Delineation and quantification of normal and abnormal brain tissues on Magnetic Resonance Images is fundamental to the diagnosis and longitudinal assessment of neurological diseases. Here we sought to develop a convolutional neural network for automated multiclass tissue segmentation of brain MRIs that was robust at typical clinical resolutions and in the presence of a variety of lesions. We trained a 3D U-Net for full brain multiclass tissue segmentation from a prior atlas-based segmentation method on an internal dataset that consisted of 558 clinical T1-weighted brain MRIs (453/52/53; training/validation/test) of patients with one of 50 different diagnostic entities (n = 362) or with a normal brain MRI (n = 196). We then used transfer learning to refine our model on an external dataset that consisted of 7 patients with hand-labeled tissue types. We evaluated the tissue-wise and intra-lesion performance with different loss functions and spatial prior information in the validation set and applied the best performing model to the internal and external test sets. The network achieved an average overall Dice score of 0.87 and volume similarity of 0.97 in the internal test set. Further, the network achieved a median intra-lesion tissue segmentation accuracy of 0.85 inside lesions within white matter and 0.61 inside lesions within gray matter. After transfer learning, the network achieved an average overall Dice score of 0.77 and volume similarity of 0.96 in the external dataset compared to human raters. The network had equivalent or better performance than the original atlas-based method on which it was trained across all metrics and produced segmentations in a hundredth of the time. We anticipate that this pipeline will be a useful tool for clinical decision support and quantitative analysis of clinical brain MRIs in the presence of lesions.

Cumulating evidence from epidemiologic studies implicates cardiovascular health and cerebrovascular function in several brain diseases in late life. We examined vascular risk factors with respect to a cerebrovascular measure of brain functioning in subjects in mid-life, which could represent a marker of brain changes in later life. Breath-hold functional MRI (fMRI) was performed in 541 women and men (mean age 50.4 years) from the Coronary Artery Risk Development in Young Adults (CARDIA) Brain MRI sub-study. Cerebrovascular reactivity (CVR) was quantified as percentage change in blood-oxygen level dependent (BOLD) signal in activated voxels, which was mapped to a common brain template and log-transformed. Mean CVR was calculated for anatomic regions underlying the default-mode network (DMN) - a network implicated in AD and other brain disorders - in addition to areas considered to be relatively spared in the disease (e.g. occipital lobe), which were utilized as reference regions. Mean CVR was significantly reduced in the posterior cingulate/precuneus (β=-0.063, 95% CI: -0.106, -0.020), anterior cingulate (β=-0.055, 95% CI: -0.101, -0.010), and medial frontal lobe (β=-0.050, 95% CI: -0.092, -0.008) relative to mean CVR in the occipital lobe, after adjustment for age, sex, race, education, and smoking status, in subjects with pre-hypertension/hypertension compared to normotensive subjects. By contrast, mean CVR was lower, but not significantly, in the inferior parietal lobe (β=-0.024, 95% CI: -0.062, 0.014) and the hippocampus (β=-0.006, 95% CI: -0.062, 0.050) relative to mean CVR in the occipital lobe. Similar results were observed in subjects with diabetes and dyslipidemia compared to those without these conditions, though the differences were non-significant. Reduced CVR may represent diminished vascular functionality for the DMN for individuals with prehypertension/hypertension in mid-life, and may serve as a preclinical marker for brain dysfunction in later life.

Objective

We examined imaging surrogates of white matter microstructural abnormalities which may precede white matter lesions (WML) and represent a relevant marker of cerebrovascular injury in adults in midlife.

