/MS/002) to conduct this research study was obtained from Global Hospitals & Health City, Chennai.
An EGG measurement system with 3 surface electrodes was developed and used to record the EGGs from healthy individuals and subjects with diabetes. Of the 3 electrodes, 2 electrodes were positioned on the outer curvature (fundus) and on the inner curvature (mid corpus) of the stomach with a separation distance of 5 cm between the electrodes, in accordance with the standard electrode placement protocol [
EGGs from all participants were acquired for a period of 10 minutes (
Block diagram of the electrogastrogram acquisition system.
Acquisition of electrogastrogram from a participant.
The EMD analysis was used to decompose the input EGG signal into different frequency components called intrinsic mode functions (IMFs) [
where
The feature extraction technique plays a vital role in achieving high classification accuracy in the analysis of biosignal processing. The process of feature extraction involves the transformation of raw EGG signals into a feature vector [
The peak frequency of healthy and diabetic EGG signals was extracted using the FFT. By taking the FFT for recorded healthy and diabetic EGG signals, the frequency components present in the EGG signal were plotted against an amplitude spectrum of a single side. Further, the frequency component with maximum amplitude was considered as the peak frequency of an EGG signal.
Hjorth parameters are used to characterize the information on the temporal dynamics of the measured biosignals. In this work, the Hjorth features activity, mobility, and complexity were extracted from healthy and diabetic EGG signals.
Activity represents the measurement of variance or the average power of an EGG signal. Activity is given as follows [
where (
Mobility represents the average frequency of an EGG signal. The mobility parameter is defined as the square root of the ratio of the variance of the first derivative of the signal and the variance of the signal. Mobility of an EGG signal is defined as follows:
The mobility parameter has a proportion of standard deviation of the power spectrum.
Complexity represents a measure of variability of an EGG signal. Complexity of an EGG signal is defined as follows:
The complexity parameter indicates the similarity between input EGG signals to a pure sine wave. The value of complexity converges to 1 as the shape of the signal gets more similar to a pure sine wave.
Entropy is defined as a measure of disorder associated with a system, and hence, it is a measure of information content, uncertainty, and complexity of the system.
The Rényi entropy of the sample
where
The Tsallis entropy is one of the most promising information theoretic methods for biosignal analysis. The Tsallis entropy (
where
Time domain and frequency domain are the two different possible ways in which the entropy of a biosignal can be computed. The spectral entropy of EGG signals shall be computed in frequency domain [
where
Fractals are mathematical sets with a high degree of geometrical complexity, which can model many classes of time series data as well as images [
The preprocessed EGG signals recorded from healthy individuals and subjects with diabetes were converted into a time corrected instantaneous frequency spectrogram using a spectrogram method. The spectrogram was plotted as an image with the intensities encoding the levels. The spectrogram had time on the x-axis and frequency on the y-axis [
Image entropy is defined as a scalar value that represents the entropy of a grayscale image. Entropy is a measure of disorder or randomness that can be used to characterize the texture of the input image. Images with lesser entropy have lot of black sky, less contrast, and a large number of pixels. Image entropy is expressed by the equation [
where
The HFD is a descriptor of the complexity of the geometry of a given set. The set can be the trajectory of any dynamical system and can be reconstructed from the measured data. Suppose that
For most values of
with sufficiently small
The GLCM is a sum of the number of times that the pixel with the gray level value
Contrast is a measure of the intensity (contrast) between a pixel and its neighbor pixel over the whole image. Contrast is 0 for a constant image. In general, the property contrast is also known as variance and inertia [
Using different feature extraction methods, a number of features can be extracted and, from them, effective informative features can be selected [
In this work, a genetic algorithm–based feature selection method was adapted to search, identify, and select potentially informative features from extracted healthy and diabetic EGG signal features for feature analysis. The flowchart of the genetic algorithm is shown in
Of the 30 features extracted from preprocessed healthy and diabetic EGG signals, the 20 best features were chosen using a genetic algorithm–based feature selection method.
Flowchart of the genetic algorithm.
Different patterns of EGG signals were observed in healthy individuals and subjects with diabetes.
(A) Typical electrogastrogram signal recorded from a healthy individual. (B) The single-sided amplitude spectrum of a healthy electrogastrogram signal.
(A) Typical electrogastrogram signal recorded from a subject with diabetes. (B) The single-sided amplitude spectrum of a diabetic electrogastrogram signal.
The variation of spectral entropy values was evaluated as a function of mobility of EGG signals recorded from healthy individuals (
The variation of HFD values was investigated as a function of mobility of EGGs recorded from healthy subjects (
Variation of spectral entropy values as a function of mobility of electrogastrogram signals. (A) Healthy individuals. (B) Subjects with diabetes.
Variation of the Hausdorff box-counting fractal dimension values as a function of mobility of electrogastrograms. (A) Healthy individuals. (B) Subjects with diabetes.
The values of MFD were evaluated as a function of mobility of EGGs acquired from healthy subjects (
The average Hjorth parameters activity, mobility, and complexity of EGG signals were recorded from healthy individuals and subjects with diabetes (
The MFD and HFD values of the EGG signals were recorded from healthy individuals and subjects with diabetes (
Variation of the Maragos fractal dimension values as a function of mobility of electrogastrograms. (A) Healthy individuals. (B) Subjects with diabetes.
Hjorth parameters (mean) of electrogastrogram signals recorded from healthy individuals and subjects with diabetes. Error bars indicate standard error.
The fractal dimension (mean) of electrogastrogram signals from healthy individuals and subjects with diabetes. Error bars indicate standard error. (A) Maragos fractal dimension. (B) Hausdorff box-counting fractal dimension.
Type 2 diabetes is a chronic disease that prevents the physiological system from using insulin efficiently. It is expected that the global number of type 2 diabetes cases will reach around 450 million by 2030. Undiagnosed diabetes is often associated with complications such as cardiovascular and kidney diseases. However, these risk factors are preventable by the early detection and diagnosis of diabetes [
Human gastric myoelectrical activity can be measured using a noninvasive technique known as EGG. However, although frequency characteristics are one of the most significant parameters, the visual analysis of EGG signals is very difficult. Subjects with diabetes who have poorly controlled diet habits are often suspected of diabetic gastroparesis. In this work, features such as time domain features, frequency domain features, and time-frequency domain features were extracted from EGG signals recorded from healthy individuals and subjects with diabetes. Further, potentially informative features were selected using a genetic algorithm–based feature selection method. Additionally, the correlation of the extracted features with the mobility of the digestive system was analyzed. Results demonstrate that the extracted features grasp individual informative characteristics that can be used for analysis. Further, the features MFD and HFD have a high degree of correlation with the mobility of healthy and diabetic EGG signals. Additionally, the spectral entropy of EGG signals recorded from healthy individuals is highly correlated with the mobility of EGG signals recorded from healthy individuals and subjects with diabetes. This work appears to be of high clinical significance, as these extracted potentially informative features can be used for the analysis and classification of digestive system disorders. In the future, deep learning techniques can be utilized for the automated classification of healthy and diabetic EGG signals.
electrogastrograms
empirical mode decomposition
fast Fourier transform
Gray Level co-occurrence matrix
Hausdorff box-counting fractal dimension
intrinsic mode function
Maragos fractal dimensions
None declared.