Unbiased Analysis of Item-Specific Multi-Voxel Activation Patterns Across Learning

Abstract

Recent work has highlighted that multi-voxel pattern analysis (MVPA) can be severely biased when BOLD response estimation involves systematic imbalance in model regressor correlations. This problem occurs in situations where trial types of interest are temporally dependent and the associated BOLD activity overlaps. For example, in learning paradigms early and late learning stage trials are inherently ordered. It has been shown empirically that MVPAs assessing consecutive learning stages can be substantially biased especially when stages are closely spaced. Here, we propose a simple technique that ensures zero bias in item-specific multi-voxel activation patterns for consecutive learning stages with stage being defined by the incremental number of individual item occurrences. For the simpler problem, when MVPA is computed irrespective of learning stage over all item occurrences within a trial sequence, our results confirm that a sufficiently large, randomly selected subset of all possible trial sequence permutations ensures convergence to zero bias – but only when different trial sequences are generated for different subjects. However, this does not help to solve the harder problem to obtain bias-free results for learning-related activation patterns regarding consecutive learning stages. Randomization over all item occurrences fails to ensure zero bias when the full trial sequence is retrospectively divided into item occurrences confined to early and late learning stages. To ensure bias-free MVPA of consecutive learning stages, trial-sequence randomization needs to be done separately for each consecutive learning stage.

Neural representation of newly instructed rule identities during early implementation trials

“…Specifically, we sought to identify distributed neural activity patterns associated with subtle representational differences regarding newly instructed individual rule identities such as ‘if the word BUTTER is displayed on the screen, then flex the middle finger’ or ‘if the word MONKEY is displayed on the screen, then flex the index finger’. To this end, we employed a recently developed multivariate pattern analysis technique (MVPA) specifically calibrated to uncover the rapidly evolving representational dynamics while implementing novel rule instructions for the first time (Ruge et al, 2018b). Importantly, this technique (see Materials and methods) ensured unbiased results by avoiding systematic imbalance in model regressor correlations through appropriate stimulus sequence construction (cf., Mumford et al, 2014; Visser et al, 2016).…”

Initial neural representation of individual rules after first-time instruction

“…Three additional subjects could not be used due to incomplete data collection. Part of the present dataset was used in a previous methods-oriented paper 15 . The sample for experiment 2 consisted of 70 human participants (39 female, 31 male; mean age: 23.9 years, range 19-33 years).…”

“…Finally, for each subject, the resulting mean difference values were averaged across task blocks separately for each learning stage before being submitted to group-level statistical evaluation. Based on previous work, trial sequences were constructed in a way that ensured unbiased multivariate results under conditions of overlapping single-trial BOLD responses within task blocks 15,64,69 . Specifically, we conducted unbiased identity-specific MVPA separately for each successive learning stage within each task block.…”

“…To this end, for each task block the overall trial sequence was composed of 2 (experiment 1) or 4 (experiment 2) independently generated 'atomic' 8-trial sequences, each comprising 2 randomly distributed occurrences of each of the 4 nouns. On average across such atomic sequences, this approach guaranties unbiased MVPA due to the circumstance that non-zero bias regarding individual atomic sequences is distributed around zero mean 15 . We furthermore took advantage of multiple novel task blocks per participant which allowed us to regress out bias-induced variance across blocks and thereby to obtain more robust results.…”

“…Specifically, we sought to identify distributed neural activity patterns associated with subtle representational differences regarding newly instructed individual rule identities such as 'if the word BUTTER is displayed on the screen, then flex the middle finger' or 'if the word MONKEY is displayed on the screen, then flex the index finger'. To this end, we employed a newly developed multivariate pattern analysis technique (MVPA) specifically calibrated to uncover the rapidly evolving representational dynamics following novel rule instructions 15 .…”