WIP + Methods Series

Abstract: Building on cortical semantic maps (Huth et al. 2016, Nature) and the recent decoding of connected speech from 3T fMRI (Tang et al. 2023, Nature Neuroscience), we propose to investigate lexical semantic representations at the level of cortical layers using 7T fMRI. Whereas prior work used voxels of 2.6 mm, ultra-high-field fMRI will allow us to approach 1 mm resolution and thus separate 2-3 cortical layers, given an average cortical thickness of 2.4 mm (Fischl and Dale 2000, PNAS). Our aim is to move semantic decoding from gross anatomy toward the mesoscopic scale of cortical columns and canonical cortical circuitry (e.g. Bastos et al. 2012, Neuron). Pilot work suggests that this will require restricting the field of view; for example, focusing on the frontal half of the brain would provide access to the frontal cortex and anterior temporal lobe - two regions central to semantic processing. Recent evidence for layer-specific sound representations in auditory cortex (Leonard et al. 2023, Nature) provides a concrete precedent for asking whether semantic representations during speech are similarly organised across cortical depth. This work forms part of a larger effort to understand the neurobiology of speech at a level that can support the next generation of non-invasive brain-to-text decoding methods.

We then present a harmonisation diagnostics tool (DHARM: Diagnose Harmonisation) designed to support the systematic evaluation of variability arising from acquisition differences within and across datasets and to assess the performance of harmonisation strategies. The tool provides a comprehensive set of metrics that capture both biological and technical sources of variability, enabling robust assessment of site and scanner effects in both raw and harmonised data across cross-sectional and longitudinal settings. Through short demonstrations, we illustrate how these diagnostics can guide method selection and improve confidence in downstream analyses.

 

We conclude by discussing practical considerations, limitations, and opportunities for community input, including how the tool could support your multisite research and directions for future development.