The effect of head motion on brain age prediction using deep convolutional neural networks.

Deep learning can be used effectively to predict participants' age from brain magnetic resonance imaging (MRI) data, and a growing body of evidence suggests that the difference between predicted and chronological age-referred to as brain-predicted age difference (brain-PAD)-is related to various neurological and neuropsychiatric disease states. A crucial aspect of the applicability of brain-PAD as a biomarker of individual brain health is whether and how brain-predicted age is affected by MR image artifacts commonly encountered in clinical settings. To investigate this issue, we trained and validated two different 3D convolutional neural network architectures (CNNs) from scratch and tested the models on a separate dataset consisting of motion-free and motion-corrupted T1-weighted MRI scans from the same participants, the quality of which were rated by neuroradiologists from a clinical diagnostic point of view. Our results revealed a systematic increase in brain-PAD with worsening image quality for both models. This effect was also observed for images that were deemed usable from a clinical perspective, with brains appearing older in medium than in good quality images. These findings were also supported by significant associations found between the brain-PAD and standard image quality metrics indicating larger brain-PAD for lower-quality images. Our results demonstrate a spurious effect of advanced brain aging as a result of head motion and underline the importance of controlling for image quality when using brain-predicted age based on structural neuroimaging data as a proxy measure for brain health.

Amygdala Volume is Associated with ADHD Risk and Severity Beyond Comorbidities in Adolescents: Clinical Testing of Brain Chart Reference Standards.

Understanding atypicalities in ADHD brain correlates is a step towards better understanding ADHD etiology. Efforts to map atypicalities at the level of brain structure have been hindered by the absence of normative reference standards. Recent publication of brain charts allows for assessment of individual variation relative to age- and sex-adjusted reference standards and thus estimation not only of case-control differences but also of intraindividual prediction. METHODS: Aim was to examine, whether brain charts can be applied in a sample of adolescents (N = 140, 38% female) to determine whether atypical brain subcortical and total volumes are associated with ADHD at-risk status and severity of parent-rated symptoms, accounting for self-rated anxiety and depression, and parent-rated oppositional defiant disorder (ODD) as well as motion. RESULTS: Smaller bilateral amygdala volume was associated with ADHD at-risk status, beyond effects of comorbidities and motion, and smaller bilateral amygdala volume was associated with inattention and hyperactivity/impulsivity, beyond effects of comorbidities except for ODD symptoms, and motion. CONCLUSIONS: Individual differences in amygdala volume meaningfully add to estimating ADHD risk and severity. Conceptually, amygdalar involvement is consistent with behavioral and functional imaging data on atypical reinforcement sensitivity as a marker of ADHD-related risk. Methodologically, results show that brain chart reference standards can be applied to address clinically informative, focused and specific questions.

Study protocol of the Hungarian Longitudinal Study of Healthy Brain Aging (HuBA).

Background and purpose:

Neuro­cog­nitive aging and the associated brain diseases impose a major social and economic burden. Therefore, substantial efforts have been put into revealing the lifestyle, the neurobiological and the genetic underpinnings of healthy neurocognitive aging. However, these studies take place almost exclusively in a limited number of highly-developed countries. Thus, it is an important open question to what extent their findings may generalize to neurocognitive aging in other, not yet investigated regions. The purpose of the Hungarian Longitudinal Study of Healthy Brain Aging (HuBA) is to collect multi-modal longitudinal data on healthy neurocognitive aging to address the data gap in this field in Central and Eastern Europe.

We adapted the Australian Ima­ging, Biomarkers and Lifestyle (AIBL) study of aging study protocol to local circumstances and collected demographic, lifestyle, men­tal and physical health, medication and medical history related information as well as re­cor­ded a series of magnetic resonance imaging (MRI) data. In addition, participants were al­so offered to participate in the collection of blood samples to assess circulating in­flam­matory biomarkers as well as a sleep study aimed at evaluating the general sleep quality based on multi-day collection of subjective sleep questionnaires and whole-night elec­troencephalographic (EEG) data.

