Deriving reproducible biomarkers from multi-site resting-state data: An Autism-based example

Abstract

Resting-state functional Magnetic Resonance Imaging (R-fMRI) holds the promise to reveal functional biomarkers of neuropsychiatric disorders. However, extracting such biomarkers is challenging for complex multi-faceted neuropathologies, such as autism spectrum disorders. Large multi-site datasets increase sample sizes to compensate for this complexity, at the cost of uncontrolled heterogeneity. This heterogeneity raises new challenges, akin to those face in realistic diagnostic applications. Here, we demonstrate the feasibility of inter-site classification of neuropsychiatric status, with an application to the Autism Brain Imaging Data Exchange (ABIDE) database, a large (N=871) multi-site autism dataset. For this purpose, we investigate pipelines that extract the most predictive biomarkers from the data. These R-fMRI pipelines build participant-specific connectomes from functionally-defined brain areas. Connectomes are then compared across participants to learn patterns of connectivity that differentiate typical controls from individuals with autism. We predict this neuropsychiatric status for participants from the same acquisition sites or different, unseen, ones. Good choices of methods for the various steps of the pipeline lead to 67% prediction accuracy on the full ABIDE data, which is significantly better than previously reported results. We perform extensive validation on multiple subsets of the data defined by different inclusion criteria. These enables detailed analysis of the factors contributing to successful connectome-based prediction. First, prediction accuracy improves as we include more subjects, up to the maximum amount of subjects available. Second, the definition of functional brain areas is of paramount importance for biomarker discovery: brain areas extracted from large R-fMRI datasets outperform reference atlases in the classification tasks.

Enhancing studies of the connectome in autism using the autism brain imaging data exchange II

“…Replication samples are needed to minimize false positives and avoid settling for ‘approximate replications’ 19 —a practice that has plagued biological psychiatry 19 and neuroscience more broadly 17 . Additionally, as recently demonstrated, the utility of datasets for prediction increases with sample size—even if heterogeneous data sources are used to amass large samples 23 .…”

“…Finally, we note that along with the challenges related to its multisite post-hoc data aggregation, ABIDE II also offers a unique opportunity to develop analytical approaches to address these challenges. For example, a recent effort based on ABIDE I demonstrated the ability to optimize classifiers for the prediction of data from previously unseen imaging sites 23 .…”

Joint prediction of multiple scores captures better individual traits from brain images

“…It is also easy to acquire and to compare across subjects, including on diminished subjects. Brain connectivity at rest is well suited to study pathologies (Greicius, 2008) as it is suitable for diminished patients and is impacted by neurodegenera-70 tive (Wang et al, 2006), or neuropsychiatric disorders (Craddock et al, 2009;Castellanos et al, 2013;Abraham et al, 2016). However, predicting clinical status of psychiatric or neurological patients from these data is challenging due to the low signal-to-noise ratio of the blood-oxygen-level dependent…”

Identifying Autism from Resting-State fMRI Using Long Short-Term Memory Networks

“…Among the single layer models, the highest subject accuracy was achieved for the LSTM with 32 hidden nodes (68.5%). Compared to the most competitive result using the majority of the ABIDE cohort [1], the difference between our accuracy and chance is over 3% higher, while our dataset contained more challenging, heterogeneous data with 25% more subjects. Furthermore, compared to the study with the closest number of subjects to ours [9], our model improved accuracy compared to chance by 9%.…”

“…Functional connectivity measures are used as predictors for classifying ASD v.s. neurotypical control, using popular learning methods such as support vector machines, random forests, or ridge regression [13,3,1]. Pairwise connections deemed important for accurate classification are then potential biomarkers for ASD.…”