Using extreme value statistics to reconceptualize psychopathology as extreme deviations from a normative reference model

Charlotte J. Fraza,Mariam Zabihi,Christian F. Beckmann,Andre F. Marquand

DOI: https://doi.org/10.1101/2022.08.23.505049

2024-11-07

Abstract:For many problems in neuroimaging, the most informative features occur in the tail of the distribution. For example, when considering psychiatric disorders as deviations from a 'norm', the tails of the distribution are considerably more informative than the bulk of the distribution for understanding risk, for stratifying and predicting such disorders and for anomaly detection. Yet, most statistical methods used in neuroimaging focus on modelling the bulk and fail to adequately capture extreme values occurring in the tails. To address this, we propose a framework that combines normative models with multivariate extreme value statistics to accurately model extreme deviations of a reference cohort for individual participants. Normative models are now widely used in clinical neuroscience and similar to the employment of normative growth charts in pediatric medicine to track a child's weight in relation to their age, normative models can be used with neuroimaging measurements to quantify individual neurophenotypic deviations from a reference cohort. However, formal statistical treatment of how to model the extreme deviations from these models has been lacking until now. In this article we provide such an approach, inspired by applications of extreme value statistics in meteorology. Since the presentation of extreme value statistics is usually quite technical, we begin with a non-technical introduction to the fundamental principles of extreme value statistics, to accurately map the tails of the normative distribution for biological markers, including mapping multivariate tail dependence across multiple markers. Next, we give a demonstration of this approach to the UK Biobank dataset and demonstrate how extreme values can be used to accurately estimate risk and detect atypicality. This framework provides a valuable tool for the statistical modelling of extreme deviations in neurobiological data, which could provide us with more accurate and effective diagnostic tools for neurological and psychiatric disorders.

Neuroscience

What problem does this paper attempt to address?

The problem that this paper attempts to solve is that in neuroimaging, traditional statistical methods mainly focus on the main part of the data distribution and ignore extreme values (i.e., data points distributed in the tail). However, in many neuroscience problems, especially when considering mental disorders, these extreme values can provide more valuable information than the main part of the distribution for understanding risks, stratified prediction of these disorders, and anomaly detection. Therefore, the paper proposes a methodological framework that combines the normative model and multivariate extreme value statistics to accurately model extreme deviations in individual participants in the reference cohort.

Specifically, the main objectives of the paper include:
1. Provide a basic instructional introduction to extreme value statistics and give implementation guidelines, as extreme value statistics is a highly technical field.
2. Demonstrate the application of this method through empirical research on neuroimaging data using multivariate extreme value statistics.
3. Establish clear pairwise tail - dependence relationships and reveal the extreme - dependence structures between different brain phenotypes.
4. Use extreme - value - based principal component analysis (PCA), namely "extreme PCA", to summarize this tail - dependence relationship.
5. Link these extreme principal components to multiple behavioral phenotypes and explore how extreme neurobiological data affect behavior.

Through these objectives, the paper aims to provide a valuable tool for statistical modeling of extreme deviations in neurobiological data, which may provide more accurate and effective diagnostic tools for neurological and mental disorders.