Baseline data collection has al­ready been accomplished for more than a hundred participants and data collection in the se­cond
session is on the way. The collected data might reveal specific local trends or could also indicate the generalizability of previous findings. Moreover, as the HuBA protocol al­so offers a sleep study designed for tho­rough characterization of participants’ sleep quality and related factors, our extended multi-modal dataset might provide a base for incorporating these measures into healthy and clinical aging research. 

Besides its straightforward na­tional benefits in terms of health ex­pen­di­ture, we hope that this Hungarian initiative could provide results valid for the whole Cent­ral and Eastern European region and could also promote aging and Alzheimer’s disease research in these countries.

Automatic brain MRI motion artifact detection based on end-to-end deep learning is similarly effective as traditional machine learning trained on image quality metrics.

Head motion artifacts in magnetic resonance imaging (MRI) are an important confounding factor concerning brain research as well as clinical practice. For this reason, several machine learning-based methods have been developed for the automatic quality control of structural MRI scans. Deep learning offers a promising solution to this problem, however, given its data-hungry nature and the scarcity of expert-annotated datasets, its advantage over traditional machine learning methods in identifying motion-corrupted brain scans is yet to be determined. In the present study, we investigated the relative advantage of the two methods in structural MRI quality control. To this end, we collected publicly available T1-weighted images and scanned subjects in our own lab under conventional and active head motion conditions. The quality of the images was rated by a team of radiologists from the point of view of clinical diagnostic use. We present a relatively simple, lightweight 3D convolutional neural network trained in an end-to-end manner that achieved a test set (N = 411) balanced accuracy of 94.41% in classifying brain scans into clinically usable or unusable categories. A support vector machine trained on image quality metrics achieved a balanced accuracy of 88.44% on the same test set. Statistical comparison of the two models yielded no significant difference in terms of confusion matrices, error rates, or receiver operating characteristic curves. Our results suggest that these machine learning methods are similarly effective in identifying severe motion artifacts in brain MRI scans, and underline the efficacy of end-to-end deep learning-based systems in brain MRI quality control, allowing the rapid evaluation of diagnostic utility without the need for elaborate image pre-processing.

Movement-related artefacts (MR-ART) dataset of matched motion-corrupted and clean structural MRI brain scans.

Magnetic Resonance Imaging (MRI) provides a unique opportunity to investigate neural changes in healthy and clinical conditions. Its large inherent susceptibility to motion, however, often confounds the measurement. Approaches assessing, correcting, or preventing motion corruption of MRI measurements are under active development, and such efforts can greatly benefit from carefully controlled datasets. We present a unique dataset of structural brain MRI images collected from 148 healthy adults which includes both motion-free and motion-affected data acquired from the same participants. This matched dataset allows direct evaluation of motion artefacts, their impact on derived data, and testing approaches to correct for them. Our dataset further stands out by containing images with different levels of motion artefacts from the same participants, is enriched with expert scoring characterizing the image quality from a clinical point of view and is also complemented with standard image quality metrics obtained from MRIQC. The goal of the dataset is to raise awareness of the issue and provide a useful resource to assess and improve current motion correction approaches.

Effect of head motion-induced artefacts on the reliability of deep learning-based whole-brain segmentation.

Due to their robustness and speed, recently developed deep learning-based methods have the potential to provide a faster and hence more scalable alternative to more conventional neuroimaging analysis pipelines in terms of whole-brain segmentation based on magnetic resonance (MR) images. These methods were also shown to have higher test-retest reliability, raising the possibility that they could also exhibit superior head motion tolerance. We investigated this by comparing the effect of head motion-induced artifacts in structural MR images on the consistency of segmentation performed by FreeSurfer and recently developed deep learning-based methods to a similar extent. We used state-of-the art neural network models (FastSurferCNN and Kwyk) and developed a new whole-brain segmentation pipeline (ReSeg) to examine whether reliability depends on choice of deep learning method. Structural MRI scans were collected from 110 participants under rest and active head motion and were evaluated for image quality by radiologists. Compared to FreeSurfer, deep learning-based methods provided more consistent segmentations across different levels of image quality, suggesting that they also have the advantage of providing more reliable whole-brain segmentations of MR images corrupted by motion-induced artifacts, and provide evidence for their practical applicability in the study of brain structural alterations in health and disease.

Investigation of the relationship between visual feature binding in short- and long-term memory in healthy aging.

