Next generation research applications for hybrid PET/MR and PET/CT imaging using deep learning

INTRODUCTION

Recently there have been significant advances in the field of machine learning and artificial intelligence (AI) centered around imaging-based applications such as computer vision. In particular, the tremendous power of deep learning algorithms, primarily based on convolutional neural network strategies, is becoming increasingly apparent and has already had direct impact on the fields of radiology and nuclear medicine. While most early applications of computer vision to radiological imaging have focused on classification of images into disease categories, it is also possible to use these methods to improve image quality. Hybrid imaging approaches, such as PET/MRI and PET/CT, are ideal for applying these methods.

METHODS

This review will give an overview of the application of AI to improve image quality for PET imaging directly and how the additional use of anatomic information from CT and MRI can lead to further benefits. For PET, these performance gains can be used to shorten imaging scan times, with improvement in patient comfort and motion artifacts, or to push towards lower radiotracer doses. It also opens the possibilities for dual tracer studies, more frequent follow-up examinations, and new imaging indications. How to assess quality and the potential effects of bias in training and testing sets will be discussed.

CONCLUSION

Harnessing the power of these new technologies to extract maximal information from hybrid PET imaging will open up new vistas for both research and clinical applications with associated benefits in patient care.

Ischemic Core and Hypoperfusion Volumes Correlate With Infarct Size 24 Hours After Randomization in DEFUSE 3

Background and Purpose- Accurate prediction of the subsequent infarct volume early after stroke onset helps determine appropriate interventions and prognosis. In the DEFUSE 3 trial (Endovascular Therapy Following Imaging Evaluation for Ischemic Stroke), we evaluated the accuracy of baseline ischemic core and hypoperfusion volumes for predicting infarct volume 24 hours after randomization to endovascular thrombectomy versus medical management. We also assessed if the union of baseline ischemic core and the volume of persistent hypoperfusion at 24 hours after randomization predicts infarct volume. Methods- Patients in DEFUSE 3 with computed tomography perfusion imaging or magnetic resonance diffusion weighted imaging/perfusion imaging acquired at baseline and at 24 hours after randomization were included. Ischemic core and Tmax >6s hypoperfusion volumes at baseline and follow-up were calculated using RAPID software and compared with the infarct volumes obtained 24 hours after randomization. Patients were stratified by reperfusion status for analyses. Results- Of 125 eligible patients, 59 patients with >90% reperfusion had a strong correlation between baseline ischemic core volume and infarct volume 24 hours postrandomization ( r=0.83; P<0.0001), and 14 patients with <10% reperfusion had a strong correlation between baseline Tmax >6s volume and infarct volume 24 hours postrandomization ( r=0.77; P<0.001). In the 52 patients with 10% to 90% reperfusion, as well as in all 125 patients, the union of the baseline ischemic core and the follow-up Tmax >6s perfusion volume was highly correlated with infarct volume 24 hours postrandomization (for N=125; r=0.83; P<0.0001), with a median absolute difference of 21.3 mL between observed and predicted infarct volumes. Conclusions- The union of the irreversibly injured ischemic core and persistently hypoperfused tissue volumes, as identified by computed tomography perfusion or magnetic resonance diffusion weighted imaging/perfusion, predicted infarct volume at 24 hours after randomization in DEFUSE 3 patients. Clinical Trial Registration- URL: https://www.clinicaltrials.gov . Unique identifier: NCT02586415.

Contralateral Hemispheric Cerebral Blood Flow Measured With Arterial Spin Labeling Can Predict Outcome in Acute Stroke

