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.

PNPLA3 rs738409 risk genotype decouples TyG index from HOMA2-IR and intrahepatic lipid content.

BACKGROUND: Recent reports suggested a different predictive value for TyG index compared to HOMA-IR in coronary artery calcification (CAC) and other atherosclerotic outcomes, despite that both indices are proposed as surrogate markers of insulin resistance. We hypothesized a key role for liver pathology as an explanation and therefore assessed the relationship among the two indices and the intrahepatic lipid content stratified by PNPLA3 rs738409 genotypes as a known non-alcoholic fatty liver disease (NAFLD) genetic risk. METHODS: Thirty-nine women from a prior GDM-genetic study were recalled with PNPLA3 rs738409 CC and GG genotypes for metabolic phenotyping and to assess hepatic triglyceride content (HTGC). 75 g OGTT was performed, fasting lipid, glucose, insulin levels and calculated insulin resistance indices (TyG and HOMA2-IR) were used. HTGC was measured by MR based methods. Mann-Whitney-U, χ2 and for the correlation analysis Spearman rank order tests were applied. RESULTS: The PNPLA3 rs738409 genotype had a significant effect on the direct correlation between the HOMA2-IR and TyG index: the correlation (R = 0.52, p = 0.0054) found in the CC group was completely abolished in those with the GG (NAFLD) risk genotype. In addition, the HOMA2-IR correlated with HTGC in the entire study population (R = 0.69, p < 0.0001) and also separately in both genotypes (CC R = 0.62, p = 0.0006, GG: R = 0.74, p = 0.0058). In contrast, the correlation between TyG index and HTGC was only significant in rs738409 CC genotype group (R = 0.42, p = 0.0284) but not in GG group. A similar pattern was observed in the correlation between TG and HTGC (CC: R = 0.41, p = 0.0335), when the components of the TyG index were separately assessed. CONCLUSIONS: PNPLA3 rs738409 risk genotype completely decoupled the direct correlation between two surrogate markers of insulin resistance: TyG and HOMA2-IR confirming our hypothesis. The liver lipid content increased in parallel with the HOMA2-IR independent of genotype, in contrast to the TyG index where the risk genotype abolished the correlation. This phenomenon seems to be related to the nature of hepatic fat accumulation and to the different concepts establishing the two insulin resistance markers.

Combined effect of pancreatic lipid content and gene variants (TCF7L2, WFS1 and 11BHSD1) on B-cell function in Middle Aged Women in a Post Hoc Analysis.

BACKGROUND: TCF7L2 rs7903146 and PNPLA3 rs738409 gene variants confer the strongest risk for type 2 diabetes mellitus (T2DM) and non-alcoholic fatty liver disease (NAFLD), respectively. Pancreatic triacylglycerol content (PTGC) was reported to have a role in T2DM development. We aimed to assess the correlation between PTGC and hepatic triacylglycerol content (HTGC) stratified by PNPLA3 rs738409 genotype and subsequently interactions between PTGC and gene variants associated with ß-cell dysfunction (TCF7L2, WFS1) and visceral adiposity (11ΒHSD1) on ß-cell function were also tested. METHODS: PTGC and HTGC were assessed using MR in a post-hoc analysis of a genotype-based (PNPLA3 rs738409) recall study of 39 (lipid- and glucose lowering) drug-naïve women. Oral glucose tolerance test, HbA1c, insulin indices, anthropometric data were evaluated. The effect of minor allele carrying of TCF7L2 (rs7903146); WFS1 (rs1801214) and 11ΒHSD1 (rs4844880) variants in combination with PTGC was studied on surrogate markers of ß-cell function. We used Spearman's rank-order, Mann-Whitney-U tests, and linear regression models. RESULTS: PTGC and HTGC values were correlated after stratification by the rs738409 variant (only in CC genotype group R = 0.67, p = 10- 4). PTGC and HbA1c values correlated in the entire study population (R = 0.58, p = 10- 4). Insulin resistance, sensitivity and disposition indices were correlated with PTGC (HOMA2-IR: R = 0.42, p = 0.008; TyG: R = 0.38, p = 0.018; Matsuda: R= - 0.48, p = 0.002; DIbasal: R=-0.33, p = 0.039; ISSI-2: R=-0.35, p = 0.028). Surrogate markers of ß-cell function (HOMA2-B, AUCinsulin/AUCglucose) correlated significantly with PTGC in subjects with the following genotypes rs7903146: CC R = 0.51, p = 0.022; rs18001214: CT + CC R = 0.55, p = 0.013; rs4844880: TA + AA R = 0.56, p = 0.016. The strongest interactions were found between PTGC and TCF7L2 rs7903146 effect on HOMA2-B (p = 0.001) and AUCinsulin/AUCglucose (p = 0.013). CONCLUSIONS: The PNPLA3 rs738409 genotype has a major effect on the correlation between PTGC and HTGC. Furthermore we first report the combined effect of PTGC and individual risk gene variants of TCF7L2, WFS1 and 11ΒHSD1 on ß-cell dysfunction. The correlation between pancreatic lipid accumulation and HbA1c also indicates an important role for the latter pathology.

