Labeling Noncontrast Head CT Reports for Common Findings Using Natural Language Processing.

BACKGROUND AND PURPOSE: Prioritizing reading of noncontrast head CT examinations through an automated triage system may improve time to care for patients with acute neuroradiologic findings. We present a natural language-processing approach for labeling findings in noncontrast head CT reports, which permits creation of a large, labeled dataset of head CT images for development of emergent-finding detection and reading-prioritization algorithms. MATERIALS AND METHODS: In this retrospective study, 1002 clinical radiology reports from noncontrast head CTs collected between 2008 and 2013 were manually labeled across 12 common neuroradiologic finding categories. Each report was then encoded using an n-gram model of unigrams, bigrams, and trigrams. A logistic regression model was then trained to label each report for every common finding. Models were trained and assessed using a combination of L2 regularization and 5-fold cross-validation. RESULTS: Model performance was strongest for the fracture, hemorrhage, herniation, mass effect, pneumocephalus, postoperative status, and volume loss models in which the area under the receiver operating characteristic curve exceeded 0.95. Performance was relatively weaker for the edema, hydrocephalus, infarct, tumor, and white-matter disease models (area under the receiver operating characteristic curve > 0.85). Analysis of coefficients revealed finding-specific words among the top coefficients in each model. Class output probabilities were found to be a useful indicator of predictive error on individual report examples in higher-performing models. CONCLUSIONS: Combining logistic regression with n-gram encoding is a robust approach to labeling common findings in noncontrast head CT reports.

Brain Network Disruption in Whiplash.

BACKGROUND AND PURPOSE: Whiplash-associated disorders frequently develop following motor vehicle collisions and often involve a range of cognitive and affective symptoms, though the neural correlates of the disorder are largely unknown. In this study, a sample of participants with chronic whiplash injuries were scanned by using resting-state fMRI to assess brain network changes associated with long-term outcome metrics. MATERIALS AND METHODS: Resting-state fMRI was collected for 23 participants and used to calculate network modularity, a quantitative measure of the functional segregation of brain region communities. This was analyzed for associations with whiplash-associated disorder outcome metrics, including scales of neck disability, traumatic distress, depression, and pain. In addition to these clinical scales, cervical muscle fat infiltration was quantified by using Dixon fat-water imaging, which has shown promise as a biomarker for assessing disorder severity and predicting recovery in chronic whiplash. RESULTS: An association was found between brain network structure and muscle fat infiltration, wherein lower network modularity was associated with larger amounts of cervical muscle fat infiltration after controlling for age, sex, body mass index, and scan motion (t = -4.02, partial R 2 = 0.49, P < .001). CONCLUSIONS: This work contributes to the existing whiplash literature by examining a sample of participants with whiplash-associated disorder by using resting-state fMRI. Less modular brain networks were found to be associated with greater amounts of cervical muscle fat infiltration suggesting a connection between disorder severity and neurologic changes, and a potential role for neuroimaging in understanding the pathophysiology of chronic whiplash-associated disorders.

Ambulatory function in motor incomplete spinal cord injury: a magnetic resonance imaging study of spinal cord edema and lower extremity muscle morphometry.

STUDY DESIGN: This research utilized a cross-sectional design. OBJECTIVES: Spinal cord edema length has been measured with T2-weighted sagittal MRI to predict motor recovery following spinal cord injury. The purpose of our study was to establish the correlational value of axial spinal cord edema using T2-weighted MRI. We hypothesized a direct relationship between the size of damage on axial MRI and walking ability, motor function and distal muscle changes seen in motor incomplete spinal cord injury (iSCI). SETTING: University-based laboratory in Chicago, IL, USA. METHODS: Fourteen participants with iSCI took part in the study. Spinal cord axial damage ratios were assessed using axial T2-weighted MRI. Walking ability was investigated using the 6-min walk test and daily stride counts. Maximum plantarflexion torque was quantified using isometric dynomometry. Muscle fat infiltration (MFI) and relative muscle cross-sectional area (rmCSA) were quantified using fat/water separation magnetic resonance imaging. RESULTS: Damage ratios were negatively correlated with distance walked in 6 min, average daily strides and maximum plantarflexion torque, and a negative linear trend was found between damage ratios and lower leg rmCSA. While damage ratios were not significantly correlated with MFI, we found significantly higher MFI in the wheelchair user participant group compared to community walkers. CONCLUSIONS: Damage ratios may be useful in prognosis of motor recovery in spinal cord injury. The results warrant a large multi-site research study to investigate the value of high-resolution axial T2-weighted imaging to predict walking recovery following motor incomplete spinal cord injury.

