Bowel-sound pattern analysis using wavelets and neural networks with application to long-term, unsupervised, gastrointestinal motility monitoring
Introduction
During the last decades, the field of biomedical engineering has received a lot of attention from researchers, allowing technology improvements to have a positive impact on medical knowledge and practice. Expert systems and artificial intelligence methods have been implemented in a variety of applications (Aruna et al., 2005, İçer et al., 2006, Jiang and Choi, 2006, Polat et al., 2006, Turkoglu et al., 2002, Subasi, 2006, Yan et al., 2006), offering better knowledge handling and diagnostic decisions support utilities. These benefits are observed in various fields of medical expertise, including the critical area of gastroenterology, where computer assisted expert systems have been reported for delivering e-health services (Aruna et al., in press, Huang and Chen, 2006, Kuo et al., 2004). However, there is a lack of a reliable and easy to apply method for long-term monitoring of intestinal contractile activity that still makes difficult to study the gastrointestinal motility physiology (Drossman, 1999, Talley et al., 2000). As a result, there are certain gastrointestinal functional disorders that are still diagnosed indirectly, via elimination of other diseases. The irritable bowel syndrome (IBS) is the most representative case, while other functional disorders such as abdominal bloating, functional dyspepsia, diarrhea, constipation and abdominal pain fall in the same category (Drossman, 1999, Talley et al., 2000). Most of these symptoms are strongly affected by various factors like nutrition, medication and stress; their influence to intestinal motility is another issue needs further research (Drossman, 1999, Holtmann and Enck, 1991).
Manometry and electromyography stand as the most common investigative methods that have been tested for medical knowledge extraction and diagnosis of the related abnormalities (Chen et al., 1993, Kellow et al., 1990, Sharna, 1989, Vantrappen et al., 1986, Weisbrodt, 1987). Their utilization has been mainly restricted to research studies, since they exhibit certain disadvantages to be used in clinical practice. Intraluminal manometry is painful and inconvenient to apply on human subjects, due to its invasiveness. Nevertheless, it is still capable of “clean”, multi-site pressure recordings. Operative electromyography seems to feature similar advantages and disadvantages. As a result, research efforts have been concentrated on non-invasive methods like cutaneous electromyography (Garcia-Casado, Martinez-de-Juan, & Ponce, 2005), as well as recording and analysis of gastrointestinal sounds (Craine et al., 1999, Craine et al., 2001, Craine et al., 2002, Dimoulas et al., 1999, Hadjileontiadis et al., 2000, Hadjileontiadis, 2005a, Hadjileontiadis, 2005b, Pastiadis et al., 1998, Tomomasa et al., 1999, Dimoulas, 1997, Campbell et al., 1989).
Bowel sound (BS) auscultation was proposed as an alternative medical study approach, in order to overcome the previous stated obstacles. There are direct relationships between manometry, electromyography and abdominal surface vibration (ASV) sensing, in how they monitor intestinal motility. All methods measure the effects of muscular convulsions, which produce pressure alterations over time and electrical myogenic activity. Implementation of BS monitoring has many advantages over traditional diagnosis methods (including X-ray screening approaches), because it is easier to apply, non-invasive, painless, does not cause discomfort to subjects, its influence to other psycho-physiology issues is limited, and can be applied for prolonged periods. However, knowledge and interpretation of BS has advanced little since Cannon’s pioneering work (Cannon, 1905), so that there is a lack of a reliable and accurate method for use in clinical practice. Most researchers seem to agree that this is not due to redundancy of diagnostic information of BS, but due to insufficient scientific support (Hadjileontiadis et al., 2000).
Section snippets
Problem definition
Based on the results provided by manometric research studies (Kellow et al., 1990, Sharna, 1989, Vantrappen et al., 1986, Weisbrodt, 1987) intestinal contractile activity is repeated periodically in cycles of silence periods, regular (in energy and repetitiveness) and irregular contractions. These initial foundations are related to the motor migrating complex (MMC cycle), and characterize the “fasting state”, where the stomach quietens, while the intestinal contractile activity is appeared in
Implementation
The current work constitutes the first stage of a hierarchical expert system, which aims to deliver BS pattern analysis. Thus, the main target within this project is to ensure rejection of the interference noise from the AS recordings, and to apply pattern classification according to the ASCC dictionary. Neural networks and supervised training techniques were chosen as the most applicable solution for their robustness and simplicity in pattern recognition (Hush & Horne, 1993), especially when
Experimental results and discussion
In the following paragraphs we will present training results of the implemented NNET topologies using the performance descriptors that previously described. Although many experiments were tried using various input vectors’ combinations, training parameters and network sizes, we will focus our attention to the configurations presented in the previous paragraphs, which they have provided the highest performance ratings. Thus, NNETs with (a) fs1, fs2 and fs3 input vectors (Eq. (24)), (b) 10, 15
Summary
The current work deals with the implementation of autonomous intestinal motility analysis system, for long-term unsupervised treatment in BS. AIMAS, which is a part of a broader intestinal motility auscultation medical diagnosis project, has been implemented in various versions of feed-forward multi-layer neural networks. The implemented wavelet neural network version, exhibits the best rating of all modules, while provides flexibility, simplicity and reduced computational cost. Experimental
References (64)
- et al.
An investigation of neuro-fuzzy systems in psychosomatic disorders
Expert Systems with Applications
(2005) - et al.
Intelligent target recognition based on wavelet packet neural network
Expert Systems with Applications
(2005) Automatic identification of sound source position employing neural networks and rough sets
Pattern Recognition Letters
(2003)- et al.
Computer analysis of BS
Computers in Biology and Medicine
(1975) - et al.
Classification methodologies of multilayer perceptrons with sigmoid activation functions
Pattern Recognition
(2005) - et al.
Feature saliency using signal-to-noise ratios in automated diagnostic systems developed for ECG beats
Expert Systems with Applications
(2005) - et al.
Recurrent neural networks employing Lyapunov exponents for EEG signals classification
Expert Systems with Applications
(2005) - et al.
Comparison of multilayer perceptron training algorithms for portal venous doppler signals in the cirrhosis disease
Expert Systems with Applications
(2006) - et al.
A cardiac sound characteristic waveform method for in-home heart disorder monitoring with electric stethoscope
Expert Systems with Applications
(2006) - et al.
Prolonged ambulant recordings of small motility demonstrate abnormalities in the irritable bowel syndrome
Gastroenterology
(1990)