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.

Efficacy of weight loss intervention can be predicted based on early alterations of fMRI food cue reactivity in the striatum.

Increased fMRI food cue reactivity in obesity, i.e. higher responses to high- vs. low-calorie food images, is a promising marker of the dysregulated brain reward system underlying enhanced susceptibility to obesogenic environmental cues. Recently, it has also been shown that weight loss interventions might affect fMRI food cue reactivity and that there is a close association between the alteration of cue reactivity and the outcome of the intervention. Here we tested whether fMRI food cue reactivity could be used as a marker of diet-induced early changes of neural processing in the striatum that are predictive of the outcome of the weight loss intervention. To this end we investigated the relationship between food cue reactivity in the striatum measured one month after the onset of the weight loss program and weight changes obtained at the end of the six-month intervention. We observed a significant correlation between BMI change measured after six months and early alterations of fMRI food cue reactivity in the striatum, including the bilateral putamen, right pallidum, and left caudate. Our findings provide evidence for diet-induced early alterations of fMRI food cue reactivity in the striatum that can predict the outcome of the weight loss intervention.

Correction to: Reducing task-based fMRI scanning time using simultaneous multislice echo planar imaging.

The original version of this article contained a mistake. The correct Affiliation 2 is Semmelweis University, János Szentágothai PhD School, MR Research Centre, Balassa Street 6, Budapest 1083, Hungary.

Reducing task-based fMRI scanning time using simultaneous multislice echo planar imaging.

PURPOSE: To maintain alertness and to remain motionless during scanning represent a substantial challenge for patients/subjects involved in both clinical and research functional magnetic resonance imaging (fMRI) examinations. Therefore, availability and application of new data acquisition protocols allowing the shortening of scan time without compromising the data quality and statistical power are of major importance. METHODS: Higher order category-selective visual cortical areas were identified individually, and rapid event-related fMRI design was used to compare three different sampling rates (TR = 2000, 1000, and 410 ms, using state-of-the-art simultaneous multislice imaging) and four different scanning lengths to match the statistical power of the traditional scanning methods to high sampling-rate design. RESULTS: The results revealed that ~ 4 min of the scan time with 1 Hz (TR = 1000 ms) sampling rate and ~ 2 min scanning at ~ 2.5 Hz (TR = 410 ms) sampling rate provide similar localization sensitivity and selectivity to that obtained with 11-min session at conventional, 0.5 Hz (TR = 2000 ms) sampling rate. CONCLUSION: Our findings suggest that task-based fMRI examination of clinical population prone to distress such as presurgical mapping experiments might substantially benefit from the reduced (20-40%) scanning time that can be achieved by the application of simultaneous multislice sequences.

Resting State fMRI Functional Connectivity Analysis Using Dynamic Time Warping.