Binding visual features into coherent object representations is essential both in short- and long-term memory. However, the relationship between feature binding processes at different memory delays remains unexplored. Here, we addressed this question by using the Mnemonic Similarity Task and a delayed-estimation working memory task on a large sample of older adults. The results revealed that higher propensity to misbind object features in working memory is associated with lower lure discrimination performance in the mnemonic similarity task, suggesting that shared feature binding processes underlie the formation of coherent short- and long-term visual object memory representations.

The influence of anticipated monetary incentives on visual working memory performance in healthy younger and older adults.

Motivation exerts substantial control over cognitive functions, including working memory. Although it is well known that both motivational control and working memory processes undergo a progressive decline with ageing, whether and to what extent their interaction is altered in old age remain unexplored. Here we aimed at uncovering the effect of reward anticipation on visual working memory performance in a large cohort of younger and older adults using a delayed-estimation task. We applied a three-component probabilistic model to dissociate the reward effects on three possible sources of error corrupting working memory performance: variability in recall, misbinding of object features and random guessing. The results showed that monetary incentives have a significant beneficial effect on overall working memory recall precision only in the group of younger adults. However, our model-based analysis resulted in significant reward effects on all three working memory component processes, which did not differ between the age groups, suggesting that model-based analysis is more sensitive to small reward-induced modulations in the case of older participants. These findings revealed that monetary incentives have a global boosting effect on working memory performance, which is deteriorated to some extent but still present in healthy older adults.

Predicting Body Mass Index From Structural MRI Brain Images Using a Deep Convolutional Neural Network.

In recent years, deep learning (DL) has become more widespread in the fields of cognitive and clinical neuroimaging. Using deep neural network models to process neuroimaging data is an efficient method to classify brain disorders and identify individuals who are at increased risk of age-related cognitive decline and neurodegenerative disease. Here we investigated, for the first time, whether structural brain imaging and DL can be used for predicting a physical trait that is of significant clinical relevance-the body mass index (BMI) of the individual. We show that individual BMI can be accurately predicted using a deep convolutional neural network (CNN) and a single structural magnetic resonance imaging (MRI) brain scan along with information about age and sex. Localization maps computed for the CNN highlighted several brain structures that strongly contributed to BMI prediction, including the caudate nucleus and the amygdala. Comparison to the results obtained via a standard automatic brain segmentation method revealed that the CNN-based visualization approach yielded complementary evidence regarding the relationship between brain structure and BMI. Taken together, our results imply that predicting BMI from structural brain scans using DL represents a promising approach to investigate the relationship between brain morphological variability and individual differences in body weight and provide a new scope for future investigations regarding the potential clinical utility of brain-predicted BMI.

Transfer learning improves resting-state functional connectivity pattern analysis using convolutional neural networks.

Background: Deep learning is gaining importance in the prediction of cognitive states and brain pathology based on neuroimaging data. Including multiple hidden layers in artificial neural networks enables unprecedented predictive power; however, the proper training of deep neural networks requires thousands of exemplars. Collecting this amount of data is not feasible in typical neuroimaging experiments. A handy solution to this problem, which has largely fallen outside the scope of deep learning applications in neuroimaging, is to repurpose deep networks that have already been trained on large datasets by fine-tuning them to target datasets/tasks with fewer exemplars. Here, we investigated how this method, called transfer learning, can aid age category classification and regression based on brain functional connectivity patterns derived from resting-state functional magnetic resonance imaging. We trained a connectome-convolutional neural network on a larger public dataset and then examined how the knowledge learned can be used effectively to perform these tasks on smaller target datasets collected with a different type of scanner and/or imaging protocol and pre-processing pipeline. Results: Age classification on the target datasets benefitted from transfer learning. Significant improvement (∼9%-13% increase in accuracy) was observed when the convolutional layers' weights were initialized based on the values learned on the public dataset and then fine-tuned to the target datasets. Transfer learning also appeared promising in improving the otherwise poor prediction of chronological age. Conclusions: Transfer learning is a plausible solution to adapt convolutional neural networks to neuroimaging data with few exemplars and different data acquisition and pre-processing protocols.

The electrophysiological correlates of integrated face and body-part perception.