Hemodynamic response function in patients with stroke-induced aphasia: implications for fMRI data analysis.

Functional MRI is based on changes in cerebral microvasculature triggered by increased neuronal oxidative metabolism. This change in blood flow follows a pattern known as the hemodynamic response function (HRF), which typically peaks 4-6 s following stimulus delivery. However, in the presence of cerebrovascular disease the HRF may not follow this normal pattern, due to either the temporal signal to noise (tSNR) ratio or delays in the HRF, which may result in misinterpretation or underestimation of fMRI signal. The present study examined the HRF and SNR in five individuals with aphasia resulting from stroke and four unimpaired participants using a lexical decision task and a long trial event-related design. T1-weighted images were acquired using an MP-RAGE sequence and BOLD T2*-weighted images were acquired using Echo Planar Imaging to measure time to peak (TTP) in the HRF. Data were analyzed using Brain Voyager in four anatomic regions known to be involved in language processing: Broca's area and the posterior perisylvian network (PPN) (including Wernicke's area, the angular and supramarginal gyri) and right hemisphere homologues of these regions. The occipital area also was examined as a control region. Analyses showed that the TTP in three out of five patients in the left perisylvian area was increased significantly as compared to normal individuals and the left primary visual cortex in the same patients. In two other patients no significant delays were detected. We also found that the SNR for BOLD signal detection may by insufficient in damaged areas. These findings indicate that obtaining physiologic (TTP) and quality assurance (tSNR) information is essential for studying activation patterns in brain-damaged patients in order to avoid errors in interpretation of the data. An example of one such misinterpretation and the need for alternative data analysis strategies is discussed.

Neural correlates of rule-based and information-integration visual category learning.

An emerging theory of the neurobiology of category learning postulates that there are separate neural systems supporting the learning of categories based on verbalizeable rules (RB) or through implicit information integration (II). The medial temporal lobe (MTL) is thought to play a crucial role in successful RB categorization, whereas the posterior regions of the caudate are hypothesized to support II categorization. Functional neuroimaging was used to assess activity in these systems during category-learning tasks with category structures designed to afford either RB or II learning. Successful RB categorization was associated with relatively increased activity in the anterior MTL. Successful II categorization was associated with increased activity in the caudate body. The dissociation observed with neuroimaging is consistent with the roles of these systems in memory and dissociations reported in patient populations. Convergent evidence from these approaches consistently reinforces the idea of multiple neural systems supporting category learning.

The posterior cingulate and medial prefrontal cortex mediate the anticipatory allocation of spatial attention.

The purpose of this study was to identify brain regions underlying internally generated anticipatory biases toward locations where significant events are expected to occur. Subjects fixated centrally and responded to peripheral targets preceded by a spatially valid (predictive), invalid (misleading), or neutral central cue while undergoing fMRI scanning. In some validly cued trials, reaction time was significantly shorter than in trials with neutral cues, indicating that the cue had successfully induced a spatial redistribution of motivational valence, manifested as expectancy. The largest cue benefits led to selectively greater activations within the posterior cingulate and medial prefrontal cortex. These two areas thus appear to establish a neural interface between attention and motivation. An inverse relationship to cue benefit was seen in the parietal cortex, suggesting that spatial expectancy may entail the inhibition of attention-related areas to reduce distractibility by events at irrelevant locations.

Neural correlates of artificial grammar learning.

Artificial grammar learning (AGL) is a form of nondeclarative memory that involves the nonconscious acquisition of abstract rules. While data from amnesic patients indicate that AGL does not depend on the medial temporal lobe, the neural basis of this type of memory is unknown and was therefore examined using event-related fMRI. Prior to scanning, participants studied letter strings constructed according to an artificial grammar. Participants then made grammaticality judgments about novel grammatical and nongrammatical strings while fMRI data were collected. The participants successfully acquired knowledge of the grammar, as evidenced by correct identification of the grammatical letter strings (57.4% correct; SE 1.9). During grammaticality judgments, widespread increases in activity were observed throughout the occipital, posterior temporal, parietal, and prefrontal cortical areas, reflecting the cognitive demands of the task. More specific analyses contrasting grammatical and nongrammatical strings identified greater activity in left superior occipital cortex and the right fusiform gyrus for grammatical stimuli. Increased activity was also observed in the left superior occipital and left angular gyrus for correct responses compared to incorrect. Comparing activity during grammaticality judgments versus a matched recognition control task again identified greater activation in the left angular gyrus. The network of areas exhibiting increased activity for grammatical stimuli appears to have more in common with studies examining word-form processing or mental calculation than the fluency effects previously reported for nondeclarative memory tasks such as priming and visual categorization. These results suggest that a novel nondeclarative memory mechanism supporting AGL exists in the left superior occipital and inferior parietal cortex.