Methods

In 698 community-dwelling adults (mean age 50 years ±3.5 SD) from the Coronary Artery Risk Development in Young Adults (CARDIA) Brain MRI sub-study, WML were identified on structural MR and fractional anisotropy (FA), representing WM microstructural integrity, was derived using Diffusion Tensor Imaging. FA and WML maps were overlaid on a parcellated T1-template, based on an expert-delineated brain atlas, which included 42 WM tract ROIs. Analyses occurred in stages: 1) WML were quantified for the different tracts (i.e., frequency, volume, volume relative to tract size); 2) the interdependence of FA in normal appearing WM (NAWM) and WML was examined across tracts; 3) associations of NAWM FA and hypertension status were assessed controlling for WML volume. In the latter analysis, both overall hypertension (i.e. hypertension vs. normotension and prehypertension vs. normotension) and hypertension categorized by antihypertensive treatment status (yes/no) and blood pressure control (e.g., diastolic <90 mmHg, systolic <140 mmHg), were assessed.

Results

WML were widely distributed across different WM tracts, however, WML volume was small. Mean NAWM FA was lower in participants with vs. participants without WML in given tracts. Hypertension was significantly associated with lower mean NAWM FA globally across tracts, both before and after adjustment for WML volume. Moreover, the magnitude of this association differed by treatment status and the level of control of the hypertension.

Conclusions

In middle-aged adults, NAWM FA could represent a relevant marker of cerebrovascular injury when WML are minimally present.

Background Although artificial intelligence (AI) shows promise across many aspects of radiology, the use of AI to create differential diagnoses for rare and common diseases at brain MRI has not been demonstrated. Purpose To evaluate an AI system for generation of differential diagnoses at brain MRI compared with radiologists. Materials and Methods This retrospective study tested performance of an AI system for probabilistic diagnosis in patients with 19 common and rare diagnoses at brain MRI acquired between January 2008 and January 2018. The AI system combines data-driven and domain-expertise methodologies, including deep learning and Bayesian networks. First, lesions were detected by using deep learning. Then, 18 quantitative imaging features were extracted by using atlas-based coregistration and segmentation. Third, these image features were combined with five clinical features by using Bayesian inference to develop probability-ranked differential diagnoses. Quantitative feature extraction algorithms and conditional probabilities were fine-tuned on a training set of 86 patients (mean age, 49 years ± 16 [standard deviation]; 53 women). Accuracy was compared with radiology residents, general radiologists, neuroradiology fellows, and academic neuroradiologists by using accuracy of top one, top two, and top three differential diagnoses in 92 independent test set patients (mean age, 47 years ± 18; 52 women). Results For accuracy of top three differential diagnoses, the AI system (91% correct) performed similarly to academic neuroradiologists (86% correct; P = .20), and better than radiology residents (56%; P < .001), general radiologists (57%; P < .001), and neuroradiology fellows (77%; P = .003). The performance of the AI system was not affected by disease prevalence (93% accuracy for common vs 85% for rare diseases; P = .26). Radiologists were more accurate at diagnosing common versus rare diagnoses (78% vs 47% across all radiologists; P < .001). Conclusion An artificial intelligence system for brain MRI approached overall top one, top two, and top three differential diagnoses accuracy of neuroradiologists and exceeded that of less-specialized radiologists. © RSNA, 2020 Online supplemental material is available for this article. See also the editorial by Zaharchuk in this issue.

Objective

To identify early changes in brain structure and function that are associated with cardiovascular risk factors (CVRF).

Design

Cross-sectional brain Magnetic Resonance I (MRI) study.

Setting

Community based cohort in three U.S. sites.

Participants

A Caucasian and African-American sub-sample (n= 680; mean age 50.3 yrs) attending the 25 year follow-up exam of the Coronary Artery Risk Development in Young Adults Study.

Primary and secondary outcomes

3T brain MR images processed for quantitative estimates of: total brain (TBV) and abnormal white matter (AWM) volume; white matter fractional anisotropy (WM-FA); and gray matter cerebral blood flow (GM-CBF). Total intracranial volume is TBV plus cerebral spinal fluid (TICV). A Global Cognitive Function (GCF) score was derived from tests of speed, memory and executive function.