Previous studies have suggested that the human visual system processes faces and bodies holistically-that is, the different body parts are integrated into a unified representation. However, the time course of this integrative process is less known. In the present study, we investigated this issue by recording event-related potentials evoked by a face and two hands presented simultaneously and in different configurations. When the hands were rotated to obtain a biologically implausible configuration, a reduction of the P2 amplitude was observed relative to the condition in which the face and hands were retained in their veridical configuration and were supplemented with visual cues to highlight further the overall body posture. Our results show that the P2 component is sensitive to manipulations affecting the configuration of face and hand stimuli and suggest that the P2 reflects the operation of perceptual mechanisms responsible for the integrated processing of visually presented body parts.

ALTERED BOLD RESPONSE WITHIN THE CORE FACE-PROCESSING NETWORK IN CONGENITAL PROSOPAGNOSIA.

BACKGROUND AND PURPOSE: Congenital prosopagnosia is a life-long disorder of face perception. To study the neural backgrounds of congenital prosopagnosia we measured the blood oxygen level-dependent response of congenital prosopagnosic participants, using functional magnetic resonance imaging. METHODS: We tested three persons of the family (father, daughter and son), having symptoms of congenital prosopagnosia, as well as healthy controls, using combined neuropsychological and functional magnetic resonance imaging methods. To reveal the neural correlates of the impairments, blood oxygen level-dependent responses within the occipito-temporal cortex were measured to faces and nonsense object images in a block-design experiment. RESULTS: Neuropsychological tests demonstrated significant impairments of face perception/recognition in each subject. We found that the activity of the fusiform and occipital face areas as well as of the lateral occipital cortex was significantly reduced in congenital prosopagnosic participants when compared to controls. Analysis of the hemodynamic response function revealed a lower peak response, but also a significantly faster and stronger decay of the blood oxygen level-dependent response in the occipito-temporal areas in congenital prosopagnosic participants when compared to controls. CONCLUSION: Our results emphasize the dysfunction of the core face processing system, as well as the lateral occipital complex, in congenital prosopagnosia. Further, the functional impairment of these areas is signalled best by the altered hemodynamic response function, showing abnormally low initial peak and stronger and faster decay in the later parts of the blood oxygen level-dependent response.

The face evoked steady-state visual potentials are sensitive to the orientation, viewpoint, expression and configuration of the stimuli.

Previous studies demonstrated that the steady-state visual-evoked potential (SSVEP) is reduced to the repetition of the same identity face when compared with the presentation of different identities, suggesting high-level neural adaptation to face identity. Here we investigated whether the SSVEP is sensitive to the orientation, viewpoint, expression and configuration of faces (Experiment 1), and whether adaptation to identity at the level of the SSVEP is robust enough to generalize across these properties (Experiment 2). In Experiment 1, repeating the same identity face with continuously changing orientation, viewpoint or expression evoked a larger SSVEP than the repetition of an unchanged face, presumably reflecting a release of adaptation. A less robust effect was observed in the case of changes affecting face configuration. In Experiment 2, we found a similar release of adaptation for faces with changing orientation, viewpoint and configuration, as there was no difference between the SSVEP for the same and different identity faces. However, we found an adaptation effect for faces with changing expressions, suggesting that face identity coding, as reflected in the SSVEP, is largely independent of the emotion displayed by faces. Taken together, these results imply that the SSVEP taps high-level face representations which abstract away from the changeable aspects of the face and likely incorporate information about face configuration, but which are specific to the orientation and viewpoint of the face.

The background of reduced face specificity of N170 in congenital prosopagnosia.

Congenital prosopagnosia is lifelong face-recognition impairment in the absence of evidence for structural brain damage. To study the neural correlates of congenital prosopagnosia, we measured the face-sensitive N170 component of the event-related potential in three members of the same family (father (56 y), son (25 y) and daughter (22 y)) and in age-matched neurotypical participants (young controls: n = 14; 24.5 y±2.1; old controls: n = 6; 57.3 y±5.4). To compare the face sensitivity of N170 in congenital prosopagnosic and neurotypical participants we measured the event-related potentials for faces and phase-scrambled random noise stimuli. In neurotypicals we found significantly larger N170 amplitude for faces compared to noise stimuli, reflecting normal early face processing. The congenital prosopagnosic participants, by contrast, showed reduced face sensitivity of the N170, and this was due to a larger than normal noise-elicited N170, rather than to a smaller face-elicited N170. Interestingly, single-trial analysis revealed that the lack of face sensitivity in congenital prosopagnosia is related to a larger oscillatory power and phase-locking in the theta frequency-band (4-7 Hz, 130-190 ms) as well as to a lower intertrial jitter of the response latency for the noise stimuli. Altogether, these results suggest that congenital prosopagnosia is due to the deficit of early, structural encoding steps of face perception in filtering between face and non-face stimuli.