Electromyography as a recording system for eyeblink conditioning with functional magnetic resonance imaging.

This study was designed to develop a suitable method of recording eyeblink responses while conducting functional magnetic resonance imaging (fMRI). Given the complexity of this behavioral setup outside of the magnet, this study sought to adapt and further optimize an approach to eyeblink conditioning that would be suitable for conducting event-related fMRI experiments. This method involved the acquisition of electromyographic (EMG) signals from the orbicularis oculi of the right eye, which were subsequently amplified and converted into an optical signal outside of the head coil. This optical signal was converted back into an electrical signal once outside the magnet room. Electromyography (EMG)-detected eyeblinks were used to measure responses in a delay eyeblink conditioning paradigm. Our results indicate that: (1) electromyography is a sensitive method for the detection of eyeblinks during fMRI; (2) minimal interactions or artifacts of the EMG signal were created from the magnetic resonance pulse sequence; and (3) no electromyography-related artifacts were detected in the magnetic resonance images. Furthermore, an analysis of the functional data showed areas of activation that have previously been shown in positron emission tomography studies of human eyeblink conditioning. Our results support the strength of this behavioral setup as a suitable method to be used in association with fMRI.

Impact of signal-to-noise on functional MRI of the human amygdala.

The impact of signal-to-noise (SNR) on fMRI of the amygdala was investigated during a picture encoding task. The SNR value required to observe reliable activation was determined by computer simulations. Blood oxygen level-dependent (BOLD) sensitivity maps were generated to indicate brain regions with sufficient SNR to test the statistical hypotheses. The results showed that the medial aspect of the amygdala had insufficient SNR to detect a 1% peak BOLD signal change for a t-test comparison in a majority of subjects. None of these subjects showed activation in regions with unacceptable SNR values, indicating a low false positive rate. Furthermore, hemispheric asymmetries in the BOLD sensitivity maps mirrored asymmetries in the activation patterns. Impoverished SNR was also found in the basal forebrain and orbitofrontal cortex. These findings emphasize the importance of considering SNR when interpreting fMRI results in the limbic forebrain.

Functional specificity of superior parietal mediation of spatial shifting.

Using event-related functional magnetic resonance imaging (fMRI) we determined how brain activity changes when an attended target shifts its location. In the main experiment, a white square could appear at 10 possible eccentricities along the horizontal meridian. It remained on the screen for a variable period of time and then changed location. At any time the stimulus could dim briefly. Subjects had to press a button when the stimulus dimmed. In order to perform this task attention had to be locked onto the target and shift with it. Half of the runs were performed overtly and half covertly. The event of interest consisted of the shift in the location of the attentional target. The state of maintained attention occurring in between the shifts constituted the baseline. The superior parietal gyrus was activated bilaterally in response to attentional shifts. No other area showed a significant response to shifting. On the left side the amplitude of the superior parietal response correlated positively with the distance of the shift. On the right side a significant correlation was present only for overt shifts. In a separate experiment we compared the maintaining of attention at a single spatial location to passive fixation: the frontal eye fields, anterior cingulate, right dorsolateral prefrontal cortex, and inferior parietal lobule were significantly activated, indicating that the absence of a shift-related response in these areas in the main experiment was due to the fact that they were equally activated by maintaining and shifting attention. The response to spatial shifts and the correlation with the distance between the original and the new location points to a specific role of the superior parietal gyrus in shifting the locus of spatial attention.

Location- or feature-based targeting of peripheral attention.

Using event-related fMRI we determined the differential effects of feature- versus location-based cues for directing peripheral attention. Pairs of same-color targets appeared on the left and on the right. A predictive cue indicated whether the subsequent targeting of attention would be based on location (left versus right) or color (red versus blue). Subjects had to press a button when the relevant pair of targets (in the cued side of space or of the cued color) were identical in shape. The feature-based cue thus also led to a "global" expectancy of targets on either side of space whereas the location-based cue elicited a more "focal" expectancy limited to one side of space. The right inferior parietal lobule was more active when attention was targeted on the basis of location than of color. There was no difference between left-sided or right-sided attention in this region, indicating that it mediated the targeting to both sides of space. These results show that the right inferior parietal cortex plays a relatively selective role in mediating location-based and spatially focal modes of attentional deployment. Its relatively equal activation for leftward and rightward attentional shifts is also consistent with models of right hemispheric dominance of spatial attention.

