Insular Gamma Power is Sufficient to Classify Between Anticipation of Palatable Versus Neutral Items. (16th November 2020)

• Record Type: Journal Article
• Authors: Gattas, Sandra; Huang, Yuhao D; Kakusa, Bina; Feng, Austin Y; Barbosa, Daniel; Shivacharan, Rajat; Parker, Jonathon J; Kuijper, Fiene M; Bohon, Cara; Mikhail, Abanoub; Halpern, Casey H
• Is Part Of: Neurosurgery. Volume 67(2010)Supplement 1
• Publication Date: 2020-11-16
• Subjects: Nervous system -- Surgery -- Periodicals; 617.48005
• DOI: 10.1093/neuros/nyaa447_624
• Languages: English
• ISSNs: 0148-396X

Abstract

Abstract: INTRODUCTION: Anticipation periods are key for the therapeutic intervention of eating disorders. Prior findings point to the nucleus accumbens as a therapeutic target during this window. Yet, a more accessible cortical area is necessary for ease of extracranial diagnostic measures and intracranial surgical targeting. The insula is a candidate region, since it hosts gustatory cortex and supports cognitive processes mediating food choices. Using human intracranial recordings, we previously identified differences in insular gamma power between anticipation of palatable and tasteless items. Here, we investigate whether a predictive model can be generated to use insular gamma power alone to predict, on a single trial basis, whether an anticipatory period is to a palatable or tasteless item. Such a predictive model is a stepping-stone to developing non-chronic neuromodulation therapy and would inform insular coding capacity regarding feeding anticipatory periods. METHODS: Intracranial recordings from the human insula were acquired while patients (n = 8) engaged in a task where images were shown cueing the solution (palatable vs. tasteless) to be subsequently delivered for taste. An ensemble subspace KNN classifier was trained, with 5-fold cross validation, to classify between anticipation of palatable and tasteless solutions. Insular sites with significantly diminished gamma power during anticipation or consumption of palatable items were utilized. The feature vector consisted of gamma power in four anticipation time windows: 0-0.5, 0.5-1, 1–2, and 2–3 sec. PCA was used to keep enough components to explain 95% of variance. RESULTS: The predictive model performed with 76.1% accuracy in classifying between the two anticipatory periods (palatable vs. neutral) and with a mean TPR of 72.5% and FPR of 27.5% across the two classes. A classifier at the specified parameters relies significantly on each of the four features (p < 0.01), with significantly more reliance on feature 2 (0.5-1 sec cue period gamma power) compared to any other feature ( P = .002). CONCLUSION: The generated model holds translational potential for closed-loop neuromodulation therapy for eating-related disorders and isolates further the involvement of the insula in feeding anticipation.