Wide-field calcium imaging of cortical activation and functional connectivity in externally- and internally-driven locomotion.

The role of the cerebral cortex in self-initiated versus sensory-driven movements is central to understanding volitional action. Whether the differences in these two movement classes are due to specific cortical areas versus more cortex-wide engagement is debated. Using wide-field Ca2+ imaging, we compared neural dynamics during spontaneous and motorized treadmill locomotion, determining the similarities and differences in cortex-wide activation and functional connectivity (FC). During motorized locomotion, the cortex exhibits greater activation globally prior to and during locomotion starting compared to spontaneous and less during steady-state walking, during stopping, and after termination. Both conditions are characterized by FC increases in anterior secondary motor cortex (M2) nodes and decreases in all other regions. There are also cortex-wide differences; most notably, M2 decreases in FC with all other nodes during motorized stopping and after termination. Therefore, both internally- and externally-generated movements widely engage the cortex, with differences represented in cortex-wide activation and FC patterns.

Sex differences in functional cortical organization reflect differences in network topology rather than cortical morphometry.

Differences in brain size between the sexes are consistently reported. However, the consequences of this anatomical difference on sex differences in intrinsic brain function remain unclear. In the current study, we investigate whether sex differences in intrinsic cortical functional organization may be associated with differences in cortical morphometry, namely different measures of brain size, microstructure, and the geodesic distance of connectivity profiles. For this, we compute a low dimensional representation of functional cortical organization, the sensory-association axis, and identify widespread sex differences. Contrary to our expectations, sex differences in functional organization do not appear to be systematically associated with differences in total surface area, microstructural organization, or geodesic distance, despite these morphometric properties being per se associated with functional organization and differing between sexes. Instead, functional sex differences in the sensory-association axis are associated with differences in functional connectivity profiles and network topology. Collectively, our findings suggest that sex differences in functional cortical organization extend beyond sex differences in cortical morphometry.

Functional magnetic resonance imaging signal has sub-second temporal accuracy.

Neuronal activation sequence information is essential for understanding brain functions. Extracting such timing information from blood-oxygenation-level-dependent functional magnetic resonance imaging (fMRI) signals is confounded by local cerebral vascular reactivity (CVR), which varies across brain locations. Thus, detecting neuronal synchrony as well as inferring inter-regional causal modulation using fMRI signals can be biased. Here we used fast fMRI measurements sampled at 10 Hz to measure the fMRI latency difference between visual and sensorimotor areas when participants engaged in a visuomotor task. The regional fMRI timing was calibrated by subtracting the CVR latency measured by a breath-holding task. After CVR calibration, the fMRI signal at the lateral geniculate nucleus (LGN) preceded that at the visual cortex by 496 ms, followed by the fMRI signal at the sensorimotor cortex with a latency of 464 ms. Sequential LGN, visual, and sensorimotor cortex activations were found in each participant after the CVR calibration. These inter-regional fMRI timing differences across and within participants were more closely related to the reaction time after the CVR calibration. Our results suggested the feasibility of mapping brain activity using fMRI with accuracy in hundreds of milliseconds.

Neighborhood structure-guided brain functional networks estimation for mild cognitive impairment identification.

The adoption and growth of functional magnetic resonance imaging (fMRI) technology, especially through the use of Pearson's correlation (PC) for constructing brain functional networks (BFN), has significantly advanced brain disease diagnostics by uncovering the brain's operational mechanisms and offering biomarkers for early detection. However, the PC always tends to make for a dense BFN, which violates the biological prior. Therefore, in practice, researchers use hard-threshold to remove weak connection edges or introduce l 1-norm as a regularization term to obtain sparse BFNs. However, these approaches neglect the spatial neighborhood information between regions of interest (ROIs), and ROI with closer distances has higher connectivity prospects than ROI with farther distances due to the principle of simple wiring costs in resent studies. Thus, we propose a neighborhood structure-guided BFN estimation method in this article. In detail, we figure the ROIs' Euclidean distances and sort them. Then, we apply the K-nearest neighbor (KNN) to find out the top K neighbors closest to the current ROIs, where each ROI's K neighbors are independent of each other. We establish the connection relationship between the ROIs and these K neighbors and construct the global topology adjacency matrix according to the binary network. Connect ROI nodes with k nearest neighbors using edges to generate an adjacency graph, forming an adjacency matrix. Based on adjacency matrix, PC calculates the correlation coefficient between ROIs connected by edges, and generates the BFN. With the purpose of evaluating the performance of the introduced method, we utilize the estimated BFN for distinguishing individuals with mild cognitive impairment (MCI) from the healthy ones. Experimental outcomes imply this method attains better classification performance than the baselines. Additionally, we compared it with the most commonly used time series methods in deep learning. Results of the performance of K-nearest neighbor-Pearson's correlation (K-PC) has some advantage over deep learning.

