Big Data and Neuroimaging

Abstract

Big Data are of increasing importance in a variety of areas, especially in the biosciences. There is an emerging critical need for Big Data tools and methods, because of the potential impact of advancements in these areas. Importantly, statisticians and statistical thinking have a major role to play in creating meaningful progress in this arena. We would like to emphasize this point in this special issue, as it highlights both the dramatic need for statistical input for Big Data analysis and for a greater number of statisticians working on Big Data problems. We use the field of statistical neuroimaging to demonstrate these points. As such, this paper covers several applications and novel methodological developments of Big Data tools applied to neuroimaging data.

Functional neuroimaging in the era of Big Data and Open Science: A modern overview

“…Each of these dimensions could conceivably grow in the future, as technology improves (spatial and temporal aspects of the image data), or as a result of open data‐sharing practices (number of subjects). Hence, for example, a method that relies on bootstrap resampling will need to retain computational efficiency over thousands of subjects, for each of which a large amount of complex high‐dimensional data are available (Webb‐Vargas et al, 2017). Divide and combine approaches (Chen & Xie, 2014), that use parallel computing, are poised to play a bigger role in this Big Data neuroimaging context.…”

“…Serious statistical research on positron emission tomography (PET) and fMRI started in the early to mid-1990s, when computing resources were much more limited than they are today, and even at that point touched on issues that are still central, for example control of familywise error rate (Forman et al, 1995) in the presence of hundreds of thousands of simultaneous tests. As noted by Webb-Vargas et al (2017), neuroimaging data in all their variations have the hallmarks of Big Data. Complex data structure (spatial and temporal correlation within a scan, across scans, across sessions, and across sites) and ever-increasing size (as resolution improves, but also due to the rise of public repositories with multiple types of data collected on thousands of individuals) necessitate the development of scalable statistical analysis techniques and the exploitation of modern solutions such as GPUs and parallel computing.…”

Hyperbolic trade-off: the importance of balancing trial and subject sample sizes in neuroimaging

“…In our modeling efforts, we compare the summary statistics approach of condition-level modeling (CLM) directly to a hierarchical framework of trial-level modeling (TLM) that explicitly takes cross-trial variability into consideration at the population level to examine the impact of cross-trial variability. We aim to provide a fresh perspective for experimental designs, and make a contribution to the discussion of 'big data' versus 'deep scanning' (Webb-Vargas et al, 2017;Gordon et al, 2017). trial-level effect estimates for the cth condition of the sth subject in the tth trial; input data at the population level y cs condition-level effects for the cth condition of the sth subject, estimated through time series regression at the subject level y cs condition-level effects for the cth condition of the sth subject, estimated through averaging across T trials at the subject level σ τ within-subject cross-trial standard deviation (referred to as "cross-trial variability") σ π cross-subject cross-trial standard deviation (referred to as "cross-subject variability") R v variability ratio: the ratio of within-subject cross-trial variability to cross-subject cross-trial variability; defined as σ τ /σ π ρ subject-level correlation between two conditions µ a contrast between two condition-level effects µ 1 and µ 2 ; defined as µ 2 − µ 1 σ standard error (or uncertainty) of a contrast estimate µ; here, the statistical efficiency or precision of the contrast estimate at the population level is denoted by σ −1 Population Subject (S)…”

Teachers’ Views towards Changes in Curriculum and the Use of Real-Time Big Data in Statistics Education

“…With the occurrence of data dumping phenomenon, hence the emergence of research on the need of developing statistical thinking in understanding big data (Ashley Steel et al, 2019;González et al, 2020;Toledo et al, 2018;Hoerl et al, 2014) so to further enhance the quality of living. Big data is an important aspect in innovation that is gaining more attention in the current age (Baig et al, 2020;Webb-Vargas et al, 2017;Storey & Song, 2017) and has become the prime stream in most current research fields (Chen et al, 2020). In the current situation of statistics education in Malaysia, the use of real-time data is still loose especially those involving big data.…”

“…However, the use of big data in statistics education curriculum is a need to equip the younger generation in facing future challenges. The skill in interpreting big data has becoming more important and it needs a major role in statistical thinking in order to produce meaningful results (Webb-Vargas et al, 2017). This increasing need should be encouraging in changing statistics curriculum so that is capable in nurturing a current generation that are not only aware in statistics, but also data minded.…”