Behavior quantification as the missing link between fields: Tools for digital psychiatry and their role in the future of neurobiology

Abstract

The great behavioral heterogeneity observed between individuals with the same psychiatric disorder and even within one individual over time complicates both clinical practice and biomedical research. However, modern technologies are an exciting opportunity to improve behavioral characterization. Existing psychiatry methods that are qualitative or unscalable, such as patient surveys or clinical interviews, can now be collected at a greater capacity and analyzed to produce new quantitative measures. Furthermore, recent capabilities for continuous collection of passive sensor streams, such as phone GPS or smartwatch accelerometer, open avenues of novel questioning that were previously entirely unrealistic. Their temporally dense nature enables a cohesive study of real-time neural and behavioral signals.

To develop comprehensive neurobiological models of psychiatric disease, it will be critical to first develop strong methods for behavioral quantification. There is huge potential in what can theoretically be captured by current technologies, but this in itself presents a large computational challenge -- one that will necessitate new data processing tools, new machine learning techniques, and ultimately a shift in how interdisciplinary work is conducted. In my thesis, I detail research projects that take different perspectives on digital psychiatry, subsequently tying ideas together with a concluding discussion on the future of the field.

Topics covered include:

Arguments for daily free-form audio journals as an underappreciated psychiatry datatype, with example scientific uses and technical validation results for a new pipeline to extract both acoustic and linguistic features from these journals.

A guide to collection of clinical interview recordings in a large, multi-site study, documenting lessons learned from the ongoing AMPSCZ project.

Results on the relationship between cognitive disorganization in psychosis and linguistic disfluency use in clinical interviews.

Results from a multimodal digital psychiatry dataset collected during a deep brain stimulation trial, showing detection of salient behaviors that would have otherwise gone unnoticed -- a first of its kind blueprint.

Novel stability theorems for multi-area recurrent neural networks (RNNs), applied to improve RNN performance on sequential classification benchmarks; stability is important for interpretability and safety of RNNs in sensitive use cases like medicine, while multi-area networks are an obvious next step for multimodal machine learning.