Decoding Respiratory Signals from LFP in Olfactory Bulb based on ConvLSTM Framework

Abstract:
Respiratory signals are crucial for decoding olfactory neural circuits and understanding respiration-entrained local field potential (LFP) rhythms. Acquiring high-precision respiratory signals often requires additional equipment, increasing the experimental burden, particularly in free-moving paradigms. However, within the current neuroscience paradigm, we have not yet found a non-invasive method for obtaining the respiratory signals of laboratory animals. To minimize surgical trauma to experimental animals, this paper introduces a ConvLSTM-based decoding method, which uses the LFP from the olfactory bulb (OB) region of animals to accurately decode respiratory signals. After validating the effectiveness of the ConvLSTM method through comparative experiments, we further investigated the factors influencing the model's decoding performance. The Γ frequency band exhibited the best decoding results, with a mean absolute error (MAE) of 0.16±0.09, which may be related to the causal direction from respiratory signals to Γ rhythms. Additionally, the ConvLSTM model demonstrated strong trend prediction capabilities in multi-sequence point predictions and forward predictions, despite a slight decline in single-point prediction accuracy. Extensive validation on publicly available datasets indicates that the ConvLSTM-based respiratory decoding approach is highly applicable, offering a non-invasive alternative for respiratory signal acquisition in animal studies.