Predictive modeling with neuroimaging data has the potential to characterize and diagnose psychiatric disorders. As our knowledge of mental health and machine learning continue to evolve, several topics have emerged related to both machine learning and psychiatry that require consideration. Here, I will provide a perspective on the current and future applications of predictive models towards understanding brain correlates of psychiatric disorders. I will begin by identifying two emerging topics from machine learning with direct implications for predictive models: 1) bias, fairness, and trustworthiness in machine learning algorithms and 2) dirty data. Next, I will highlight that the level of interpretation is crucial towards research design and depends on the goal of the investigation. Finally, I will demonstrate a role for predictive models in two emerging topics in psychiatry: transdiagnostic research and targeting imaging based brain markers. By highlighting and considering these emerging issues, I aim to accelerate the utility and applications of neuroimaging-based predictive models for psychiatric research.

 

 

 

 

 

 

 

Dustin Scheinost
Assistant professor of Radiology & Biomedical Imaging
Biomedical Engineering, Statistics & Data Science and Child Study Center, Yale School of Medicine

Dustin Scheinost – An eye towards the future of neuroimaging predictive models in psychiatry