Detection of Smoothly Distributed Spatial Outliers with Applications to Identifying the Pulmonary Distribution of Parenchymal Hyperresponsiveness

摘要

Methacholine challenge tests are used to measure changes in pulmonary function that indicate symptoms of asthma. In addition to pulmonary function tests, which measure global changes in pulmonary function, computed tomography (CT) images taken before and after administration of methacholine provide local air volume changes at individual acinar units. Some of the acini may have extreme air volume changes relative to the global average, indicating hyperresponsiveness, and those extreme values may occur in clusters. We propose a Gaussian mixture model with a spatial smoothness penalty to improve prediction of hyperresponsive locations that occur in spatial clusters. A simulation study provides evidence that the spatial smoothness penalty improves prediction under different data-generating mechanisms. We apply this method to CT data from Seoul National University Hospital on five healthy and ten asthmatic subjects.