Explaining deep learning-based representations of resting state functional connectivity data: focusing on interpreting nonlinear patterns in autism spectrum disorder.

Background: Resting state Functional Magnetic Resonance Imaging fMRI (rs-fMRI) has been used extensively to study brain function in psychiatric disorders, yielding insights into brain organization. However, the high dimensionality of the rs-fMRI data presents significant challenges for data analysis. Variational autoencoders (VAEs), a type of neural network, have been instrumental in extracting low-dimensional latent representations of resting state functional connectivity (rsFC) patterns, thereby addressing the complex nonlinear structure of rs-fMRI data. Despite these advances, interpreting these latent representations remains a challenge. This paper aims to address this gap by developing explainable VAE models and testing their utility using rs-fMRI data in autism spectrum disorder (ASD). Methods: One-thousand one hundred and fifty participants (601 healthy controls [HC] and 549 patients with ASD) were included in the analysis. RsFC correlation matrices were extracted from the preprocessed rs-fMRI data using the Power atlas, which includes 264 regions of interest (ROIs). Then VAEs were trained in an unsupervised manner. Lastly, we introduce our latent contribution scores to explain the relationship between estimated representations and the original rs-fMRI brain measures. Results: We quantified the latent contribution scores for both the ASD and HC groups at the network level. We found that both ASD and HC groups share the top network connectivitives contributing to all estimated latent components. For example, latent 0 was driven by rsFC within ventral attention network (VAN) in both the ASD and HC. However, we found significant differences in the latent contribution scores between the ASD and HC groups within the VAN for latent 0 and the sensory/somatomotor network for latent 2. Conclusion: This study introduced latent contribution scores to interpret nonlinear patterns identified by VAEs. These scores effectively capture changes in each observed rsFC feature as the estimated latent representation changes, enabling an explainable deep learning model that better understands the underlying neural mechanisms of ASD.

Intrusive Traumatic Re-Experiencing Domain: Functional Connectivity Feature Classification by the ENIGMA PTSD Consortium.

Background: Intrusive traumatic re-experiencing domain (ITRED) was recently introduced as a novel perspective on posttraumatic psychopathology, proposing to focus research of posttraumatic stress disorder (PTSD) on the unique symptoms of intrusive and involuntary re-experiencing of the trauma, namely, intrusive memories, nightmares, and flashbacks. The aim of the present study was to explore ITRED from a neural network connectivity perspective. Methods: Data were collected from 9 sites taking part in the ENIGMA (Enhancing Neuro Imaging Genetics through Meta Analysis) PTSD Consortium (n= 584) and included itemized PTSD symptom scores and resting-state functional connectivity (rsFC) data. We assessed the utility of rsFC in classifying PTSD, ITRED-only (no PTSD diagnosis), and trauma-exposed (TE)-only (no PTSD or ITRED) groups using a machine learning approach, examining well-known networks implicated in PTSD. A random forest classification model was built on a training set using cross-validation, and the averaged cross-validation model performance for classification was evaluated using the area under the curve. The model was tested using a fully independent portion of the data (test dataset), and the test area under the curve was evaluated. Results: rsFC signatures differentiated TE-only participants from PTSD and ITRED-only participants at about 60% accuracy. Conversely, rsFC signatures did not differentiate PTSD from ITRED-only individuals (45% accuracy). Common features differentiating TE-only participants from PTSD and ITRED-only participants mainly involved default mode network-related pathways. Some unique features, such as connectivity within the frontoparietal network, differentiated TE-only participants from one group (PTSD or ITRED-only) but to a lesser extent from the other group. Conclusions: Neural network connectivity supports ITRED as a novel neurobiologically based approach to classifying posttrauma psychopathology.

Smaller total and subregional cerebellar volumes in posttraumatic stress disorder: a mega-analysis by the ENIGMA-PGC PTSD workgroup.