Traditional resting-state network concept is based on calculating linear dependence of spontaneous low frequency fluctuations of the BOLD signals of different brain areas, which assumes temporally stable zero-lag synchrony across regions. However, growing amount of experimental findings suggest that functional connectivity exhibits dynamic changes and a complex time-lag structure, which cannot be captured by the static zero-lag correlation analysis. Here we propose a new approach applying Dynamic Time Warping (DTW) distance to evaluate functional connectivity strength that accounts for non-stationarity and phase-lags between the observed signals. Using simulated fMRI data we found that DTW captures dynamic interactions and it is less sensitive to linearly combined global noise in the data as compared to traditional correlation analysis. We tested our method using resting-state fMRI data from repeated measurements of an individual subject and showed that DTW analysis results in more stable connectivity patterns by reducing the within-subject variability and increasing robustness for preprocessing strategies. Classification results on a public dataset revealed a superior sensitivity of the DTW analysis to group differences by showing that DTW based classifiers outperform the zero-lag correlation and maximal lag cross-correlation based classifiers significantly. Our findings suggest that analysing resting-state functional connectivity using DTW provides an efficient new way for characterizing functional networks.

Neural basis of identity information extraction from noisy face images.

Previous research has made significant progress in identifying the neural basis of the remarkably efficient and seemingly effortless face perception in humans. However, the neural processes that enable the extraction of facial information under challenging conditions when face images are noisy and deteriorated remains poorly understood. Here we investigated the neural processes underlying the extraction of identity information from noisy face images using fMRI. For each participant, we measured (1) face-identity discrimination performance outside the scanner, (2) visual cortical fMRI responses for intact and phase-randomized face stimuli, and (3) intrinsic functional connectivity using resting-state fMRI. Our whole-brain analysis showed that the presence of noise led to reduced and increased fMRI responses in the mid-fusiform gyrus and the lateral occipital cortex, respectively. Furthermore, the noise-induced modulation of the fMRI responses in the right face-selective fusiform face area (FFA) was closely associated with individual differences in the identity discrimination performance of noisy faces: smaller decrease of the fMRI responses was accompanied by better identity discrimination. The results also revealed that the strength of the intrinsic functional connectivity within the visual cortical network composed of bilateral FFA and bilateral object-selective lateral occipital cortex (LOC) predicted the participants' ability to discriminate the identity of noisy face images. These results imply that perception of facial identity in the case of noisy face images is subserved by neural computations within the right FFA as well as a re-entrant processing loop involving bilateral FFA and LOC.

FMRI repetition suppression for voices is modulated by stimulus expectations.

According to predictive coding models of sensory processing, stimulus expectations have a profound effect on sensory cortical responses. This was supported by experimental results, showing that fMRI repetition suppression (fMRI RS) for face stimuli is strongly modulated by the probability of stimulus repetitions throughout the visual cortical processing hierarchy. To test whether processing of voices is also affected by stimulus expectations, here we investigated the effect of repetition probability on fMRI RS in voice-selective cortical areas. Changing ('alt') and identical ('rep') voice stimulus pairs were presented to the listeners in blocks, with a varying probability of alt and rep trials across blocks. We found auditory fMRI RS in the nonprimary voice-selective cortical regions, including the bilateral posterior STS, the right anterior STG and the right IFC, as well as in the IPL. Importantly, fMRI RS effects in all of these areas were strongly modulated by the probability of stimulus repetition: auditory fMRI RS was reduced or not present in blocks with low repetition probability. Our results revealed that auditory fMRI RS in higher-level voice-selective cortical regions is modulated by repetition probabilities and thus suggest that in audition, similarly to the visual modality, processing of sensory information is shaped by stimulus expectation processes.

Dissociating the effect of noise on sensory processing and overall decision difficulty.

It has been proposed that perceptual decision making involves a task-difficulty component, which detects perceptual uncertainty and guides allocation of attentional resources. It is thought to take place immediately after the early extraction of sensory information and is specifically reflected in a positive component of the event related potentials, peaking at ∼ 220 ms after stimulus onset. However, in the previous research, neural processes associated with the monitoring of overall task difficulty were confounded by those associated with the increased sensory processing demands as a result of adding noise to the stimuli. Here we dissociated the effect of phase noise on sensory processing and overall decision difficulty using a face gender categorization task. Task difficulty was manipulated either by adding noise to the stimuli or by adjusting the female/male characteristics of the face images. We found that it is the presence of noise and not the increased overall task difficulty that affects the electrophysiological responses in the first 300 ms following stimulus onset in humans. Furthermore, we also showed that processing of phase-randomized as compared to intact faces is associated with increased fMRI responses in the lateral occipital cortex. These results revealed that noise-induced modulation of the early electrophysiological responses reflects increased visual cortical processing demands and thus failed to provide support for a task-difficulty component taking place between the early sensory processing and the later sensory accumulation stages of perceptual decision making.

