Colonic Pressure Data Processing Based On Independent Component Analysis

The Colonic motor function is critical information to help doctors understand the pathology of some intestinal diseases like slow transit constipation (STC) and colonic inertia (CI). The colonic motor function is carried out by colonic contractions, so the colonic pressure data can reflect the characteristics of the colonic motor function.Recent years, researchers found several patterns of colonic motor activity through 24-h ambulatory manometry, making it a hot issue to study the physiological significance of these patterns. For example, it is found that the high-amplitude propagated contractions play an important role in the human defecatory mechanism. However, the colonic motility activity is complex and variable, consists of patterns of colonic motility, artefacts caused by subjects'movements, noises from other physiological activities and etc, which makes it difficult to observe the information we really need.ICA is a recently developed method for finding underlying factors or components from multidimensional data. It has attracted broad attention and been successfully used in many fields, especially in biomedical signal processing. ICA needs little prior knowledge about the source signals; this makes it very suitable to the colonic pressure data, whose underlying components are still unclear to us. So far, most studies on ICA were focused on the complete cases where there are as many sensors as source signals.In this paper, we first introduced the manometry system, and next, the theories and different algorithms of independent component analysis (ICA). Making comparisons of these algorithms, we chose the Fast-ICA algorithm because it was most suitable for the colonic pressure signals. Using ICA and the Joint Time-Frequency analysis as an assistant, we found that the 26 subjects'colonic motility could be divided into 3 types: regular rhythm samples (12 subjects); slow rhythm samples (8 subjects) and disordered samples (6 subjects). Therefore, with the results of ICA, we can make preliminary judgments to the subjects'colonic motor functions. Meanwhile, ICA extracted signals of colonic motor patterns as independent components, which could greatly facilitate the further research on their physiological significance.In response to the noises from other physiological activities and the constraint of basic ICA model that sources must be as many as the sensors, this work proposed a noisy overdetermined ICA model, which used the Principal Component Analysis (PCA) as an preprocessing to ICA. PCA estimated the number of sources and reduced the dimension of observed signals to be equal to that of sources, by which, the noise were largely eliminated and the overdetermind ICA problem turned to be a complete one. Comparing to the basic ICA model, the overdetermined ICA estimated the number of colonic motor patterns that happened simultaneously, and eliminated the noise components largely. The results in the work proved the validity of this method.