CLASSIFICATION OF COLONIC SIGNALS FOR CONSTIPATION DETECTION BASED ON EWT AND T-FORWARD FEATURE SELECTION

Slow transit constipation (STC) is a kind of functional constipation with high clinical incidence. Colonic pressure signals play a crucial role in the diagnosis of slow transit constipation (STC). However, there are currently no ideal diagnostic tests for STC, making rapid and accurate identification essential. In this study, a method for constipation recognition based on Empirical Wavelet Transform (EWT) and T-forward feature selection was proposed based on gastrointestinal pressure signals obtained from patients and normal subjects ingesting wireless capsules containing pressure sensors. Initially, the signal undergoes multi-scale analysis using the Empirical Wavelet Transform, decomposing it into three distinct frequency bands. These frequency segments are then used to derive the integrated feature set. Here, a combination of nonlinear and time-domain features was utilized. The median values of the nonlinear features, Sample Entropy (SampEn) and Correlation Dimension (CorrDim), were consistently higher in patients across all frequency bands compared to those in healthy subjects, indicating that the dynamic behavior of gastrointestinal pressure signals in STC patients is more complex. Subsequently, an independent T-test is employed for the first round of feature screening based on [Formula: see text]-values, followed by a second round of feature selection through forward search according to the significance of the [Formula: see text]-values. After the features are selected, they are applied to the classification model. An evaluation of 113 relevant data instances was conducted and the results indicated that the classification accuracy for constipation reached 91.18%. The findings highlight the promising capability of the proposed algorithm for categorizing colonic signals and detecting episodes of constipation.