Neural mechanisms for voice recognition.

We investigated neural mechanisms that support voice recognition in a training paradigm with fMRI. The same listeners were trained on different weeks to categorize the mid-regions of voice-morph continua as an individual's voice. Stimuli implicitly defined a voice-acoustics space, and training explicitly defined a voice-identity space. The pre-defined centre of the voice category was shifted from the acoustic centre each week in opposite directions, so the same stimuli had different training histories on different tests. Cortical sensitivity to voice similarity appeared over different time-scales and at different representational stages. First, there were short-term adaptation effects: increasing acoustic similarity to the directly preceding stimulus led to haemodynamic response reduction in the middle/posterior STS and in right ventrolateral prefrontal regions. Second, there were longer-term effects: response reduction was found in the orbital/insular cortex for stimuli that were most versus least similar to the acoustic mean of all preceding stimuli, and, in the anterior temporal pole, the deep posterior STS and the amygdala, for stimuli that were most versus least similar to the trained voice-identity category mean. These findings are interpreted as effects of neural sharpening of long-term stored typical acoustic and category-internal values. The analyses also reveal anatomically separable voice representations: one in a voice-acoustics space and one in a voice-identity space. Voice-identity representations flexibly followed the trained identity shift, and listeners with a greater identity effect were more accurate at recognizing familiar voices. Voice recognition is thus supported by neural voice spaces that are organized around flexible 'mean voice' representations.

Electrophysiological correlates of learning-induced modulation of visual motion processing in humans.

Training on a visual task leads to increased perceptual and neural responses to visual features that were attended during training as well as decreased responses to neglected distractor features. However, the time course of these attention-based modulations of neural sensitivity for visual features has not been investigated before. Here we measured event related potentials (ERP) in response to motion stimuli with different coherence levels before and after training on a speed discrimination task requiring object-based attentional selection of one of the two competing motion stimuli. We found that two peaks on the ERP waveform were modulated by the strength of the coherent motion signal; the response amplitude associated with motion directions that were neglected during training was smaller than the response amplitude associated with motion directions that were attended during training. The first peak of motion coherence-dependent modulation of the ERP responses was at 300 ms after stimulus onset and it was most pronounced over the occipitotemporal cortex. The second peak was around 500 ms and was focused over the parietal cortex. A control experiment suggests that the earlier motion coherence-related response modulation reflects the extraction of the coherent motion signal whereas the later peak might index accumulation and readout of motion signals by parietal decision mechanisms. These findings suggest that attention-based learning affects neural responses both at the sensory and decision processing stages.

Semantic classification of diseases in discharge summaries using a context-aware rule-based classifier.

OBJECTIVE Automated and disease-specific classification of textual clinical discharge summaries is of great importance in human life science, as it helps physicians to make medical studies by providing statistically relevant data for analysis. This can be further facilitated if, at the labeling of discharge summaries, semantic labels are also extracted from text, such as whether a given disease is present, absent, questionable in a patient, or is unmentioned in the document. The authors present a classification technique that successfully solves the semantic classification task. DESIGN The authors introduce a context-aware rule-based semantic classification technique for use on clinical discharge summaries. The classification is performed in subsequent steps. First, some misleading parts are removed from the text; then the text is partitioned into positive, negative, and uncertain context segments, then a sequence of binary classifiers is applied to assign the appropriate semantic labels. Measurement For evaluation the authors used the documents of the i2b2 Obesity Challenge and adopted its evaluation measures: F(1)-macro and F(1)-micro for measurements. RESULTS On the two subtasks of the Obesity Challenge (textual and intuitive classification) the system performed very well, and achieved a F(1)-macro = 0.80 for the textual and F(1)-macro = 0.67 for the intuitive tasks, and obtained second place at the textual and first place at the intuitive subtasks of the challenge. CONCLUSIONS The authors show in the paper that a simple rule-based classifier can tackle the semantic classification task more successfully than machine learning techniques, if the training data are limited and some semantic labels are very sparse.

