Using Kohonen self organising feature maps for the analysis of ambulatory oesophageal manometry

Abstract

Gastrointestinal (GI) diseases, the third most common cause of deaths in the UK (59,685 in 2000) have increased by 25% in the last ten years. Gastro-oesophageal reflux disease (GORD) has a prevalence of 20-40% in the population. Ten to 15% of GORD patients have Barrett's oesophagus, a change in the cell lining of the oesophagus linked to oesophageal cancer. With a 5 year survival of 9% and annually 7000 new cases (12 per 100,000) oesophageal cancer has increased by 50% in the last 20 years. Barrett's oesophagus is associated with poor motility, reflux and hiatus hernia [2]. Oesophageal manometry is used to investigate oesophageal motility. Patterns of pressure changes are seen as peristaltic waves pass over transducers on a catheter within the oesophagus. The standard investigation has many problems, as symptoms rarely occur during the short investigation time period. Also, despite widespread use it is not clear what values and patterns are important. Ambulatory manometry, introduced in 1985 to address these issues, has proved unpopular due to the large amounts of data generated which has proved difficult to analyse. Manual analysis is very time consuming while current software analysis, using predefined rules to detect peaks and classify patterns, is not always trusted and has not advanced over the last 20 years. Our approach has been to use Kohonen self organising feature maps to classify the patterns of data that occur over 24 hours. Following a simple process of identifying candidate periods (feature vectors) this technique was used to automatically find patterns or clusters within the raw data. The early results have shown that patterns of oesophageal manometry data can be identified. Different patterns can be analysed and variation in occurrence rates detected during specific symptomatic periods. The results suggest that this approach may enable detection and help quantify differences in 24 hour manometry data between healthy controls and patients with Barrett's oesophagus.

The Application of Biomedical Engineering Techniques to the Diagnosis and Management of Tropical Diseases: A Review

“…Therefore, the SOM can be considered as an exploratory data analysis tool for generating hypotheses on the relationships among the data. The SOM has been widely used in medical applications [ 49 , 50 , 51 , 52 , 53 ]. Typical SOM can be visualized by using the U-matrix and the component planes.…”