Background and Purpose- Imaging is frequently used to select acute stroke patients for intra-arterial therapy. Quantitative cerebral blood flow can be measured noninvasively with arterial spin labeling magnetic resonance imaging. Cerebral blood flow levels in the contralateral (unaffected) hemisphere may affect capacity for collateral flow and patient outcome. The goal of this study was to determine whether higher contralateral cerebral blood flow (cCBF) in acute stroke identifies patients with better 90-day functional outcome. Methods- Patients were part of the prospective, multicenter iCAS study (Imaging Collaterals in Acute Stroke) between 2013 and 2017. Consecutive patients were enrolled after being diagnosed with anterior circulation acute ischemic stroke. Inclusion criteria were ischemic anterior circulation stroke, baseline National Institutes of Health Stroke Scale score ≥1, prestroke modified Rankin Scale score ≤2, onset-to-imaging time <24 hours, with imaging including diffusion-weighted imaging and arterial spin labeling. Patients were dichotomized into high and low cCBF groups based on median cCBF. Outcomes were assessed by day-1 and day-5 National Institutes of Health Stroke Scale; and day-30 and day-90 modified Rankin Scale. Multivariable logistic regression was used to test whether cCBF predicted good neurological outcome (modified Rankin Scale score, 0-2) at 90 days. Results- Seventy-seven patients (41 women) met the inclusion criteria with median (interquartile range) age of 66 (55-76) yrs, onset-to-imaging time of 4.8 (3.6-7.7) hours, and baseline National Institutes of Health Stroke Scale score of 13 (9-20). Median cCBF was 38.9 (31.2-44.5) mL per 100 g/min. Higher cCBF predicted good outcome at day 90 (odds ratio, 4.6 [95% CI, 1.4-14.7]; P=0.01), after controlling for baseline National Institutes of Health Stroke Scale, diffusion-weighted imaging lesion volume, and intra-arterial therapy. Conclusions- Higher quantitative cCBF at baseline is a significant predictor of good neurological outcome at day 90. cCBF levels may inform decisions regarding stroke triage, treatment of acute stroke, and general outcome prognosis. Clinical Trial Registration- URL: https://www.clinicaltrials.gov. Unique identifier: NCT02225730.

Identifying Hypoperfusion in Moyamoya Disease With Arterial Spin Labeling and an [15O]-Water Positron Emission Tomography/Magnetic Resonance Imaging Normative Database

Background and Purpose- Noninvasive imaging of brain perfusion has the potential to elucidate pathophysiological mechanisms underlying Moyamoya disease and enable clinical imaging of cerebral blood flow (CBF) to select revascularization therapies for patients. We used hybrid positron emission tomography (PET)/magnetic resonance imaging (MRI) technology to characterize the distribution of hypoperfusion in Moyamoya disease and its relationship to vessel stenosis severity, through comparisons with a normative perfusion database of healthy controls. Methods- To image CBF, we acquired [¹⁵O]-water PET as a reference and simultaneously acquired arterial spin labeling (ASL) MRI scans in 20 Moyamoya patients and 15 age-matched, healthy controls on a PET/MRI scanner. The ASL MRI scans included a standard single-delay ASL scan with postlabel delay of 2.0 s and a multidelay scan with 5 postlabel delays (0.7-3.0s) to estimate and account for arterial transit time in CBF quantification. The percent volume of hypoperfusion in patients (determined as the fifth percentile of CBF values in the healthy control database) was the outcome measure in a logistic regression model that included stenosis grade and location. Results- Logistic regression showed that anterior ( P<0.0001) and middle cerebral artery territory regions ( P=0.003) in Moyamoya patients were susceptible to hypoperfusion, whereas posterior regions were not. Cortical regions supplied by arteries with stenosis on MR angiography showed more hypoperfusion than normal arteries ( P=0.001), but the extent of hypoperfusion was not different between mild-moderate versus severe stenosis. Multidelay ASL did not perform differently from [¹⁵O]-water PET in detecting perfusion abnormalities, but standard ASL overestimated the extent of hypoperfusion in patients ( P=0.003). Conclusions- This simultaneous PET/MRI study supports the use of multidelay ASL MRI in clinical evaluation of Moyamoya disease in settings where nuclear medicine imaging is not available and application of a normative perfusion database to automatically identify abnormal CBF in patients.

Summary of the First ISMRM-SNMMI Workshop on PET/MRI: Applications and Limitations

Since the introduction of simultaneous PET/MRI in 2011, there have been significant advancements. In this review, we highlight several technical advancements that have been made primarily in attenuation and motion correction and discuss the status of multiple clinical applications using PET/MRI. This review is based on the experience at the first PET/MRI conference cosponsored by the International Society for Magnetic Resonance in Medicine and the Society of Nuclear Medicine and Molecular Imaging.