Learning to filter out visual distractors.

When learning to master a visual task in a cluttered natural environment, it is important to optimize the processing of task-relevant information and to efficiently filter out distractors. However, the mechanisms that suppress task-irrelevant information are not well understood. Here we show that training leads to a selective increase in motion coherence detection thresholds for task-irrelevant motion directions that interfered with the processing of task-relevant directions during training. Furthermore, using functional magnetic resonance imaging we found that training attenuated neural responses associated with the task-irrelevant direction compared with the task-relevant direction in the visual cortical areas involved in processing of visual motion. The strongest suppression of functional magnetic resonance imaging responses to task-irrelevant motion information was observed in human area MT+. These findings reveal that perceptual learning leads to the suppression and efficient filtering of task-irrelevant visual information.

Attentional modulation of perceived pain intensity in capsaicin-induced secondary hyperalgesia.

Perceived pain intensity is modulated by attention. However, it is not known that how pain intensity ratings are affected by attention in capsaicin-induced secondary hyperalgesia. Here we show that perceived pain intensity in secondary hyperalgesia is decreased when attention is distracted away from the painful pinprick stimulus with a visual task. Furthermore, it was found that the magnitude of attentional modulation in secondary hyperalgesia is very similar to that of capsaicin-untreated, control condition. Our findings, showing no interaction between capsaicin treatment and attentional modulation suggest that capsaicin-induced secondary hyperalgesia and attention might affect mechanical pain through independent mechanisms.

Flawless visual short-term memory for facial emotional expressions.

Facial emotions are important cues of human social interactions. Emotional expressions are continuously changing and thus should be monitored, memorized, and compared from time to time during social intercourse. However, it is not known how efficiently emotional expressions can be stored in short-term memory. Here we show that emotion discrimination is not impaired when the faces to be compared are separated by several seconds, requiring storage of fine-grained emotion-related information in short-term memory. Likewise, we found no significant effect of increasing the delay between the sample and the test face in the case of facial identity discrimination. Furthermore, a second experiment conducted on a large subject sample (N = 160) revealed flawless short-term memory for both facial emotions and facial identity also when observers performed the discrimination tasks only twice with novel faces. We also performed an fMRI experiment, which confirmed that discrimination of fine-grained emotional expressions in our experimental paradigm involved processing of high-level facial emotional attributes. Significantly stronger fMRI responses were found in a cortical network--including the posterior superior temporal sulcus--that is known to be involved in processing of facial emotional expression during emotion discrimination than during identity discrimination. These findings reveal flawless, high-resolution visual short-term memory for emotional expressions, which might underlie efficient monitoring of continuously changing facial emotions.

Smelling human sex hormone-like compounds affects face gender judgment of men.

Although strong cross-sensory interactions between visual, tactile and auditory modalities have already been shown, we know little about how chemosensory information affects processing in other sensory modalities. We studied whether smelling gender-specific odorous sex hormone-like steroids: 5-alpha-androgenst-16-en-3-one (androgen) or oestra-1, 3, 5 (10), 16-tetraen-3-ol (estrogen) can bias face gender discrimination. We found that, as a result of inhalation of androgen, men perceive faces to be more masculine as compared to when they are exposed to estrogen. Our results provide evidence for specific cross-sensory effects of the gender-specific chemosensory cues on the categorization of visual face gender.

Analysis of multidimensional neural activity via CNN-UM.

In this paper we show that the Cellular Nonlinear Network Universal Machine (CNN-UM) is an excellent tool for analyzing time series of multidimensional binary signals. The developed algorithm is dedicated to process electrophysiological multi-neuron recordings: our aim is to find specific multidimensional activity patterns, which may reflect higher order functional cell-assemblies. The analysis consists of two parts: first, the occurrences of different patterns are counted, then the statistical significance of each occurrence frequency is calculated separately.