Interpreting models interpreting brain dynamics

Abstract

Brain dynamics are highly complex and yet hold the key to understanding brain function and dysfunction. The dynamics captured by resting-state functional magnetic resonance imaging data are noisy, high-dimensional, and not readily interpretable. The typical approach of reducing this data to low-dimensional features and focusing on the most predictive features comes with strong assumptions and can miss essential aspects of the underlying dynamics. In contrast, introspection of discriminatively trained deep learning models may uncover disorder-relevant elements of the signal at the level of individual time points and spatial locations. Yet, the difficulty of reliable training on high-dimensional low sample size datasets and the unclear relevance of the resulting predictive markers prevent the widespread use of deep learning in functional neuroimaging. In this work, we introduce a deep learning framework to learn from high-dimensional dynamical data while maintaining stable, ecologically valid interpretations. Results successfully demonstrate that the proposed framework enables learning the dynamics of resting-state fMRI directly from small data and capturing compact, stable interpretations of features predictive of function and dysfunction.

Explainable Fuzzy Clustering Framework Reveals Divergent Default Mode Network Connectivity Dynamics in Schizophrenia

“…In this study, we used the Functional Imaging Biomedical Informatics Research Network (FBIRN) dataset, consisting of rs-fMRI recordings from 151 SZs and 160 HCs [54]. The dataset has been used in many studies, both related to fMRI dFNC clustering and classification [21], [24], [43], [47]. In addition to neuroimaging data, positive and negative symptom severity scores from the Positive and Negative Syndrome Scale (PANSS) [55].…”

“…Lastly, whereas the mean feature would capture how strongly a participant resided within a given state and the variance feature would capture how the similarity of a participant to each fuzzy state varied over time, the range feature sought to give insight into the more extreme probabilities for each time-series that might otherwise be obscured. The utility of this feature makes sense given that previous studies have identified the effects of SZ upon the brain to be highly localized [24].…”

“…These include electroencephalography (EEG), magnetoencephalography (MEG), and fMRI. All three modalities – EEG [16]–[18], MEG [19], and fMRI [13], [20]–[24] - have been used extensively for SZ analysis. EEG and MEG capture much higher resolution temporal information.…”

“…Many studies have analyzed resting state fMRI data within the context of SZ and other neuropsychiatric disorders. For example, studies have used independent components (ICs) [24], spectral features [9] [33], and functional network connectivity. Functional network connectivity offers a unique benefit in that it provides insights into the interaction of different brain regions and networks.…”

Pairing Explainable Deep Learning Classification with Clustering to Uncover Effects of Schizophrenia Upon Whole Brain Functional Network Connectivity Dynamics

“…Given previous studies comparing local and global explanations (59), one can reasonably conclude that averaging or using global explanations could obscure the heterogeneous effects of disorders across individuals. Additionally, with the exception of a couple studies that quantify aspects of explanations (9,46), these studies often do not provide much insight into how disorders affect brain dynamics over time. Clustering analyses are very useful in providing insights into dynamics that would typically be obscured by classification approaches.…”