Allgaier, N. | bioRxiv Science

bioRxiv2021

Performance Scaling for Structural MRI Surface Parcellations: A Machine Learning Analysis in the ABCD Study

The use of pre-defined parcellations on surface-based representations of the brain as a method for data reduction is common across neuroimaging studies. In particular, prediction-based studies typically employ parcellation-driven summaries of brain measures as input to predictive algorithms, but the choice of parcellation and its influence on performance is often ignored. Here we employed pre-processed structural magnetic resonance imaging data (sMRI) from the ABCD Study(R) to examine the relationship between 220 parcellations and out-of-sample predictive performance across 45 phenotypic measures in a large sample of 9-10-year-old children (N=9,432). Choice of Machine Learning (ML) pipeline and use of alternative multiple parcellation-based strategies were also assessed. Relative parcellation performance was dependent on the spatial resolution of the parcellation, with larger numbers of parcels (up to [~]4000) outperforming coarser parcellations, according to a power-law scaling of between 1/4 and 1/3. Performance was further influenced by the type of parcellation, ML pipeline, and general strategy, with existing literature-based parcellations, a support vector based pipeline, and ensembling across multiple parcellations, respectively, as the highest performing. These findings highlight the choice of parcellation as an important influence on downstream predictive performance, showing in some cases that switching to a higher resolution parcellation can yield a relatively large boost to performance.

neuroscience↗

bioRxiv2020

The ABCD Stop Signal Data: Response to Bissett et al.

This paper responds to a recent critique by Bissett and colleagues (Bissett et al., eLife, In Press) of the fMRI Stop task being used in the Adolescent Brain Cognitive DevelopmentSM Study (ABCD Study(R)). The critique focuses primarily on a design feature of the task that the authors contend lead to a violation of race model assumptions (i.e., that the Go and Stop processes are fully independent) which are relevant to the calculation of the Stop Signal Reaction Time, a measure of the inhibition process. Bissett and colleagues also raise a number of secondary concerns. In this response we note that satisfying race model assumptions is a pernicious challenge for Stop task designs but also that the race model is quite robust against violations of its assumptions. Most importantly, while Bissett et al. raise conceptual concerns with the task we focus here on analyses of both the performance and the neuroimaging data and we conclude that the concerns appear to have minimal impact on the neuroimaging data (the validity of which do not rely on race model assumptions) and have far less of an impact on the performance data than the critique suggests. We note that Bissett et al. did not apply any performance-based exclusions to the data they analyzed, that a number of the trial coding errors that they flagged were already identified and corrected in the ABCD annual data releases, that a number of the secondary concerns reflect sensible design decisions and, indeed, that their own computational modeling of the ABCD Stop task suggests the problems they identify have just a modest impact on the rank ordering of individual differences in subject performance. In this paper, we list some adjustments that have been made to the task and some new flags that are now added to the annual, curated data releases. We stress that the ABCD data are fully available to the scientific community who are empowered to apply whatever inclusion and exclusion criteria they deem appropriate for their analyses and we conclude that the ABCD Stop task yields valuable data that researchers can use to track adolescent neurodevelopment.

neuroscience↗