Clinical Focus


  • Diagnostic Radiology

Academic Appointments


Professional Education


  • Fellowship: Stanford University Neuroradiology Fellowship (2019) CA
  • Board Certification: American Board of Radiology, Diagnostic Radiology (2018)
  • Fellowship: UCSF Neuroradiology Fellowship (2018) CA
  • Residency: University of Virgina School of Medicine (2016) VA
  • Internship: Wake Forest Baptist Medical Center (2012)
  • Medical Education: University of California, San Diego (2011) CA

All Publications


  • Machine Learning Approach to Differentiation of Peripheral Schwannomas and Neurofibromas: A Multi-Center Study. Neuro-oncology Zhang, M., Tong, E., Wong, S., Hamrick, F., Mohammadzadeh, M., Rao, V., Pendleton, C., Smith, B. W., Hug, N. F., Biswal, S., Seekins, J., Napel, S., Spinner, R. J., Mahan, M. A., Yeom, K. W., Wilson, T. J. 2021

    Abstract

    BACKGROUND: Non-invasive differentiation between schwannomas and neurofibromas is important for appropriate management, preoperative counseling, and surgical planning, but has proven difficult using conventional imaging. The objective of this study was to develop and evaluate machine learning approaches for differentiating peripheral schwannomas from neurofibromas.METHODS: We assembled a cohort of schwannomas and neurofibromas from 3 independent institutions and extracted high-dimensional radiomic features from gadolinium-enhanced, T1-weighted MRI using the PyRadiomics package on Quantitative Imaging Feature Pipeline. Age, sex, neurogenetic syndrome, spontaneous pain, and motor deficit were recorded. We evaluated the performance of 6 radiomics-based classifier models with and without clinical features and compared model performance against human expert evaluators.RESULTS: 107 schwannomas and 59 neurofibroma were included. The primary models included both clinical and imaging data. The accuracy of the human evaluators (0.765) did not significantly exceed the no-information rate (NIR), whereas the Support Vector Machine (0.929), Logistic Regression (0.929), and Random Forest (0.905) classifiers exceeded the NIR. Using the method of DeLong, the AUC for the Logistic Regression (AUC=0.923) and K Nearest Neighbor (AUC=0.923) classifiers was significantly greater than the human evaluators (AUC=0.766; p = 0.041).CONCLUSIONS: The radiomics-based classifiers developed here proved to be more accurate and had a higher AUC on the ROC curve than expert human evaluators. This demonstrates that radiomics using routine MRI sequences and clinical features can aid in differentiation of peripheral schwannomas and neurofibromas.

    View details for DOI 10.1093/neuonc/noab211

    View details for PubMedID 34487172

  • Machine-learning Approach to Differentiation of Benign and Malignant Peripheral Nerve Sheath Tumors: A Multicenter Study Zhang, M., Tong, E., Hamrick, F., Pendleton, C., Smith, B., Hug, N., Mattonen, S., Napel, S., Spinner, R., Yeom, K., Wilson, T., Mahan, M. AMER ASSOC NEUROLOGICAL SURGEONS. 2021
  • Machine-Learning Approach to Differentiation of Benign and Malignant Peripheral Nerve Sheath Tumors: A Multicenter Study. Neurosurgery Zhang, M., Tong, E., Hamrick, F., Lee, E. H., Tam, L. T., Pendleton, C., Smith, B. W., Hug, N. F., Biswal, S., Seekins, J., Mattonen, S. A., Napel, S., Campen, C. J., Spinner, R. J., Yeom, K. W., Wilson, T. J., Mahan, M. A. 2021