Medial to lateral frontal functional connectivity mapping reveals the organization of cingulate cortex.

The functional organization of the frontal lobe is a source of debate, focusing on broad functional subdivisions, large-scale networks, or local refined specificities. Multiple neurocognitive models have tried to explain how functional interactions between cingulate and lateral frontal regions contribute to decision making and cognitive control, but their neuroanatomical bases remain unclear. We provide a detailed description of the functional connectivity between cingulate and lateral frontal regions using resting-state functional MRI in rhesus macaques. The analysis focuses on the functional connectivity of the rostral part of the cingulate sulcus with the lateral frontal cortex. Data-driven and seed-based analysis revealed three clusters within the cingulate sulcus organized along the rostro-caudal axis: the anterior, mid, and posterior clusters display increased functional connectivity with, respectively, the anterior lateral prefrontal regions, face-eye lateral frontal motor cortical areas, and hand lateral frontal motor cortex. The location of these clusters can be predicted in individual subjects based on morphological landmarks. These results suggest that the anterior cluster corresponds to the anterior cingulate cortex, whereas the posterior clusters correspond to the face-eye and hand cingulate motor areas within the anterior midcingulate cortex. These data provide a comprehensive framework to identify cingulate subregions based on functional connectivity and local organization.

Concurrent tDCS-fMRI after stroke reveals link between attention network organization and motor improvement.

Restoring motor function after stroke necessitates involvement of numerous cognitive systems. However, the impact of damage to motor and cognitive network organization on recovery is not well understood. To discover correlates of successful recovery, we explored imaging characteristics in chronic stroke subjects by combining noninvasive brain stimulation and fMRI. Twenty stroke survivors (6 months or more after stroke) were randomly assigned to a single session of transcranial direct current stimulation (tDCS) or sham during image acquisition. Twenty healthy subjects were included as controls. tDCS was limited to 10 min at 2 mA to serve as a mode of network modulation rather than therapeutic delivery. Fugl-Meyer Assessments (FMA) revealed significant motor improvement in the chronic stroke group receiving active stimulation (p = 0.0005). Motor changes in this group were correlated in a data-driven fashion with imaging features, including functional connectivity (FC), surface-based morphometry, electric field modeling and network topology, focusing on relevant regions of interest. We observed stimulation-related changes in FC in supplementary motor (p = 0.0029), inferior frontal gyrus (p = 0.0058), and temporo-occipital (p = 0.0095) areas, though these were not directly related to motor improvement. The feature most strongly associated with FMA improvement in the chronic stroke cohort was graph topology of the dorsal attention network (DAN), one of the regions surveyed and one with direct connections to each of the areas with FC changes. Chronic stroke subjects with a greater degree of motor improvement had lower signal transmission cost through the DAN (p = 0.029). While the study was limited by a small stroke cohort with moderate severity and variable lesion location, these results nevertheless suggest a top-down role for higher order areas such as attention in helping to orchestrate the stroke recovery process.

Brain dynamics supported by a hierarchy of complex correlation patterns defining a robust functional architecture.

Functional magnetic resonance imaging (fMRI) provides insights into cognitive processes with significant clinical potential. However, delays in brain region communication and dynamic variations are often overlooked in functional network studies. We demonstrate that networks extracted from fMRI cross-correlation matrices, considering time lags between signals, show remarkable reliability when focusing on statistical distributions of network properties. This reveals a robust brain functional connectivity pattern, featuring a sparse backbone of strong 0-lag correlations and weaker links capturing coordination at various time delays. This dynamic yet stable network architecture is consistent across rats, marmosets, and humans, as well as in electroencephalogram (EEG) data, indicating potential universality in brain dynamics. Second-order properties of the dynamic functional network reveal a remarkably stable hierarchy of functional correlations in both group-level comparisons and test-retest analyses. Validation using alcohol use disorder fMRI data uncovers broader shifts in network properties than previously reported, demonstrating the potential of this method for identifying disease biomarkers.

Decoding acceptance and reappraisal strategies from resting state macro networks.