Although the cerebellum contributes to higher-order cognitive and emotional functions relevant to posttraumatic stress disorder (PTSD), prior research on cerebellar volume in PTSD is scant, particularly when considering subregions that differentially map on to motor, cognitive, and affective functions. In a sample of 4215 adults (PTSD n = 1642; Control n = 2573) across 40 sites from the ENIGMA-PGC PTSD working group, we employed a new state-of-the-art deep-learning based approach for automatic cerebellar parcellation to obtain volumetric estimates for the total cerebellum and 28 subregions. Linear mixed effects models controlling for age, gender, intracranial volume, and site were used to compare cerebellum volumes in PTSD compared to healthy controls (88% trauma-exposed). PTSD was associated with significant grey and white matter reductions of the cerebellum. Compared to controls, people with PTSD demonstrated smaller total cerebellum volume, as well as reduced volume in subregions primarily within the posterior lobe (lobule VIIB, crus II), vermis (VI, VIII), flocculonodular lobe (lobule X), and corpus medullare (all p-FDR < 0.05). Effects of PTSD on volume were consistent, and generally more robust, when examining symptom severity rather than diagnostic status. These findings implicate regionally specific cerebellar volumetric differences in the pathophysiology of PTSD. The cerebellum appears to play an important role in higher-order cognitive and emotional processes, far beyond its historical association with vestibulomotor function. Further examination of the cerebellum in trauma-related psychopathology will help to clarify how cerebellar structure and function may disrupt cognitive and affective processes at the center of translational models for PTSD.

Neuroimaging-based classification of PTSD using data-driven computational approaches: A multisite big data study from the ENIGMA-PGC PTSD consortium.

BACKGROUND: Recent advances in data-driven computational approaches have been helpful in devising tools to objectively diagnose psychiatric disorders. However, current machine learning studies limited to small homogeneous samples, different methodologies, and different imaging collection protocols, limit the ability to directly compare and generalize their results. Here we aimed to classify individuals with PTSD versus controls and assess the generalizability using a large heterogeneous brain datasets from the ENIGMA-PGC PTSD Working group. METHODS: We analyzed brain MRI data from 3,477 structural-MRI; 2,495 resting state-fMRI; and 1,952 diffusion-MRI. First, we identified the brain features that best distinguish individuals with PTSD from controls using traditional machine learning methods. Second, we assessed the utility of the denoising variational autoencoder (DVAE) and evaluated its classification performance. Third, we assessed the generalizability and reproducibility of both models using leave-one-site-out cross-validation procedure for each modality. RESULTS: We found lower performance in classifying PTSD vs. controls with data from over 20 sites (60 % test AUC for s-MRI, 59 % for rs-fMRI and 56 % for d-MRI), as compared to other studies run on single-site data. The performance increased when classifying PTSD from HC without trauma history in each modality (75 % AUC). The classification performance remained intact when applying the DVAE framework, which reduced the number of features. Finally, we found that the DVAE framework achieved better generalization to unseen datasets compared with the traditional machine learning frameworks, albeit performance was slightly above chance. CONCLUSION: These results have the potential to provide a baseline classification performance for PTSD when using large scale neuroimaging datasets. Our findings show that the control group used can heavily affect classification performance. The DVAE framework provided better generalizability for the multi-site data. This may be more significant in clinical practice since the neuroimaging-based diagnostic DVAE classification models are much less site-specific, rendering them more generalizable.

Explaining Deep Learning-Based Representations of Resting State Functional Connectivity Data: Focusing on Interpreting Nonlinear Patterns in Autism Spectrum Disorder.

