Inverse transformed encoding models – a solution to the problem of correlated trial-by-trial parameter estimates in fMRI decoding

A tightly controlled fMRI dataset for receptive field mapping in human visual cortex

Four right-handed, healthy subjects participated in a visual stimulation experiment. Subjects were viewing a dartboard-shaped flickering checkerboard stimulus, divided into 4 rings and 12 segments, defining 48 sectors in the visual field. Local contrast in each sector was continuously varying across four levels and updated every 3 s. To maintain fixation, subjects had to respond to a stimulus at the center of the visual field. During the entire experiment, in which subjects performed 8 runs, each consisting of 100 trials, brain activity was measured with functional magnetic resonance imaging (MRI). Using a 3-T Siemens Trio MRI scanner, 220 echo-planar images were acquired in each run, with a repetition time of 1.5 s and voxel size of 3 x 3 x 3 mm. The dataset is publicly available from OpenNeuro and additionally includes region of interest maps for visual areas V1 to V4, left and right, obtained from another retinotopic mapping experiment. As such, the dataset allows for accurate mapping of receptive fields and their properties across several stages of human visual cortex.

Improving the accuracy of single-trial fMRI response estimates using GLMsingle

Advances in artificial intelligence have inspired a paradigm shift in human neuroscience, yielding large-scale functional magnetic resonance imaging (fMRI) datasets that provide high-resolution brain responses to thousands of naturalistic visual stimuli. Because such experiments necessarily involve brief stimulus durations and few repetitions of each stimulus, achieving sufficient signal-to-noise ratio can be a major challenge. We address this challenge by introducing GLMsingle, a scalable, user-friendly toolbox available in MATLAB and Python that enables accurate estimation of single-trial fMRI responses (glmsingle.org). Requiring only fMRI time-series data and a design matrix as inputs, GLMsingle integrates three techniques for improving the accuracy of trial-wise general linear model (GLM) beta estimates. First, for each voxel, a custom hemodynamic response function (HRF) is identified from a library of candidate functions. Second, cross-validation is used to derive a set of noise regressors from voxels unrelated to the experiment. Third, to improve the stability of beta estimates for closely spaced trials, betas are regularized on a voxel-wise basis using ridge regression. Applying GLMsingle to the Natural Scenes Dataset and BOLD5000, we find that GLMsingle substantially improves the reliability of beta estimates across visually-responsive cortex in all subjects. Comparable improvements in reliability are also observed in a smaller-scale auditory dataset from the StudyForrest experiment. These improvements translate into tangible benefits for higher-level analyses relevant to systems and cognitive neuroscience. We demonstrate that GLMsingle: (i) helps decorrelate response estimates between trials nearby in time; (ii) enhances representational similarity between subjects within and across datasets; and (iii) boosts one-versus-many decoding of visual stimuli. GLMsingle is a publicly available tool that can significantly improve the quality of past, present, and future neuroimaging datasets sampling brain activity across many experimental conditions.

Bayesian model selection favors parametric over categorical fMRI subsequent memory models in young and older adults

Subsequent memory paradigms allow to identify neural correlates of successful encoding by separating brain responses as a function of memory performance during later retrieval. In functional magnetic resonance imaging (fMRI), the paradigm typically elicits activations of medial temporal lobe, prefrontal and parietal cortical structures in young, healthy participants. This categorical approach is, however, limited by insufficient memory performance in older and particularly memory-impaired individuals. A parametric modulation of encoding-related activations with memory confidence could overcome this limitation. Here, we applied cross-validated Bayesian model selection (cvBMS) for first-level fMRI models to a visual subsequent memory paradigm in young (18-35 years) and older (51-80 years) adults. Nested cvBMS revealed that parametric models, especially with non-linear transformations of memory confidence ratings, outperformed categorical models in explaining the fMRI signal variance during encoding. We thereby provide a framework for improving the modeling of encoding-related activations and for applying subsequent memory paradigms to memory-impaired individuals.