Acceptance and reappraisal are considered adaptive emotion regulation strategies. While previous studies have explored the neural underpinnings of these strategies using task-based fMRI and sMRI, a gap exists in the literature concerning resting-state functional brain networks' contributions to these abilities, especially regarding acceptance. Another intriguing question is whether these strategies rely on similar or different neural mechanisms. Building on the well-known improved emotion regulation and increased cognitive flexibility of individuals who rely on acceptance, we expected to find decreased activity inside the affective network and increased activity inside the executive and sensorimotor networks to be predictive of acceptance. We also expect that these networks may be associated at least in part with reappraisal, indicating a common mechanism behind different strategies. To test these hypotheses, we conducted a functional connectivity analysis of resting-state data from 134 individuals (95 females; mean age: 30.09 ± 12.87 years, mean education: 12.62 ± 1.41 years). To assess acceptance and reappraisal abilities, we used the Cognitive Emotion Regulation Questionnaire (CERQ) and a group-ICA unsupervised machine learning approach to identify resting-state networks. Subsequently, we conducted backward regression to predict acceptance and reappraisal abilities. As expected, results indicated that acceptance was predicted by decreased affective, and executive, and increased sensorimotor networks, while reappraisal was predicted by an increase in the sensorimotor network only. Notably, these findings suggest both distinct and overlapping brain contributions to acceptance and reappraisal strategies, with the sensorimotor network potentially serving as a core common mechanism. These results not only align with previous findings but also expand upon them, illustrating the complex interplay of cognitive, affective, and sensory abilities in emotion regulation.

Predictions enable top-down pattern separation in the macaque face-processing hierarchy.

Distinguishing faces requires well distinguishable neural activity patterns. Contextual information may separate neural representations, leading to enhanced identity recognition. Here, we use functional magnetic resonance imaging to investigate how predictions derived from contextual information affect the separability of neural activity patterns in the macaque face-processing system, a 3-level processing hierarchy in ventral visual cortex. We find that in the presence of predictions, early stages of this hierarchy exhibit well separable and high-dimensional neural geometries resembling those at the top of the hierarchy. This is accompanied by a systematic shift of tuning properties from higher to lower areas, endowing lower areas with higher-order, invariant representations instead of their feedforward tuning properties. Thus, top-down signals dynamically transform neural representations of faces into separable and high-dimensional neural geometries. Our results provide evidence how predictive context transforms flexible representational spaces to optimally use the computational resources provided by cortical processing hierarchies for better and faster distinction of facial identities.

An end-to-end deep learning pipeline to derive blood input with partial volume corrections for automated parametric brain PET mapping.

Dynamic 2-[18F] fluoro-2-deoxy-D-glucose positron emission tomography (dFDG-PET) for human brain imaging has considerable clinical potential, yet its utilization remains limited. A key challenge in the quantitative analysis of dFDG-PET is characterizing a patient-specific blood input function, traditionally reliant on invasive arterial blood sampling. This research introduces a novel approach employing non-invasive deep learning model-based computations from the internal carotid arteries (ICA) with partial volume (PV) corrections, thereby eliminating the need for invasive arterial sampling. We present an end-to-end pipeline incorporating a 3D U-Net based ICA-net for ICA segmentation, alongside a Recurrent Neural Network (RNN) based MCIF-net for the derivation of a model-corrected blood input function (MCIF) with PV corrections. The developed 3D U-Net and RNN was trained and validated using a 5-fold cross-validation approach on 50 human brain FDG PET scans. The ICA-net achieved an average Dice score of 82.18% and an Intersection over Union of 68.54% across all tested scans. Furthermore, the MCIF-net exhibited a minimal root mean squared error of 0.0052. The application of this pipeline to ground truth data for dFDG-PET brain scans resulted in the precise localization of seizure onset regions, which contributed to a successful clinical outcome, with the patient achieving a seizure-free state after treatment. These results underscore the efficacy of the ICA-net and MCIF-net deep learning pipeline in learning the ICA structure's distribution and automating MCIF computation with PV corrections. This advancement marks a significant leap in non-invasive neuroimaging.

Convolutional neural networks can identify brain interactions involved in decoding spatial auditory attention.

