MICRO-MACRO MULTILEVEL ANALYSIS FOR MORPHOMETRIC DATA OF COMPUTED TOMOGRAPHY IMAGES: TWO ALTERNATIVE APPROACHES

Abstract

The computed tomography images data posse an intrinsic hierarchical structure; that is, when images (lower-level or micro-level units) are nested within individuals (higher-level or macro-level units). The problem with two-level data, where it is expected that a dependent variable measured on the individuals is influenced by variables measured on the images is the focus of this article. Methodology for the analysis of this type of scenario, known as micro-macro situation, has been little investigated. In this article, a latent variable modeling approach is proposed, where morphometric information of multiple images and characteristics of the patients, such as age and sex, are combined in a single logit model. Multilevel Factor Analysis is used to obtain latent variables at the macro level, which can be interpreted as synthetic indicators of morphometric differences in images between individuals. To illustrate the methodology, real data of patients with cerebral hemorrhage were used. The results are compared with those obtained by an alternative approach. The empirical study revealed a comparable performance between the two methodologies, although the new approach suggests a relative superiority in terms of predictive capacity.