Visualization of High-Dimensional Clinically Acquired Geographic Data Using the Self-Organizing Maps

Abstract

The objective of this study is to visualize high-dimensional data vectors using popular data reduction algorithms. The study reports on the effectiveness and expressiveness of a set of data reduction algorithms in the visualization of geospatial data sets derived from clinical records of patients. The experiments show that when the SOM algorithm is combined with GIS methods together they are even more powerful tools for exploratory analysis than when each is applied separately. The visual approach provides a very useful exploration environment to support the formulation of new and better study hypotheses regarding the spatial distribution of a particular disease. While it was apparent that the spatial distribution and patterns of asthma were predominately located near the major roadways and the Peace Bridge Complex, obstructive sleep apnea is slightly more widespread even in the suburbs and surrounding neighborhoods. The spatial patterns discovered between the original features of adult and childhood asthma are consistent with the SOM-trained data, but a slight difference emerges for the SOM-trained obstructive sleep apnea data set. This study is successful at gaining significant novel insights into the spatial characteristics of patient data in relation to key environmental factors.

Keywords: self-organizing maps, geocomputations, disease, GIS, visualization, data exploration, clustering, pattern recognition