Human listeners have the ability to direct their attention to a single speaker in a multi-talker environment. The neural correlates of selective attention can be decoded from a single trial of electroencephalography (EEG) data. In this study, leveraging the source-reconstructed and anatomically-resolved EEG data as inputs, we sought to employ CNN as an interpretable model to uncover task-specific interactions between brain regions, rather than simply to utilize it as a black box decoder. To this end, our CNN model was specifically designed to learn pairwise interaction representations for 10 cortical regions from five-second inputs. By exclusively utilizing these features for decoding, our model was able to attain a median accuracy of 77.56% for within-participant and 65.14% for cross-participant classification. Through ablation analysis together with dissecting the features of the models and applying cluster analysis, we were able to discern the presence of alpha-band-dominated inter-hemisphere interactions, as well as alpha- and beta-band dominant interactions that were either hemisphere-specific or were characterized by a contrasting pattern between the right and left hemispheres. These interactions were more pronounced in parietal and central regions for within-participant decoding, but in parietal, central, and partly frontal regions for cross-participant decoding. These findings demonstrate that our CNN model can effectively utilize features known to be important in auditory attention tasks and suggest that the application of domain knowledge inspired CNNs on source-reconstructed EEG data can offer a novel computational framework for studying task-relevant brain interactions.

Disparity in temporal and spatial relationships between resting-state electrophysiological and fMRI signals.

Resting-state brain networks (RSNs) have been widely applied in health and disease, but the interpretation of RSNs in terms of the underlying neural activity is unclear. To address this fundamental question, we conducted simultaneous recordings of whole-brain resting-state functional magnetic resonance imaging (rsfMRI) and electrophysiology signals in two separate brain regions of rats. Our data reveal that for both recording sites, spatial maps derived from band-specific local field potential (LFP) power can account for up to 90% of the spatial variability in RSNs derived from rsfMRI signals. Surprisingly, the time series of LFP band power can only explain to a maximum of 35% of the temporal variance of the local rsfMRI time course from the same site. In addition, regressing out time series of LFP power from rsfMRI signals has minimal impact on the spatial patterns of rsfMRI-based RSNs. This disparity in the spatial and temporal relationships between resting-state electrophysiology and rsfMRI signals suggests that electrophysiological activity alone does not fully explain the effects observed in the rsfMRI signal, implying the existence of an rsfMRI component contributed by 'electrophysiology-invisible' signals. These findings offer a novel perspective on our understanding of RSN interpretation.

The brain contains many cells known as neurons that send and receive messages in the form of electrical signals. The neurons in different regions of the brain must coordinate their activities to enable the brain to operate properly. Researchers often use a method called resting-state functional magnetic resonance imaging (rsfMRI) to study how different areas of the brain work together. This method indirectly measures brain activity by detecting the changes in blood flow to different areas of the brain. Regions that are working together will become active (that is, have higher blood flow and corresponding rsfMRI signal) and inactive (have lower blood flow and a lower rsfMRI signal) at the same time. These coordinated patterns of brain activity are known as "resting-state brain networks" (RSNs). Previous studies have identified RSNs in many different situations, but we still do not fully understand how these changes in blood flow are related to what is happening in the neurons themselves. To address this question, Tu et al. performed rsfMRI while also measuring the electrical activity (referred to as electrophysiology signals) in two distinct regions of the brains of rats. The team then used the data to generate maps of RSNs in those brain regions. This revealed that rsfMRI signals and electrophysiology signals produced almost identical maps in terms of the locations of the RSNs. However, the electrophysiology signals only contributed a small amount to the changes in the local rsfMRI signals over time at the same recording site. This suggests that RSNs may arise from cell activities that are not detectable by electrophysiology but do regulate blood flow to neurons. The findings of Tu et al. offer a new perspective for interpreting how rsfMRI signals relate to the activities of neurons. Further work is needed to explore all the features of the electrophysiology signals and test other methods to compare these features with rsfMRI signals in the same locations.

Dysfunction of thalamocortical circuits in early-onset schizophrenia.

Previous studies have demonstrated that the thalamus is involved in multiple functional circuits in participants with schizophrenia. However, less is known about the thalamocortical circuit in the rare subtype of early-onset schizophrenia. A total of 110 participants with early-onset schizophrenia (47 antipsychotic-naive patients) and 70 matched healthy controls were recruited and underwent resting-state functional and diffusion-weighted magnetic resonance imaging scans. A data-driven parcellation method that combined the high spatial resolution of diffusion magnetic resonance imaging and the high sensitivity of functional magnetic resonance imaging was used to divide the thalamus. Next, the functional connectivity between each thalamic subdivision and the cortex/cerebellum was investigated. Compared to healthy controls, individuals with early-onset schizophrenia exhibited hypoconnectivity between subdivisions of the thalamus and the frontoparietal network, visual network, ventral attention network, somatomotor network and cerebellum, and hyperconnectivity between subdivisions of thalamus and the parahippocampal and temporal gyrus, which were included in limbic network. The functional connectivity between the right posterior cingulate cortex and 1 subdivision of the thalamus (region of interest 1) was positively correlated with the general psychopathology scale score. This study showed that the specific thalamocortical dysconnection in individuals with early-onset schizophrenia involves the prefrontal, auditory and visual cortices, and cerebellum. This study identified thalamocortical connectivity as a potential biomarker and treatment target for early-onset schizophrenia.