Background: Resting state Functional Magnetic Resonance Imaging fMRI (rs-fMRI) has been used to study brain function in psychiatric disorders, yielding insight into brain organization. However, the high dimensionality of the rs-fMRI data presents challenges, and requires dimensionality reduction before applying machine learning techniques. Neural networks, specifically variational autoencoders (VAEs), have been instrumental in extracting low-dimensional latent representations of resting state functional connectivity patterns, addressing the complex nonlinear structure of rs-fMRI. However, interpreting those latent representations remains a challenge. This paper aims to address this gap by creating explainable VAE models and testing their utility using rs-fMRI data in autism spectrum disorder (ASD). Methods: One-thousand one hundred and fifty participants (601 HC and 549 patients with ASD) were included in the analysis. We extracted functional connectivity correlation matrices from the preprocessed rs-fMRI data using Power atlas with 264 ROIs. Then VAEs were trained in an unsupervised fashion. Lastly, we introduce our latent contribution scores to explain the relationship between estimated representations and the original rs-fMRI brain measures. Results: We quantified the latent contribution scores for the ASD and control groups at the network level. We found that both ASD and control groups share the top network connectivity that contribute to all estimated latent components. For example, latent 0 was driven by resting state functional connectivity patterns (rsFC) within ventral attention network in both the ASD and control. However, significant differences in the latent contribution scores between the ASD and control groups were discovered within the ventral attention network in latent 0 and the sensory/somatomotor network in latent 2. Conclusion: This study introduced latent contribution scores to interpret nonlinear patterns identified by VAEs. These scores effectively capture changes in each observed rsFC features as estimated latent representation changes, enabling an explainable deep learning model to better understand the underlying neural mechanism of ASD.

Structural brain features signaling trauma, PTSD, or resilience? A systematic exploration.

BACKGROUND: Studies have searched for neurobiological markers of trauma exposure, posttraumatic stress disorder (PTSD) diagnosis, and resilience to trauma to identify therapeutic targets for PTSD. Despite some promising results, findings are inconsistent. AIMS: The present study adopted a data-driven approach to systematically explore whether structural brain markers of trauma, PTSD, or resilience emerge when all are explored. MATERIALS & METHODS: Differences between clusters in the proportion of PTSD, healthy controls (HC), and trauma-exposed healthy controls (TEHC) served to indicate the presence of PTSD, trauma, and resilience markers, respectively. A total of 129 individuals, including 46 with PTSD, 49 TEHCs, and 34 HCs not exposed to trauma were scanned. Volumes, cortical thickness, and surface areas of interest were obtained from T1 structural MRI and used to identify data-driven clusters. RESULTS: Two clusters were identified, differing in the proportion of TEHCs but not of PTSDs or HCs. The cluster with the higher proportion of TEHCs, referred to as the resilience cluster, was characterized by higher volume in brain regions implicated in trauma exposure, especially the thalamus and rostral middle frontal gyrus. Cross-validation established the robustness and consistency of the identified clusters. DISCUSSION & CONCLUSION: Findings support the existence of structural brain markers of resilience.

Association of White Matter Integrity With Executive Function and Antidepressant Treatment Outcome in Patients With Late-Life Depression.

OBJECTIVE: While patients with late-life depression (LLD) often exhibit microstructural white matter alterations that can be identified with diffusion tensor imaging (DTI), there is a dearth of information concerning the links between DTI findings and specific cognitive performance, as well as between DTI measures and antidepressant treatment outcomes. DESIGN: Neuroimaging and cognitive tests were conducted at baseline in 71 older adults participating in a larger, 8-week duration antidepressant randomized controlled trial. Correlations between DTI measures of white matter integrity evaluated with tract-based spatial statistics, baseline neurocognitive performance, and prospective antidepressant treatment outcome were evaluated. RESULTS: Fractional anisotropy (FA), an index of white matter integrity, was significantly positively associated with better cognitive function as measured by the Initiation/Perseveration subscale of the Dementia Rating Scale in the bilateral superior longitudinal fasciculus (SLF), bilateral SLF-temporal, and right corticospinal tract (CST). An exploratory analysis limited to these tracts revealed that increased FA in the right CST, right SLF, and right SLF-temporal tracts was correlated with a greater decrease in depressive symptoms. Increased FA in the right CST predicted a greater chance of remission, while increased FA in the right CST and the right SLF predicted a greater chance of treatment response. CONCLUSION: In late-life depression LLD subjects, white matter integrity was positively associated with executive function in white matter tracts which act as key connecting structures underlying the cognitive control network. These tracts may play a role as a positive prognostic factor in antidepressant treatment outcome.