MRI safety and devices: An update and expert consensus

The use of magnetic resonance imaging (MRI) is increasing globally, and MRI safety issues regarding medical devices, which are constantly being developed or upgraded, represent an ongoing challenge for MRI personnel. To assist the MRI community, a panel of 10 radiologists with expertise in MRI safety from nine high-volume academic centers formed, with the objective of providing clarity on some of the MRI safety issues for the 10 most frequently questioned devices. Ten device categories were identified. The panel reviewed the literature, including key MRI safety issues regarding screening and adverse event reports, in addition to the manufacturer's Instructions For Use. Using a Delphi-inspired method, 36 practical recommendations were generated with 100% consensus that can aid the clinical MRI community. Level of Evidence: 5 Technical Efficacy Stage: 5 J. Magn. Reson. Imaging 2020;51:657-674.

Applications of Deep Learning to Neuro-Imaging Techniques

Many clinical applications based on deep learning and pertaining to radiology have been proposed and studied in radiology for classification, risk assessment, segmentation tasks, diagnosis, prognosis, and even prediction of therapy responses. There are many other innovative applications of AI in various technical aspects of medical imaging, particularly applied to the acquisition of images, ranging from removing image artifacts, normalizing/harmonizing images, improving image quality, lowering radiation and contrast dose, and shortening the duration of imaging studies. This article will address this topic and will seek to present an overview of deep learning applied to neuroimaging techniques.

Ultra-low-dose PET reconstruction using generative adversarial network with feature matching and task-specific perceptual loss

PURPOSE

Our goal was to use a generative adversarial network (GAN) with feature matching and task-specific perceptual loss to synthesize standard-dose amyloid Positron emission tomography (PET) images of high quality and including accurate pathological features from ultra-low-dose PET images only.

METHODS

Forty PET datasets from 39 participants were acquired with a simultaneous PET/MRI scanner following injection of 330 ± 30 MBq of the amyloid radiotracer 18F-florbetaben. The raw list-mode PET data were reconstructed as the standard-dose ground truth and were randomly undersampled by a factor of 100 to reconstruct 1% low-dose PET scans. A 2D encoder-decoder network was implemented as the generator to synthesize a standard-dose image and a discriminator was used to evaluate them. The two networks contested with each other to achieve high-visual quality PET from the ultra-low-dose PET. Multi-slice inputs were used to reduce noise by providing the network with 2.5D information. Feature matching was applied to reduce hallucinated structures. Task-specific perceptual loss was designed to maintain the correct pathological features. The image quality was evaluated by peak signal-to-noise ratio (PSNR), structural similarity (SSIM), and root mean square error (RMSE) metrics with and without each of these modules. Two expert radiologists were asked to score image quality on a 5-point scale and identified the amyloid status (positive or negative).

RESULTS

With only low-dose PET as input, the proposed method significantly outperformed Chen et al.'s method (Chen et al. Radiology. 2018;290:649-656) (which shows the best performance in this task) with the same input (PET-only model) by 1.87 dB in PSNR, 2.04% in SSIM, and 24.75% in RMSE. It also achieved comparable results to Chen et al.'s method which used additional magnetic resonance imaging (MRI) inputs (PET-MR model). Experts' reading results showed that the proposed method could achieve better overall image quality and maintain better pathological features indicating amyloid status than both PET-only and PET-MR models proposed by Chen et al. CONCLUSION: Standard-dose amyloid PET images can be synthesized from ultra-low-dose images using GAN. Applying adversarial learning, feature matching, and task-specific perceptual loss are essential to ensure image quality and the preservation of pathological features.

Identifying cardiovascular risk factors that impact cerebrovascular reactivity: An ASL MRI study

BACKGROUND

To maintain cerebral blood flow (CBF), cerebral blood vessels dilate and contract in response to blood supply through cerebrovascular reactivity (CR).

PURPOSE

Cardiovascular (CV) disease is associated with increased stroke risk, but which risk factors specifically impact CR is unknown.

STUDY TYPE

Prospective longitudinal.

SUBJECTS

Fifty-three subjects undergoing carotid endarterectomy or stenting.

FIELD STRENGTH/SEQUENCE

3T, 3D pseudo-continuous arterial spin labeling (PCASL) ASL, and T₁ 3D fast spoiled gradient echo (FSPGR).

ASSESSMENT

We evaluated group differences in CBF changes for multiple cardiovascular risk factors in patients undergoing carotid revascularization surgery.

