Regression‐based machine‐learning approaches to predict task activation using resting‐state fMRI

Abstract

Resting-state fMRI has shown the ability to predict task activation on an individual basis by using a general linear model (GLM) to map resting-state network features to activation z-scores. The question remains whether the relatively simplistic GLM is the best approach to accomplish this prediction. In this study, several regressionbased machine-learning approaches were compared, including GLMs, feed-forward neural networks, and random forest bootstrap aggregation (bagging). Resting-state and task data from 350 Human Connectome Project subjects were analyzed. First, the effect of the number of training subjects on the prediction accuracy was evaluated. In addition, the prediction accuracy and Dice coefficient were compared across models. Prediction accuracy increased with the training number up to 200 subjects; however, an elbow in the prediction curve occurred around 30-40 training subjects.All models performed well with correlation matrices, which displayed correlation between actual and predicted task activation for all subjects, exhibiting a strong diagonal trend for all tasks. Overall, the neural network and random forest bagging techniques outperformed the GLM. These approaches, however, require additional computing power and processing time. These results show that, while the GLM performs well, resting-state fMRI prediction of task activation could benefit from more complex machine learning approaches. K E Y W O R D SfMRI, machine learning, neural networks, random-forest bootstrap aggregation, resting state

Jumping over Baselines with New Methods to Predict Activation Maps from Resting-state fMRI

“…Motivated by recent progress in establishing a stronger link between spontaneous and task-evoked activity, we examine the problem of mapping rsfMRI measurements to patterns of activity elicited during tfMRI-based experimental paradigms in individual subjects. We show additional evidence that it is indeed possible to predict task activity maps from patterns of rsfMRI FC, as previously reported [11][12][13][14][15][16][17][18]23 . However, we emphasized early on that observing higher intra-subject prediction scores compared to inter-subject scores was not a useful observation we believed provided informative predictions-they needed perform better than what any naive group averaging could predict on the cortical surface.…”

“…performance for a large number of target contrasts. Notably, these results not only predict individual subject differences, i.e., 'connectome-fingerprints' 21,22 , as many have previously shown [12][13][14][15][16][17][18]23 ; they provide support that whole-cortex prediction by a model exceeds what any kind of group averaging, i.e., baselines, could achieve-a point we will reiterate the importance of.…”

“…The relationship between rsfMRI and tfMRI for individual subject prediction can be captured by a regression problem, as illustrated in figure 1. This topic has been addressed in numerous studies [11][12][13][14][15][16][17][18] . Here we re-examined methods that address this problem using machine learning techniques with only functional data.…”

Predicting Sex from Resting-State fMRI Across Multiple Independent Acquired Datasets

“…Thus, investigating sex differences at the level of BOLD fluctuation may reveal if there is strong evidence of sex differences. Recently, machine learning (ML) techniques have been used widely to perform classification and regression on neuroscience data (Al Zoubi, Awad, & Kasabov, 2018;Al Zoubi, Ki Wong, et al, 2018;Campbell et al, 2020;Cohen, Chen, Parker Jones, Niu, & Wang, 2020;Du, Fu, & Calhoun, 2018;Garner et al, 2019;Kazeminejad & Sotero, 2019;Saccà et al, 2019). Some works focused on using ML for classifying subjects into male and female using functional (Ktena et al, 2018;Zhang, Dougherty, Baum, White, & Michael, 2018) and structural data (Chekroud et al, 2016;Feis, Brodersen, von Cramon, Luders, & Tittgemeyer, 2013;Rosenblatt, 2016).…”

Jumping over baselines with new methods to predict activation maps from resting-state fMRI

“…1. This topic has been addressed in numerous studies [11][12][13][14][15][16][17][18] . Here we re-examined methods that address this problem using machine learning techniques with only functional data.…”