Altering second-order configurations reduces the adaptation effects on early face-sensitive event-related potential components.

The spatial distances among the features of a face are commonly referred to as second-order relations, and the coding of these properties is often regarded as a cornerstone in face recognition. Previous studies have provided mixed results regarding whether the N170, a face-sensitive component of the event-related potential, is sensitive to second-order relations. Here we investigated this issue in a gender discrimination paradigm following long-term (5 s) adaptation to normal or vertically stretched male and female faces, considering that the latter manipulation substantially alters the position of the inner facial features. Gender-ambiguous faces were more likely judged to be female following adaptation to a male face and vice versa. This aftereffect was smaller but statistically significant after being adapted to vertically stretched when compared to unstretched adapters. Event-related potential recordings revealed that adaptation effects measured on the amplitude of the N170 show strong modulations by the second-order relations of the adapter: reduced N170 amplitude was observed, however, this reduction was smaller in magnitude after being adapted to stretched when compared to unstretched faces. These findings suggest early face-processing, as reflected in the N170 component, proceeds by extracting the spatial relations of inner facial features.

Phase noise reveals early category-specific modulation of the event-related potentials.

Previous studies have found that the amplitude of the early event-related potential (ERP) components evoked by faces, such as N170 and P2, changes systematically as a function of noise added to the stimuli. This change has been linked to an increased perceptual processing demand and to enhanced difficulty in perceptual decision making about faces. However, to date it has not yet been tested whether noise manipulation affects the neural correlates of decisions about face and non-face stimuli similarly. To this end, we measured the ERPs for faces and cars at three different phase noise levels. Subjects performed the same two-alternative age-discrimination task on stimuli chosen from young-old morphing continua that were created from faces as well as cars and were calibrated to lead to similar performances at each noise-level. Adding phase noise to the stimuli reduced performance and enhanced response latency for the two categories to the same extent. Parallel to that, phase noise reduced the amplitude and prolonged the latency of the face-specific N170 component. The amplitude of the P1 showed category-specific noise dependence: it was enhanced over the right hemisphere for cars and over the left hemisphere for faces as a result of adding phase noise to the stimuli, but remained stable across noise levels for cars over the left and for faces over the right hemisphere. Moreover, noise modulation altered the category-selectivity of the N170, while the P2 ERP component, typically associated with task decision difficulty, was larger for the more noisy stimuli regardless of stimulus category. Our results suggest that the category-specificity of noise-induced modulations of ERP responses starts at around 100 ms post-stimulus.

Face distortion aftereffects evoked by featureless first-order stimulus configurations.

After prolonged exposure to a distorted face with expanded or contracted inner features, a subsequently presented normal face appears distorted toward the opposite direction. This phenomenon, termed as face distortion aftereffect (FDAE), is thought to occur as a result of changes in the mechanisms involved in higher order visual processing. However, the extent to which FDAE is mediated by face-specific configural processing is less known. In the present study, we investigated whether similar aftereffects can be induced by stimuli lacking all the typical characteristics of a human face except for its first-order configural properties. We found a significant FDAE after adaptation to a stimulus consisting of three white dots arranged in a triangular fashion and placed in a gray oval. FDAEs occurred also when the adapting and test stimuli differed in size or when the contrast polarity of the adaptor image was changed. However, the inversion of the adapting image as well as the reduction of its contrast abolished the aftereffect entirely. Taken together, our results suggest that higher-level visual areas, which are involved in the processing of facial configurations, mediate the FDAE. Further, while adaptation seems to be largely invariant to contrast polarity, it appears sensitive to orientation and to lower level manipulations that affect the saliency of the inner features.