STATISTICAL TESTS

PRE (baseline), POST (48-hour postop), and 6MO (6 months postop) whole-brain CBF measurements, as 129 CBF maps from 53 subjects were modeled as within-subject analysis of variance (ANOVA). To identify CV risk factors associated with CBF change, the CBF change from PRE to POST, POST to 6MO, and PRE to 6MO were modeled as multiple linear regression with each CV risk factor as an independent variable. Statistical models were performed controlling for age on a voxel-by-voxel basis using SPM8. Significant clusters were reported if familywise error (FWE)-corrected cluster-level was P < 0.05, while the voxel-level significance threshold was set for P < 0.001.

RESULTS

The entire group showed significant (cluster-level P < 0.001) CBF increase from PRE to POST, decrease from POST to 6MO, and no significant difference (all voxels with P > 0.001) from PRE to 6MO. Of multiple CV risk factors evaluated, only elevated systolic blood pressure (SBP, P = 0.001), chronic renal insufficiency (CRI, P = 0.026), and history of prior stroke (CVA, P < 0.001) predicted lower increases in CBF PRE to POST. Over POST to 6MO, obesity predicted lower (P > 0.001) and cholesterol greater CBF decrease (P > 0.001).

DATA CONCLUSION

The CV risk factors of higher SBP, CRI, CVA, BMI, and cholesterol may indicate altered CR, and may warrant different stroke risk mitigation and special consideration for CBF change evaluation.

LEVEL OF EVIDENCE

1 Technical Efficacy: Stage 5 J. Magn. Reson. Imaging 2020;51:734-747.

Rigid Motion Correction for Brain PET/MR Imaging using Optical Tracking

A significant challenge during high-resolution PET brain imaging on PET/MR scanners is patient head motion. This challenge is particularly significant for clinical patient populations who struggle to remain motionless in the scanner for long periods of time. Head motion also affects the MR scan data. An optical motion tracking technique, which has already been demonstrated to perform MR motion correction during acquisition, is used with a list-mode PET reconstruction algorithm to correct the motion for each recorded event and produce a corrected reconstruction. The technique is demonstrated on real Alzheimer's disease patient data for the GE SIGNA PET/MR scanner.

Erroneous Resting-State fMRI Connectivity Maps Due to Prolonged Arterial Arrival Time and How to Fix Them

In resting-state functional MRI (rs-fMRI), functional networks are assessed utilizing the temporal correlation between spontaneous blood oxygen level-dependent signal fluctuations of spatially remote brain regions. Recently, several groups have shown that temporal shifts are present in rs-fMRI maps in patients with cerebrovascular disease due to spatial differences in arterial arrival times, and that this can be exploited to map arrival times in the brain. This suggests that rs-fMRI connectivity mapping may be similarly sensitive to such temporal shifts, and that standard rs-fMRI analysis methods may fail to identify functional connectivity networks. To investigate this, we studied the default mode network (DMN) in Moyamoya disease patients and compared it with normal healthy volunteers. Our results show that using standard independent component analysis (ICA) and seed-based approaches, arterial arrival delays lead to inaccurate incomplete characterization of functional connectivity within the DMN in Moyamoya disease patients. Furthermore, we propose two techniques to correct these errors, for seed-based and ICA methods, respectively. Using these methods, we demonstrate that it is possible to mitigate the deleterious effects of arterial arrival time on the assessment of functional connectivity of the DMN. As these corrections have not been applied to the vast majority of >200 prior rs-fMRI studies in patients with cerebrovascular disease, we suggest that they be interpreted with great caution. Correction methods should be applied in any rs-fMRI connectivity study of subjects expected to have abnormally delayed arterial arrival times.

Deep learning enables automatic detection and segmentation of brain metastases on multisequence MRI

BACKGROUND

Detecting and segmenting brain metastases is a tedious and time-consuming task for many radiologists, particularly with the growing use of multisequence 3D imaging.

PURPOSE

To demonstrate automated detection and segmentation of brain metastases on multisequence MRI using a deep-learning approach based on a fully convolution neural network (CNN).

STUDY TYPE

Retrospective.

POPULATION

In all, 156 patients with brain metastases from several primary cancers were included.