Development of a noninvasive olfactory stimulation fMRI system in marmosets.

Olfactory dysfunction is associated with aging and the earliest stages of neurodegenerative diseases, such as Alzheimer's and Parkinson's diseases; it is thought to be an early biomarker of cognitive decline. In marmosets, a small non-human primate model used in brain research, olfactory pathway activity during olfactory stimulation has not been well studied because of the difficulty in clearly switching olfactory stimuli inside a narrow MRI. Here, we developed an olfactory-stimulated fMRI system using a small-aperture MRI machine. The olfactory presentation system consisted of two tubes, one for supply and one for suction of olfactory stimulants and a balloon valve. A balloon valve installed in the air supply tube controlled the presentation of the olfactory stimulant, which enabled sharp olfactory stimulation within MRI, such as 30 s of stimulation repeated five times at five-minute intervals. The olfactory stimulation system was validated in vivo and in a simulated system. fMRI analysis showed a rapid increase in signal values within 30 s of olfactory stimulation in eight regions related to the sense of smell. As these regions include those associated with Alzheimer's and Parkinson's diseases, olfactory stimulation fMRI may be useful in clarifying the relationship between olfactory dysfunction and dementia in non-human primates.

Spin-lock based fast whole-brain 3D macromolecular proton fraction mapping of relapsing-remitting multiple sclerosis.

A sensitive and efficient imaging technique is required to assess the subtle abnormalities occurring in the normal-appearing white matter (NAWM) and normal-appearing grey matter (NAGM) in patients with relapsing-remitting multiple sclerosis (RRMS). In this study, a fast 3D macromolecular proton fraction (MPF) quantification based on spin-lock (fast MPF-SL) sequence was proposed for brain MPF mapping. Thirty-four participants, including 17 healthy controls and 17 RRMS patients were prospectively recruited. We conducted group comparison and correlation between conventional MPF-SL, fast MPF-SL, and DWI, and compared differences in quantified parameters within MS lesions and the regional NAWM, NAGM, and normal-appearing deep grey matter (NADGN). MPF of MS lesions was significantly reduced (7.17% ± 1.15%, P < 0.01) compared to all corresponding normal-appearing regions. MS patients also showed significantly reduced mean MPF values compared with controls in NAGM (4.87% ± 0.38% vs 5.21% ± 0.32%, P = 0.01), NAWM (9.49% ± 0.69% vs 10.32% ± 0.59%, P < 0.01) and NADGM (thalamus 5.59% ± 0.67% vs 6.00% ± 0.41%, P = 0.04; caudate 5.10% ± 0.55% vs 5.53% ± 0.58%, P = 0.03). MPF and ADC showed abnormalities in otherwise normal appearing close to lesion areas (P < 0.01). In conclusion, time-efficient MPF mapping of the whole brain can be acquired efficiently (< 3 min) using fast MPF-SL. It offers a promising alternative way to detect white matter abnormalities in MS.

Cortical cognitive processing during reading captured using functional-near infrared spectroscopy.

Neuroimaging studies using functional magnetic resonance imaging (fMRI) have provided unparalleled insights into the fundamental neural mechanisms underlying human cognitive processing, such as high-level linguistic processes during reading. Here, we build upon this prior work to capture sentence reading comprehension outside the MRI scanner using functional near infra-red spectroscopy (fNIRS) in a large sample of participants (n = 82). We observed increased task-related hemodynamic responses in prefrontal and temporal cortical regions during sentence-level reading relative to the control condition (a list of non-words), replicating prior fMRI work on cortical recruitment associated with high-level linguistic processing during reading comprehension. These results lay the groundwork towards developing adaptive systems to support novice readers and language learners by targeting the underlying cognitive processes. This work also contributes to bridging the gap between laboratory findings and more real-world applications in the realm of cognitive neuroscience.

The causal role of the subthalamic nucleus in the inhibitory network.