Hunger selectively modulates corticolimbic activation to food stimuli in humans.

Functional magnetic resonance imaging (fMRI) was used to determine whether visual responses to food in the human amygdala and related corticolimbic structures would be selectively altered by changes in states of hunger. Participants viewed images of motivationally relevant (food) and motivationally irrelevant (tool) objects while undergoing fMRI in alternately hungry and satiated conditions. Food-related visual stimuli elicited greater responses in the amygdala, parahippocampal gyrus. and anterior fusiform gyrus when participants were in a hungry state relative to a satiated state. The state-dependent activation of these brain structures did not generalize to the motivationally irrelevant objects. These results support the hypothesis that the amygdala and associated inferotemporal regions are involved in the integration of subjective interoceptive states with relevant sensory cues processed along the ventral visual stream.

Heterogeneity of cingulate contributions to spatial attention.

Functional magnetic resonance imaging was used to investigate activation patterns within the cingulate region during tasks based on spatial attention. Subjects were asked to detect targets which appeared either at the site indicated by a cue or on the opposite side. A "cue effect" was identified by the presence of shorter reaction times to validly than invalidly cued targets, showing that an anticipatory bias had been generated in the direction of the cue. Target detection accuracy was consistently above 90% although cue effects and reaction times displayed substantial variations, from one task session to another. Activation within the anterior cingulate region was seen in 16 of the 26 sessions but showed no correlation with reaction time. Posterior cingulate activation was seen in only 6 of the 26 sessions. However, a random effects analysis showed that the task-related signal change in this region was strongly correlated with the speed of target detection. A post hoc analysis indicated that this correlation was significant only when cue effects were present. No other part of the cerebral cortex displayed significant correlations with reaction times or cue effects. These results suggest that the cingulate component of the attentional network has at least two functionally segregated sectors, an anterior one in BA 24/32 and a posterior cingulo-retrosplenial one in BA 23/29/30. The posterior sector appears to be associated with the speed of detecting spatial targets, especially when attention is under the influence of a cue-induced anticipatory bias. The anterior cingulate focus did not display such a relationship in our tasks and is likely to mediate other aspects of attentional deployment such as performance monitoring, response selection or target identification.

The development of specialized brain systems in reading and oral-language.

Functional magnetic resonance imaging (fMRI) was used to examine differences between children (9-12 years) and adults (21-31 years) in the distribution of brain activation during word processing. Orthographic, phonologic, semantic and syntactic tasks were used in both the auditory and visual modalities. Our two principal results were consistent with the hypothesis that development is characterized by increasing specialization. Our first analysis compared activation in children versus adults separately for each modality. Adults showed more activation than children in the unimodal visual areas of middle temporal gyrus and fusiform gyrus for processing written word forms and in the unimodal auditory areas of superior temporal gyrus for processing spoken word forms. Children showed more activation than adults for written word forms in posterior heteromodal regions (Wernicke's area), presumably for the integration of orthographic and phonologic word forms. Our second analysis compared activation in the visual versus auditory modality separately for children and adults. Children showed primarily overlap of activation in brain regions for the visual and auditory tasks. Adults showed selective activation in the unimodal auditory areas of superior temporal gyrus when processing spoken word forms and selective activation in the unimodal visual areas of middle temporal gyrus and fusiform gyrus when processing written word forms.

Impact of signal-to-noise on functional MRI.

Functional magnetic resonance imaging (fMRI) has recently been adopted as an investigational tool in the field of neuroscience. The signal changes induced by brain activations are small ( approximately 1-2%) at 1.5T. Therefore, the signal-to-noise ratio (SNR) of the time series used to calculate the functional maps is critical. In this study, the minimum SNR required to detect an expected MR signal change is determined using computer simulations for typical fMRI experimental designs. These SNR results are independent of manufacturer, site environment, field strength, coil type, or type of cognitive task used. Sensitivity maps depicting the minimum detectable signal change can be constructed. These sensitivity maps can be used as a mask of the activation map to help remove false positive activations as well as identify regions of the brain where it is not possible to confidently reject the null hypothesis due to a low SNR.

Hybrid technique for dynamic imaging.

A number of data acquisition strategies have been introduced to speed up the image acquisition in dynamic settings. One such technique is the keyhole approach, which is based on reducing k-space coverage, and consequently the spatial resolution, of the dynamic information. Another is based on reducing the field of view of the dynamic information. These two techniques are complementary in that one reduces the field of view in k-space and the other does so in the spatial domain. A hybrid approach which combines the two is described in this study. In numerical simulations and experimental studies, this hybrid approach more accurately depicts the signal changes, outperforming the two techniques from which it is derived. Magn Reson Med 44:51-55, 2000.