FIELD STRENGTH

1.5T and 3T. [Correction added on May 24, 2019, after first online publication: In the preceding sentence, the first field strength listed was corrected.] SEQUENCE: Pretherapy MR images included pre- and postgadolinium T₁ -weighted 3D fast spin echo (CUBE), postgadolinium T₁ -weighted 3D axial IR-prepped FSPGR (BRAVO), and 3D CUBE fluid attenuated inversion recovery (FLAIR).

ASSESSMENT

The ground truth was established by manual delineation by two experienced neuroradiologists. CNN training/development was performed using 100 and 5 patients, respectively, with a 2.5D network based on a GoogLeNet architecture. The results were evaluated in 51 patients, equally separated into those with few (1-3), multiple (4-10), and many (>10) lesions.

STATISTICAL TESTS

Network performance was evaluated using precision, recall, Dice/F1 score, and receiver operating characteristic (ROC) curve statistics. For an optimal probability threshold, detection and segmentation performance was assessed on a per-metastasis basis. The Wilcoxon rank sum test was used to test the differences between patient subgroups.

RESULTS

The area under the ROC curve (AUC), averaged across all patients, was 0.98 ± 0.04. The AUC in the subgroups was 0.99 ± 0.01, 0.97 ± 0.05, and 0.97 ± 0.03 for patients having 1-3, 4-10, and >10 metastases, respectively. Using an average optimal probability threshold determined by the development set, precision, recall, and Dice score were 0.79 ± 0.20, 0.53 ± 0.22, and 0.79 ± 0.12, respectively. At the same probability threshold, the network showed an average false-positive rate of 8.3/patient (no lesion-size limit) and 3.4/patient (10 mm³ lesion size limit).

DATA CONCLUSION

A deep-learning approach using multisequence MRI can automatically detect and segment brain metastases with high accuracy.

LEVEL OF EVIDENCE

3 Technical Efficacy Stage: 2 J. Magn. Reson. Imaging 2020;51:175-182.

Simultaneous phase-contrast MRI and PET for noninvasive quantification of cerebral blood flow and reactivity in healthy subjects and patients with cerebrovascular disease

BACKGROUND

H₂ ¹⁵ O-positron emission tomography (PET) is considered the reference standard for absolute cerebral blood flow (CBF). However, this technique requires an arterial input function measured through continuous sampling of arterial blood, which is invasive and has limitations with tracer delay and dispersion.

PURPOSE

To demonstrate a new noninvasive method to quantify absolute CBF with a PET/MRI hybrid scanner. This blood-free approach, called PC-PET, takes the spatial CBF distribution from a static H₂ ¹⁵ O-PET scan, and scales it to the whole-brain average CBF value measured by simultaneous phase-contrast MRI.

STUDY TYPE

Observational.

SUBJECTS

Twelve healthy controls (HC) and 13 patients with Moyamoya disease (MM) as a model of chronic ischemic disease.

FIELD STRENGTH/SEQUENCES

3T/2D cardiac-gated phase-contrast MRI and H₂ ¹⁵ O-PET.

ASSESSMENT

PC-PET CBF values from whole brain (WB), gray matter (GM), and white matter (WM) in HCs were compared with literature values since 2000. CBF and cerebrovascular reactivity (CVR), which is defined as the percent CBF change between baseline and post-acetazolamide (vasodilator) scans, were measured by PC-PET in MM patients and HCs within cortical regions corresponding to major vascular territories. Statistical Tests: Linear, mixed effects models were created to compare CBF and CVR, respectively, between patients and controls, and between different degrees of stenosis.

RESULTS

The mean CBF values in WB, GM, and WM in HC were 42 ± 7 ml/100 g/min, 50 ± 7 ml/100 g/min, and 23 ± 3 ml/100 g/min, respectively, which agree well with literature values. Compared with normal regions (57 ± 23%), patients showed significantly decreased CVR in areas with mild/moderate stenosis (47 ± 17%, P = 0.011) and in severe/occluded areas (40 ± 16%, P = 0.016). Data Conclusion: PC-PET identifies differences in cerebrovascular reactivity between healthy controls and cerebrovascular patients. PC-PET is suitable for CBF measurement when arterial blood sampling is not accessible, and warrants comparison to fully quantitative H₂ ¹⁵ O-PET in future studies.