Results

Adjusting for TICV and demographic factors, current smoking was significantly associated with lower GM-CBF and TBV, and more AWM (all <0.05); SA with lower GM-CBF, WM-FA and TBV (p=0.01); increasing BMI with decreasing GM-CBF (p<0003); hypertension with lower GM-CBF, WM-FA, and TBV and higher AWM (all <0.05); and diabetes with lower TBV (p=0.007). The GCS was lower as TBV decreased, AWM increased, and WM-FA (all p<0.01).

Conclusion

In middle age adults, CVRF are associated with brain health, reflected in MRI measures of structure and perfusion, and cognitive functioning. These findings suggest markers of mid-life cardiovascular and brain health should be considered as indication for early intervention and future risk of late-life cerebrovascular disease and dementia.

Objective

To study how type 2 diabetes adversely affects brain volumes, changes in volume, and cognitive function.

Research design and methods

Regional brain volumes and ischemic lesion volumes in 1,366 women, aged 72-89 years, were measured with structural brain magnetic resonance imaging (MRI). Repeat scans were collected an average of 4.7 years later in 698 women. Cross-sectional differences and changes with time between women with and without diabetes were compared. Relationships that cognitive function test scores had with these measures and diabetes were examined.

Results

The 145 women with diabetes (10.6%) at the first MRI had smaller total brain volumes (0.6% less; P = 0.05) and smaller gray matter volumes (1.5% less; P = 0.01) but not white matter volumes, both overall and within major lobes. They also had larger ischemic lesion volumes (21.8% greater; P = 0.02), both overall and in gray matter (27.5% greater; P = 0.06), in white matter (18.8% greater; P = 0.02), and across major lobes. Overall, women with diabetes had slightly (nonsignificant) greater loss of total brain volumes (3.02 cc; P = 0.11) and significant increases in total ischemic lesion volumes (9.7% more; P = 0.05) with time relative to those without diabetes. Diabetes was associated with lower scores in global cognitive function and its subdomains. These relative deficits were only partially accounted for by brain volumes and risk factors for cognitive deficits.

Conclusions

Diabetes is associated with smaller brain volumes in gray but not white matter and increasing ischemic lesion volumes throughout the brain. These markers are associated with but do not fully account for diabetes-related deficits in cognitive function.

Background

Clinical and epidemiological studies of older persons have implicated clusterin in Alzheimer's disease (AD) pathogenesis. In the context of identifying early biomarkers of risk, we examined associations of plasma clusterin and characteristics of AD in middle-aged individuals from the community.

Materials and methods

Subjects were 639 cognitively normal individuals (mean age 50 ± 3.5) from the Coronary Artery Risk Development in Young Adults (CARDIA) Brain MRI sub-study. Clusterin was quantified using ELISA (mean 255± 31 ng/ml). Associations were assessed between clusterin and volumes of brain regions known to atrophy in early AD, including entorhinal cortex (ECV), hippocampus (HV), and medial temporal lobe (MTLV) volumes (cm3). Total brain volume (TBV) and volumes of structures affected in later AD were examined for comparison.

Results

In multivariable models, higher clusterin had a negative non-linear association with ECV (combined left and right hemispheres), and this association was influenced by the highest clusterin levels. Compared to mean clusterin, 1 and 2 standard deviation (SD) level increases in clusterin were associated with -2.1% (95% CI: -3.3,-0.9) and -7.3% (95% CI: -11.3,-3.3) lower ECV, respectively. Similar relationships were observed between clusterin and HV, although the relationship was stronger for left-side HV than the right-side. However, the association was not significant after adjusting for covariates. Negative non-linear associations between clusterin and MTLV were strongest for the left side: compared to mean clusterin, 1 and 2 SD level increases in clusterin were associated with -0.9% (95% CI: -1.9, 0.1) and -3.7% (95% CI: -7.1, -0.3) lower MTLV. There were no significant associations between clusterin and brain structures affected in later AD.

Conclusions

In middle-aged adults unselected for AD, plasma clusterin was associated with lower volume of the entorhinal cortex, an area that atrophies early in AD. Clusterin could be informative as part of a multi-component preclinical marker for AD.

Objective

We hypothesized that greater cardiorespiratory fitness is associated with lower odds of having unfavorable brain MRI findings.

