A Machine Learning Approach Towards the Differentiation Between Interoceptive and Exteroceptive Attention

Interoception, the representation of the bodys internal state, plays a central role in emotion, motivation, and wellbeing. Interoceptive attention is qualitatively different from attention to the external senses and may recruit a distinct neural system, but the neural separability of interoceptive and exteroceptive attention is unclear. We used a machine learning approach to classify neural correlates of interoceptive and exteroceptive attention in a randomized control trial of interoceptive training (MABT). Participants in the training and control groups attended fMRI assessment before and after an 8-week intervention period (N = 44 scans). The imaging paradigm manipulated attention targets (breath vs. visual stimulus) and reporting demands (active reporting vs. passive monitoring). Machine learning models achieved high accuracy in distinguishing between interoceptive and exteroceptive attention using both in-sample and more stringent out-of-sample tests. We then explored the potential of these classifiers in "reading out" mental states in a sustained interoceptive attention task. Participants were classified as maintaining an active reporting state for only [~]90s of each 3-minute sustained attention period. Within this active period, interoceptive training enhanced participants ability to sustain interoceptive attention. These findings demonstrate that interoceptive and exteroceptive attention engage reliable and distinct neural networks; machine learning classifiers trained on this distinction show promise for assessing the stability of interoceptive attention, with implications for the future assessment of mental health and treatment response.