LEVEL OF EVIDENCE

3 Technical Efficacy Stage: 2 J. Magn. Reson. Imaging 2019. J. Magn. Reson. Imaging 2020;51:183-194.

Ultra-Low-Dose 18F-Florbetaben Amyloid PET Imaging Using Deep Learning with Multi-Contrast MRI Inputs

Purpose To reduce radiotracer requirements for amyloid PET/MRI without sacrificing diagnostic quality by using deep learning methods. Materials and Methods Forty data sets from 39 patients (mean age ± standard deviation [SD], 67 years ± 8), including 16 male patients and 23 female patients (mean age, 66 years ± 6 and 68 years ± 9, respectively), who underwent simultaneous amyloid (fluorine 18 [18F]-florbetaben) PET/MRI examinations were acquired from March 2016 through October 2017 and retrospectively analyzed. One hundredth of the raw list-mode PET data were randomly chosen to simulate a low-dose (1%) acquisition. Convolutional neural networks were implemented with low-dose PET and multiple MR images (PET-plus-MR model) or with low-dose PET alone (PET-only) as inputs to predict full-dose PET images. Quality of the synthesized images was evaluated while Bland-Altman plots assessed the agreement of regional standard uptake value ratios (SUVRs) between image types. Two readers scored image quality on a five-point scale (5 = excellent) and determined amyloid status (positive or negative). Statistical analyses were carried out to assess the difference of image quality metrics and reader agreement and to determine confidence intervals (CIs) for reading results. Results The synthesized images (especially from the PET-plus-MR model) showed marked improvement on all quality metrics compared with the low-dose image. All PET-plus-MR images scored 3 or higher, with proportions of images rated greater than 3 similar to those for the full-dose images (-10% difference [eight of 80 readings], 95% CI: -15%, -5%). Accuracy for amyloid status was high (71 of 80 readings [89%]) and similar to intrareader reproducibility of full-dose images (73 of 80 [91%]). The PET-plus-MR model also had the smallest mean and variance for SUVR difference to full-dose images. Conclusion Simultaneously acquired MRI and ultra-low-dose PET data can be used to synthesize full-dose-like amyloid PET images. © RSNA, 2018 Online supplemental material is available for this article. See also the editorial by Catana in this issue.

Abstract WP61: Union of Ischemic Core and Hypoperfusion Volume Correlates With 24-hour Infarct Size in DEFUSE 3

Background and Purpose: Accurate prediction of 24-hour infarct volume early after stroke onset helps determine appropriate interventions and prognosis. In the Endovascular Therapy Following Imaging Evaluation for Ischemic Stroke (DEFUSE 3) trial, we evaluated the accuracy of early ischemic core and hypoperfusion volumes for predicting infarct volume 24 hours after randomization to endovascular thrombectomy vs. medical management. We also assessed if the volume of persistent hypoperfusion at 24 hours predicts infarct volume.

Methods: Patients in DEFUSE 3 with CT or MRI DWI/perfusion imaging acquired at baseline and at 24 hours after randomization were included. Ischemic core and Tmax>6s hypoperfusion volumes at baseline and 24 hours were calculated using RAPID software and compared with the 24-hour infarct volumes. Patients were stratified by reperfusion status for analyses.

Results: Of 125 eligible patients, 59 patients with >90% reperfusion had a strong correlation between baseline ischemic core volumes and 24-hour infarct volume (r = 0.83; p < 0.0001), and 14 patients with <10% reperfusion had a strong correlation between baseline Tmax>6s volume and 24-hour infarct volume (r = 0.77; p < 0.001). In the 52 patients with 10-90% reperfusion, as well as in all 125 patients, the union of the baseline ischemic core and the Tmax>6s perfusion volume at 24 hours was highly correlated with 24-hour infarct volume (for N=125, r = 0.83; p < 0.0001), with a median absolute difference of 21.3 ml between observed and predicted infarct volumes.

Conclusions: The union of the irreversibly injured ischemic core and critically hypoperfused tissue volumes, as identified by CT perfusion or MR DWI/perfusion, predicted infarct volume at 24 hours in DEFUSE 3 patients.