The neural network mediating successful response inhibition mainly includes right hemisphere activation of the pre-supplementary motor area, inferior frontal gyrus (IFG), subthalamic nucleus (STN), and caudate nucleus. However, the causal role of these regions in the inhibitory network is undefined. Five patients with Parkinson's disease were assessed prior to and after therapeutic thermal ablation of the right STN in two separate functional magnetic resonance imaging (fMRI) sessions while performing a stop-signal task. Initiation times were faster but motor inhibition with the left hand (contralateral to the lesion) was significantly impaired as evident in prolonged stop-signal reaction times. Reduced inhibition after right subthalamotomy was associated (during successful inhibition) with the recruitment of basal ganglia regions outside the established inhibitory network. They included the putamen and caudate together with the anterior cingulate cortex and IFG of the left hemisphere. Subsequent network connectivity analysis (with the seed over the nonlesioned left STN) revealed a new inhibitory network after right subthalamotomies. Our results highlight the causal role of the right STN in the neural network for motor inhibition and the possible basal ganglia mechanisms for compensation upon losing a key node of the inhibition network.

Retinotopy drives the variation in scene responses across visual field map divisions of the occipital place area.

The occipital place area (OPA) is a scene-selective region on the lateral surface of human occipitotemporal cortex that spatially overlaps multiple visual field maps, as well as portions of cortex that are not currently defined as retinotopic. Here we combined population receptive field modeling and responses to scenes in a representational similarity analysis (RSA) framework to test the prediction that the OPA's visual field map divisions contribute uniquely to the overall pattern of scene selectivity within the OPA. Consistent with this prediction, the patterns of response to a set of complex scenes were heterogeneous between maps. To explain this heterogeneity, we tested the explanatory power of seven candidate models using RSA. These models spanned different scene dimensions (Content, Expanse, Distance), low- and high-level visual features, and navigational affordances. None of the tested models could account for the variation in scene response observed between the OPA's visual field maps. However, the heterogeneity in scene response was correlated with the differences in retinotopic profiles across maps. These data highlight the need to carefully examine the relationship between regions defined as category-selective and the underlying retinotopy, and they suggest that, in the case of the OPA, it may not be appropriate to conceptualize it as a single scene-selective region.

Spatiotemporal whole-brain activity and functional connectivity of melodies recognition.

Music is a non-verbal human language, built on logical, hierarchical structures, that offers excellent opportunities to explore how the brain processes complex spatiotemporal auditory sequences. Using the high temporal resolution of magnetoencephalography, we investigated the unfolding brain dynamics of 70 participants during the recognition of previously memorized musical sequences compared to novel sequences matched in terms of entropy and information content. Measures of both whole-brain activity and functional connectivity revealed a widespread brain network underlying the recognition of the memorized auditory sequences, which comprised primary auditory cortex, superior temporal gyrus, insula, frontal operculum, cingulate gyrus, orbitofrontal cortex, basal ganglia, thalamus, and hippocampus. Furthermore, while the auditory cortex responded mainly to the first tones of the sequences, the activity of higher-order brain areas such as the cingulate gyrus, frontal operculum, hippocampus, and orbitofrontal cortex largely increased over time during the recognition of the memorized versus novel musical sequences. In conclusion, using a wide range of analytical techniques spanning from decoding to functional connectivity and building on previous works, our study provided new insights into the spatiotemporal whole-brain mechanisms for conscious recognition of auditory sequences.

Nonlinear latent representations of high-dimensional task-fMRI data: Unveiling cognitive and behavioral insights in heterogeneous spatial maps.

Finding an interpretable and compact representation of complex neuroimaging data is extremely useful for understanding brain behavioral mapping and hence for explaining the biological underpinnings of mental disorders. However, hand-crafted representations, as well as linear transformations, may inadequately capture the considerable variability across individuals. Here, we implemented a data-driven approach using a three-dimensional autoencoder on two large-scale datasets. This approach provides a latent representation of high-dimensional task-fMRI data which can account for demographic characteristics whilst also being readily interpretable both in the latent space learned by the autoencoder and in the original voxel space. This was achieved by addressing a joint optimization problem that simultaneously reconstructs the data and predicts clinical or demographic variables. We then applied normative modeling to the latent variables to define summary statistics ('latent indices') and establish a multivariate mapping to non-imaging measures. Our model, trained with multi-task fMRI data from the Human Connectome Project (HCP) and UK biobank task-fMRI data, demonstrated high performance in age and sex predictions and successfully captured complex behavioral characteristics while preserving individual variability through a latent representation. Our model also performed competitively with respect to various baseline models including several variants of principal components analysis, independent components analysis and classical regions of interest, both in terms of reconstruction accuracy and strength of association with behavioral variables.