Real-time monitoring of eye movements using infrared video-oculography during functional magnetic resonance imaging of the frontal eye fields.

Monitoring eye movements is a critical aspect of experimental design for studies of spatial attention and visual perception. However, obtaining online eye-movement recordings has been technologically difficult during functional magnetic resonance (MR) imaging studies. Previous approaches to monitoring eye movements either have distorted the MR images or have shown MR-related interference in the recordings. We report a technique using long-range infrared video-oculography to record eye movements without causing artifacts in the MR images. Analysis of the MR signal from a phantom obtained with the eye-movement equipment turned on or off confirmed the absence of significant additional noise in the MR time series. Eye movements of three subjects were monitored while they performed tasks of covert and overt shifts of spatial attention. Activation of the frontal eye fields during the covert task was seen even when the eye-movement recordings demonstrated no significant difference in saccadic eye movements between the baseline and the active conditions.

Neuroanatomic overlap of working memory and spatial attention networks: a functional MRI comparison within subjects.

Frontal and posterior parietal activations have been reported in numerous studies of working memory and visuospatial attention. To directly compare the brain regions engaged by these two cognitive functions, the same set of subjects consecutively participated in tasks of working memory and spatial attention while undergoing functional MRI (fMRI). The working memory task required the subject to maintain an on-line representation of foveally displayed letters against a background of distracters. The spatial attention task required the subject to shift visual attention covertly in response to a centrally presented directional cue. The spatial attention task had no working memory requirement, and the working memory task had no covert spatial attention requirement. Subjects' ability to maintain central fixation was confirmed outside the MRI scanner using infrared oculography. According to cognitive conjunction analysis, the set of activations common to both tasks included the intraparietal sulcus, ventral precentral sulcus, supplementary motor area, frontal eye fields, thalamus, cerebellum, left temporal neocortex, and right insula. Double-subtraction analyses yielded additional activations attributable to verbal working memory in premotor cortex, left inferior prefrontal cortex, right inferior parietal lobule, precuneus, and right cerebellum. Additional activations attributable to covert spatial attention included the occipitotemporal junction and extrastriate cortex. The use of two different tasks in the same set of subjects allowed us to provide an unequivocal demonstration that the neural networks subserving spatial attention and working memory intersect at several frontoparietal sites. These findings support the view that major cognitive domains are represented by partially overlapping large-scale neural networks. The presence of this overlap also suggests that spatial attention and working memory share common cognitive features related to the dynamic shifting of attentional resources.

Relationship of MRI delayed contrast enhancement to irreversible injury, infarct age, and contractile function.

BACKGROUND: Contrast MRI enhancement patterns in several pathophysiologies resulting from ischemic myocardial injury are controversial or have not been investigated. We compared contrast enhancement in acute infarction (AI), after severe but reversible ischemic injury (RII), and in chronic infarction. METHODS AND RESULTS: In dogs, a large coronary artery was occluded to study AI and/or chronic infarction (n = 18), and a second coronary artery was chronically instrumented with a reversible hydraulic occluder and Doppler flowmeter to study RII (n = 8). At 3 days after surgery, cine MRI revealed reduced wall thickening in AI (5+/-6% versus 33+/-6% in normal, P<0.001). In RII, wall thickening before, during, and after inflation of the occluder for 15 minutes was 35+/-5%, 1+/-8%, and 21+/-10% and Doppler flow was 19.8+/-5.3, 0.2+/-0.5, and 56.3+/-17.7 (peak hyperemia) cm/s, respectively, confirming occlusion, transient ischemia, and reperfusion. Gd-DTPA-enhanced MR images acquired 30 minutes after contrast revealed hyperenhancement of AI (294+/-96% of normal, P<0.001) but not of RII (98+/-6% of normal, P = NS). Eight weeks later, the chronically infarcted region again hyperenhanced (253+/-54% of normal, n = 8, P<0.001). High-resolution (0.5 x 0.5 x 0.5 mm) ex vivo MRI demonstrated that the spatial extent of hyperenhancement was the same as the spatial extent of myocyte necrosis with and without reperfusion at 1 day (R = 0.99, P<0.001) and 3 days (R = 0.99, P<0.001) and collagenous scar at 8 weeks (R = 0.97, P<0.001). CONCLUSIONS: In the pathophysiologies investigated, contrast MRI distinguishes between reversible and irreversible ischemic injury independent of wall motion and infarct age.