Clinical Trial Registration - URL: http://www.clinicaltrials.gov. Unique identifier: NCT02586415

Deep Generative Adversarial Neural Networks for Compressive Sensing MRI

Undersampled magnetic resonance image (MRI) reconstruction is typically an ill-posed linear inverse task. The time and resource intensive computations require tradeoffs between accuracy and speed. In addition, state-of-the-art compressed sensing (CS) analytics are not cognizant of the image diagnostic quality. To address these challenges, we propose a novel CS framework that uses generative adversarial networks (GAN) to model the (low-dimensional) manifold of high-quality MR images. Leveraging a mixture of least-squares (LS) GANs and pixel-wise l1/l2 cost, a deep residual network with skip connections is trained as the generator that learns to remove the aliasing artifacts by projecting onto the image manifold. The LSGAN learns the texture details, while the l1/l2 cost suppresses high-frequency noise. A discriminator network, which is a multilayer convolutional neural network (CNN), plays the role of a perceptual cost that is then jointly trained based on high-quality MR images to score the quality of retrieved images. In the operational phase, an initial aliased estimate (e.g., simply obtained by zero-filling) is propagated into the trained generator to output the desired reconstruction. This demands a very low computational overhead. Extensive evaluations are performed on a large contrast-enhanced MR dataset of pediatric patients. Images rated by expert radiologists corroborate that GANCS retrieves higher quality images with improved fine texture details compared with conventional Wavelet-based and dictionary-learning-based CS schemes as well as with deep-learning-based schemes using pixel-wise training. In addition, it offers reconstruction times of under a few milliseconds, which are two orders of magnitude faster than the current state-of-the-art CS-MRI schemes.

Revealing sub-voxel motions of brain tissue using phase-based amplified MRI (aMRI)

PURPOSE

Amplified magnetic resonance imaging (aMRI) was recently introduced as a new brain motion detection and visualization method. The original aMRI approach used a video-processing algorithm, Eulerian video magnification (EVM), to amplify cardio-ballistic motion in retrospectively cardiac-gated MRI data. Here, we strive to improve aMRI by incorporating a phase-based motion amplification algorithm.

METHODS

Phase-based aMRI was developed and tested for correct implementation and ability to amplify sub-voxel motions using digital phantom simulations. The image quality of phase-based aMRI was compared with EVM-based aMRI in healthy volunteers at 3T, and its amplified motion characteristics were compared with phase-contrast MRI. Data were also acquired on a patient with Chiari I malformation, and qualitative displacement maps were produced using free form deformation (FFD) of the aMRI output.

RESULTS

Phantom simulations showed that phase-based aMRI has a linear dependence of amplified displacement on true displacement. Amplification was independent of temporal frequency, varying phantom intensity, Rician noise, and partial volume effect. Phase-based aMRI supported larger amplification factors than EVM-based aMRI and was less sensitive to noise and artifacts. Abnormal biomechanics were seen on FFD maps of the Chiari I malformation patient.

CONCLUSION

Phase-based aMRI might be used in the future for quantitative analysis of minute changes in brain motion and may reveal subtle physiological variations of the brain as a result of pathology using processing of the fundamental harmonic or by selectively varying temporal harmonics. Preliminary data shows the potential of phase-based aMRI to qualitatively assess abnormal biomechanics in Chiari I malformation.

Deep Learning in Neuroradiology

Deep learning is a form of machine learning using a convolutional neural network architecture that shows tremendous promise for imaging applications. It is increasingly being adapted from its original demonstration in computer vision applications to medical imaging. Because of the high volume and wealth of multimodal imaging information acquired in typical studies, neuroradiology is poised to be an early adopter of deep learning. Compelling deep learning research applications have been demonstrated, and their use is likely to grow rapidly. This review article describes the reasons, outlines the basic methods used to train and test deep learning models, and presents a brief overview of current and potential clinical applications with an emphasis on how they are likely to change future neuroradiology practice. Facility with these methods among neuroimaging researchers and clinicians will be important to channel and harness the vast potential of this new method.

