The energy metabolic footprint of predictive processing in the human brain

Abstract

AbstractNeural activity is a highly energy-intensive process. In the human brain, signaling consumes up to 75% of the available energy resources with postsynaptic potentials as the largest factor. Visual processing is especially costly, with increases in energy consumption of up to 20% in the visual cortex. In recent years, vision has been cast as a constructive process, harnessing prior knowledge in a constant feedback loop of top-down prediction and bottom-up sensory input. Interestingly, sensory input that is in line with our predictions might be processed at lower energy metabolic cost. However, there is no evidence for this claim yet, possibly due to the scarcity of measures that quantify energy consumption in the human brain.Here, we used a novel MR method measuring the cerebral metabolic rate of oxygen during sensory stimulation of visual sequences that varied in their predictability. Since predictive processing is driven by estimates of uncertainty, we assessed how confident subjects were in their knowledge of the underlying patterns. We found that processing predictable sequences steeply decreased in energetic cost with increasing confidence. Strikingly, these energetic effects were not limited to visual areas, summing up to a cortical difference of 13% between high and low levels of confidence. Furthermore, sequences deviating from expectations were energetically cheaper than predictable ones for low confidence levels, but costlier for high levels. These results speak for a major role of predictive processing in balancing the brain’s energy budget and emphasize the impact of interindividual differences when learning predictive patterns.