Research uses machine learning to distinguish threats
Nearly 100 years after Pavlov first proposed conditioned threat associations, researchers from McGovern Medical School are using machine learning to understand how the brain detects and responds to threats, paving the way to new treatments for fear-based and anxiety disorders.
Recently published in Nature Communications, “Distributed neural representations of conditioned threat in the human brain” helps to explain how the brain recognizes and responds to threat at a regional level. Led by Zhenfu Wen, PhD, assistant professor of psychiatry and first author, and Mohammed Milad, PhD, professor of psychiatry, John S. Dunn Endowed Chair, and senior author, the research includes collaborators from New York University, Harvard University, Massachusetts General Hospital, and Sweden.
Combining functional MRI data across varying threat conditioning and negative affect paradigms from 1,465 participants, researchers created a computational decoder, which distinguishes threat from safety in the human brain.
Twenty-four regions of participants’ brains were monitored as they viewed neutral pictures (pictures of a blue or yellow light) that were paired with mild electrical stimulation to their fingers. The study team used machine-learning methods to examine patterns of activations within those brain areas while people watched these pictures as they became associated to threat (the mild shock).
“In this work, we employ multivariate pattern analysis to investigate the neural representations of conditioned threat throughout conditioning, extinction learning, and extinction memory recall using two analytic approaches,” the researchers wrote. “In the first approach, we examine fMRI-based neural patterns within the ‘threat circuit’ in decoding stimuli that have been conditioned to signal threat from those that have been conditioned to signal safety. In the second approach, we examine neural patterns beyond the ‘threat circuit’ in contributing to conditioned threat and safety representations.”
Past research within this field has intensively focused on the roles of a few subcortical and cortical structures in how the association between the conditioned and the unconditioned stimuli is formed, and how the defensive responses are generated and subsequently extinguished. This approach has led to the notion of the ‘threat circuit’ that includes subregions of the amygdala, periaqueductal gray, hippocampus, medial prefrontal cortex, and insular cortex.
“Our results point to the important contribution of multiple sensory and cognitive neural nodes to the decoding of stimuli associated with threat detection and responding,” Wen said. “While the threat circuit plays an important role in the detection and response to threat, there is a need for studying how this circuit interacts with other cognitive and sensory neural networks to accomplish the complex process during the encounter of threat.”
The results from this study revealed a neural signature of threat detection, a finding that was validated across nine data samples in the United States, Europe, and China. This research supports the integration of broad networks required for the implicit and explicit threat processing. The findings of this research represent a significant advance in the field of neuroscience by providing a ‘fingerprint’ for the where and how the brain detects threatening events or stimuli. The results of this study may provide new brain targets that can be manipulated to alleviate excessive anxiety and fear in patients suffering from fear-based and anxiety disorders.