ISLES 2016 and 2017-Benchmarking Ischemic Stroke Lesion Outcome Prediction Based on Multispectral MRI

Performance of models highly depend not only on the used algorithm but also the data set it was applied to. This makes the comparison of newly developed tools to previously published approaches difficult. Either researchers need to implement others' algorithms first, to establish an adequate benchmark on their data, or a direct comparison of new and old techniques is infeasible. The Ischemic Stroke Lesion Segmentation (ISLES) challenge, which has ran now consecutively for 3 years, aims to address this problem of comparability. ISLES 2016 and 2017 focused on lesion outcome prediction after ischemic stroke: By providing a uniformly pre-processed data set, researchers from all over the world could apply their algorithm directly. A total of nine teams participated in ISLES 2015, and 15 teams participated in ISLES 2016. Their performance was evaluated in a fair and transparent way to identify the state-of-the-art among all submissions. Top ranked teams almost always employed deep learning tools, which were predominately convolutional neural networks (CNNs). Despite the great efforts, lesion outcome prediction persists challenging. The annotated data set remains publicly available and new approaches can be compared directly via the online evaluation system, serving as a continuing benchmark (www.isles-challenge.org).

Consensus statement on current and emerging methods for the diagnosis and evaluation of cerebrovascular disease

Cerebrovascular disease (CVD) remains a leading cause of death and the leading cause of adult disability in most developed countries. This work summarizes state-of-the-art, and possible future, diagnostic and evaluation approaches in multiple stages of CVD, including (i) visualization of sub-clinical disease processes, (ii) acute stroke theranostics, and (iii) characterization of post-stroke recovery mechanisms. Underlying pathophysiology as it relates to large vessel steno-occlusive disease and the impact of this macrovascular disease on tissue-level viability, hemodynamics (cerebral blood flow, cerebral blood volume, and mean transit time), and metabolism (cerebral metabolic rate of oxygen consumption and pH) are also discussed in the context of emerging neuroimaging protocols with sensitivity to these factors. The overall purpose is to highlight advancements in stroke care and diagnostics and to provide a general overview of emerging research topics that have potential for reducing morbidity in multiple areas of CVD.

Evaluation of Thick-Slab Overlapping MIP Images of Contrast-Enhanced 3D T1-Weighted CUBE for Detection of Intracranial Metastases: A Pilot Study for Comparison of Lesion Detection, Interpretation Time, and Sensitivity with Nonoverlapping CUBE MIP, CUBE, and Inversion-Recovery-Prepared Fast-Spoiled Gradient Recalled Brain Volume

BACKGROUND AND PURPOSE

Early and accurate identification of cerebral metastases is important for prognostication and treatment planning although this process is often time consuming and labor intensive, especially with the hundreds of images associated with 3D volumetric imaging. This study aimed to evaluate the benefits of thick-slab overlapping MIPs constructed from contrast-enhanced T1-weighted CUBE (overlapping CUBE MIP) for the detection of brain metastases in comparison with traditional CUBE and inversion-recovery prepared fast-spoiled gradient recalled brain volume (IR-FSPGR-BRAVO) and nonoverlapping CUBE MIP.

MATERIALS AND METHODS

A retrospective review of 48 patients with cerebral metastases was performed at our institution from June 2016 to October 2017. Brain MRIs, which were acquired on multiple 3T scanners, included gadolinium-enhanced T1-weighted IR-FSPGR-BRAVO and CUBE, with subsequent generation of nonoverlapping CUBE MIP and overlapping CUBE MIP. Two blinded radiologists identified the total number and location of metastases on each image type. The Cohen κ was used to determine interrater agreement. Sensitivity, interpretation time, and lesion contrast-to-noise ratio were assessed.

RESULTS

Interrater agreement for identification of metastases was fair-to-moderate for all image types (κ = 0.222-0.598). The total number of metastases identified was not significantly different across the image types. Interpretation time for CUBE MIPs was significantly shorter than for CUBE and IR-FSPGR-BRAVO, saving at least 50 seconds per case on average (P < .001). The mean lesion contrast-to-noise ratio for both CUBE MIPs was higher than for IR-FSPGR-BRAVO. The mean contrast-to-noise ratio for small lesions (<4 mm) was lower for nonoverlapping CUBE MIP (1.55) than for overlapping CUBE MIP (2.35). For both readers, the sensitivity for lesion detection was high for all image types but highest for overlapping CUBE MIP and CUBE (0.93-0.97).

CONCLUSIONS

This study suggests that the use of overlapping CUBE MIP or nonoverlapping CUBE MIP for the detection of brain metastases can reduce interpretation time without sacrificing sensitivity, though the contrast-to-noise ratio of lesions is highest for overlapping CUBE MIP.