Abstract:Recognition of the detrimental impact of participant motion on functional connectivity measures has led to the adoption of increasingly stringent quality control standards to minimize potential motion artifacts. These stringent standards can lead to the exclusion of many participants, creating a tension with a countervailing requirement for large sample sizes that can provide adequate statistical power, particularly for brain-behavior association studies. Here, we test and validate two techniques aimed at mitigating the impact of head motion on functional connectivity estimates, and show that these techniques enable the retention of a substantial proportion of participants who would otherwise be excluded based on motion criteria, such as a minimum mean framewise displacement (FD) threshold. Specifically, we first show that functional connectomes computed using time series data that have been ordered according to motion (i.e., framewise displacement - FD) and (1) subsetted to include the lowest-motion time points ("motion ordered") or (2) subsetted and resampled ("bagged") are reproducible, in that they enable the successful identification of an individual from a group using functional connectome fingerprinting. Second, we demonstrate that motion-ordered and bagged functional connectomes yield robust brain-behavior associations, which, when examined as a function of sample size, are comparable to those obtained using the standard full time series. Finally, we show that the utility of both approaches lies in maximizing participant inclusivity by allowing for the retention of high-motion participants that would otherwise be discarded. Given equivalent performance of the two approaches across these tests of reproducibility, validity, and utility, we conclude by recommending motion-ordering to enable data rescue, maximize inclusivity, and address the need for adequately powered samples in functional connectivity research, while maintaining stringent data quality standards. While our findings were reproducible across different head motion thresholds and edges in the functional connectome, we outline possibilities for further validation and assessment of generalization using other behavioral phenotypes and consortia datasets.

Data rescue in high-motion youth cohorts for robust and reproducible brain-behavior relationships