Promises and pitfalls of deep neural networks in neuroimaging-based psychiatric research

TL;DR

This paper reviews the use of deep neural networks in neuroimaging-based psychiatric research, highlighting methodological advances, applications, and current challenges such as data limitations and bias.

Contribution

It provides a comprehensive overview of deep learning methods in psychiatric neuroimaging and discusses key challenges impacting their clinical utility.

Findings

Deep neural networks enable improved psychiatric diagnosis and biomarker development.

Challenges include data heterogeneity, label validity, and algorithmic bias.

Methodological concepts like transfer learning are promising but face practical limitations.

Abstract

By promising more accurate diagnostics and individual treatment recommendations, deep neural networks and in particular convolutional neural networks have advanced to a powerful tool in medical imaging. Here, we first give an introduction into methodological key concepts and resulting methodological promises including representation and transfer learning, as well as modelling domain-specific priors. After reviewing recent applications within neuroimaging-based psychiatric research, such as the diagnosis of psychiatric diseases, delineation of disease subtypes, normative modeling, and the development of neuroimaging biomarkers, we discuss current challenges. This includes for example the difficulty of training models on small, heterogeneous and biased data sets, the lack of validity of clinical labels, algorithmic bias, and the influence of confounding variables.