Dissertation Title:  Wearable Gut and Brain Interfaces for Modulating Good and Bad Emotions

Abstract:

Bad emotions are stronger than good emotions. Negative emotions trigger our survival mechanisms that lead to stronger, faster, and longer-lasting impacts. Emotions are intertwined with anxiety and depression symptoms, which now affect up to 32.3% of U.S. adults. Mainline treatments such as therapy and medication are limited by long-term failure rates, lack of access, and adverse side-effects. 

Emotion detection interfaces have shown promise in mediating our emotional health through improved diagnosis, self-tracking, social support systems, mindfulness, and biofeedback training, however, they are limited by their current capacities. The most popular methods falter when distinguishing between “good” and “bad” - or, lead to privacy or social issues. Brain and gut interfaces can serve as an alternative, but often require complex setups with many electrodes, large datasets, and the usage of significant training to achieve benchmark emotion detection performance. 

I present novel, wearable gut- and brain-interfaces for "good" and "bad" emotions that can be made feasible with as few as two electrodes, minimal training and statistical analysis. I coin and define the area of gut-brain computer interfacing (GBCI), while further developing the field of affective brain-computer interfacing (aBCI). I take a novel approach by using the stomach signal and motivational direction models as an alternative to traditional affective modalities and models. I evaluate my approach using video games, machine and deep learning models, and replication studies. 

I present Joie, a joy-based electroencephalography (EEG) brain-computer interface (BCI). I present JoyNet, a neural network for joy detection with EEG; and KALM, for EEG, electrodermal activity (EDA) and breathing rate multimodal fusion. I also present Serosa, an electrogastrography (EGG) GBCI which non-invasively records indices of stomach neurons and provides real-time feedback on gut-brain interactions tied to levels of stress. I also share arts collaborations with these projects.

This thesis presents findings and innovations in research and application: first, offline affect detection models to contextualize gut (gastric) and brain (cortical) activity with the more common methods (electrodermal activity, heart and breathing-rate), and evaluate how each signal influences affect detection performance. Second, novel real-time interfaces are implemented and evaluated with placebo-controlled laboratory studies. They evaluate neurofeedback efficacy, workload and performance, user needs and perspectives, and potential applications for neural emotion modulation. The findings are interpreted through the lens of neural and embodied emotion theory. 

Committee:

Pattie Maes
Germeshausen Professor of Media Arts and Sciences
Director of Fluid Interfaces Research Group
MIT Media Lab

Rosalind Picard 
Professor of Media Arts and Sciences
Director of the Affective Computing Research Group
MIT Media Lab

Gregory Abowd
Dean of the College of Engineering
Professor of Electrical and Computer Engineering
Northeastern University

Brendan Allison 
Author and Visiting Scholar
Department of Cognitive Science
University of California San Diego