Methods

We studied 565 healthy, middle-aged, black and white men and women in the CARDIA (Coronary Artery Risk Development in Young Adults) Study. The fitness measure was symptom-limited maximal treadmill test duration (Maxdur); brain MRI was measured 5 years later. Brain MRI measures were analyzed as means and as proportions below the 15th percentile (above the 85th percentile for white matter abnormal tissue volume).

Results

Per 1-minute-higher Maxdur, the odds ratio for having less whole brain volume was 0.85 (p = 0.04) and for having low white matter integrity was 0.80 (p = 0.02), adjusted for age, race, sex, clinic, body mass index, smoking, alcohol, diet, physical activity, education, blood pressure, diabetes, total cholesterol, and lung function (plus intracranial volume for white matter integrity). No significant associations were observed between Maxdur and abnormal tissue volume or blood flow in white matter. Findings were similar for associations with continuous brain MRI measures.

Conclusions

Greater physical fitness was associated with more brain volume and greater white matter integrity measured 5 years later in middle-aged adults.

Unlabelled

Arterial spin labeling and phase contrast magnetic resonance imaging provide independent non-invasive methods for measuring cerebral blood flow. We compared global cerebral blood flow measurements obtained using pseudo-continuous arterial spin labeling and phase contrast in 436 middle-aged subjects acquired at two sites in the NHLBI CARDIA multisite study. Cerebral blood flow measured by phase contrast (CBFPC: 55.76 ± 12.05 ml/100 g/min) was systematically higher (p < 0.001) and more variable than cerebral blood flow measured by pseudo-continuous arterial spin labeling (CBFPCASL: 47.70 ± 9.75). The correlation between global cerebral blood flow values obtained from the two modalities was 0.59 (p < 0.001), explaining less than half of the observed variance in cerebral blood flow estimates. Well-established correlations of global cerebral blood flow with age and sex were similarly observed in both CBFPCASL and CBFPC CBFPC also demonstrated statistically significant site differences, whereas no such differences were observed in CBFPCASL No consistent velocity-dependent effects on pseudo-continuous arterial spin labeling were observed, suggesting that pseudo-continuous labeling efficiency does not vary substantially across typical adult carotid and vertebral velocities, as has previously been suggested.

Conclusions

Although CBFPCASL and CBFPC values show substantial similarity across the entire cohort, these data do not support calibration of CBFPCASL using CBFPC in individual subjects. The wide-ranging cerebral blood flow values obtained by both methods suggest that cerebral blood flow values are highly variable in the general population.

Background

The classic view of blood pressure (BP) reactivity to psychological stress in relation to cardiovascular risks assumes that excess reactivity is worse and lower reactivity is better. Evidence addressing how stress-induced BP reactivity in young adults is associated with midlife cognitive function is sparse.

Methods and results

We assessed BP reactivity during a star tracing task and a video game in adults aged 20 to 32 years. Twenty-three years later, cognitive function was assessed with use of the Digit Symbol Substitution Test (a psychomotor speed test), the Rey Auditory Verbal Learning Test (a verbal memory test), and the modified Stroop test (an executive function test). At the time of follow-up, participants (n=3021) had a mean age of 50.2 years; 56% were women, and 44% were black. In linear regression models adjusted for demographic and clinical characteristics including baseline and follow-up resting BP, lower systolic BP (SBP) reactivity during the star tracing and video game was associated with worse Digit Symbol Substitution Test scores (β [SE]: 0.11 [0.02] and 0.05 [0.02], respectively) and worse performance on the Stroop test (β [SE]: -0.06 [0.02] and -0.05 [0.02]; all P<0.01). SBP reactivity was more consistently associated than diastolic BP reactivity with cognitive function scores. The associations between SBP reactivity and cognitive function were mostly similar between blacks and whites.

Conclusions

Lower psychological stress-induced SBP reactivity in younger adults was associated with lower cognitive function in midlife. BP reactivity to psychological stressors may have different associations with target organs in hypertension.