    Abstract

    BACKGROUND: Clinicoradiologic differentiation between benign and malignant peripheral nerve sheath tumors (PNSTs) has important management implications.OBJECTIVE: To develop and evaluate machine-learning approaches to differentiate benign from malignant PNSTs.METHODS: We identified PNSTs treated at 3 institutions and extracted high-dimensional radiomics features from gadolinium-enhanced, T1-weighted magnetic resonance imaging (MRI) sequences. Training and test sets were selected randomly in a 70:30 ratio. A total of 900 image features were automatically extracted using the PyRadiomics package from Quantitative Imaging Feature Pipeline. Clinical data including age, sex, neurogenetic syndrome presence, spontaneous pain, and motor deficit were also incorporated. Features were selected using sparse regression analysis and retained features were further refined by gradient boost modeling to optimize the area under the curve (AUC) for diagnosis. We evaluated the performance of radiomics-based classifiers with and without clinical features and compared performance against human readers.RESULTS: A total of 95 malignant and 171 benign PNSTs were included. The final classifier model included 21 imaging and clinical features. Sensitivity, specificity, and AUC of 0.676, 0.882, and 0.845, respectively, were achieved on the test set. Using imaging and clinical features, human experts collectively achieved sensitivity, specificity, and AUC of 0.786, 0.431, and 0.624, respectively. The AUC of the classifier was statistically better than expert humans (P=.002). Expert humans were not statistically better than the no-information rate, whereas the classifier was (P=.001).CONCLUSION: Radiomics-based machine learning using routine MRI sequences and clinical features can aid in evaluation of PNSTs. Further improvement may be achieved by incorporating additional imaging sequences and clinical variables into future models.

    View details for DOI 10.1093/neuros/nyab212

    View details for PubMedID 34131749

  • Dopamine-Related Alterations of Frontostriatal Habit Circuitry Underlie Stimulus-Response Binge Eating Wang, A., Kuijper, F. M., Barbosa, D., Hagan, K., Lee, E., Tong, E., Bohon, C., Halpern, C. H. ELSEVIER SCIENCE INC. 2021: S233-S234
  • Maximizing the use of batch production of 18F-FDOPA for imaging of brain tumors to increase availability of hybrid PET/MR imaging in clinical setting. Neuro-oncology practice Aboian, M., Barajas, R., Shatalov, J., Ravanfar, V., Bahroos, E., Tong, E., Taylor, J. W., Bush, N. O., Sneed, P., Seo, Y., Cha, S., Hernandez-Pampaloni, M. 2021; 8 (1): 91–97

    Abstract

    Background: Amino acid PET imaging of brain tumors has been shown to play an important role in predicting tumor grade, delineation of tumor margins, and differentiating tumor recurrence from the background of postradiation changes, but is not commonly used in clinical practice because of high cost. We propose that PET/MRI imaging of patients grouped to the day of tracer radiosynthesis will significantly decrease the cost of PET imaging, which will improve patient access to PET.Methods: Seventeen patients with either primary brain tumors or metastatic brain tumors were recruited for imaging on 3T PET/MRI and were scanned on 4 separate days in groups of 3 to 5 patients. The first group of consecutively imaged patients contained 3 patients, followed by 2 groups of 5 patients, and a last group of 4 patients.Results: For each of the patients, standard of care gadolinium-enhanced MRI and dynamic PET imaging with 18F-FDOPA amino acid tracer was obtained. The total cost savings of scanning 17 patients in batches of 4 as opposed to individual radiosynthesis was 48.5% ($28 321). Semiquantitative analysis of tracer uptake in normal brain were performed with appropriate accumulation and expected subsequent washout.Conclusion: Amino acid PET tracers have been shown to play a critical role in the characterization of brain tumors but their adaptation to clinical practice has been limited because of the high cost of PET. Scheduling patient imaging to maximally use the radiosynthesis of imaging tracer significantly reduces the cost of PET and results in increased availability of PET tracer use in neuro-oncology.

    View details for DOI 10.1093/nop/npaa065

    View details for PubMedID 33664973

  • High-resolution Structural Magnetic Resonance Imaging and Quantitative Susceptibility Mapping. Magnetic resonance imaging clinics of North America Yedavalli, V., DiGiacomo, P., Tong, E., Zeineh, M. 2021; 29 (1): 13–39

    Abstract

    High-resolution 7-T imaging and quantitative susceptibility mapping produce greater anatomic detail compared with conventional strengths because of improvements in signal/noise ratio and contrast. The exquisite anatomic details of deep structures, including delineation of microscopic architecture using advanced techniques such as quantitative susceptibility mapping, allows improved detection of abnormal findings thought to be imperceptible on clinical strengths. This article reviews caveats and techniques for translating sequences commonly used on 1.5 or 3 T to high-resolution 7-T imaging. It discusses for several broad disease categories how high-resolution 7-T imaging can advance the understanding of various diseases, improve diagnosis, and guide management.

    View details for DOI 10.1016/j.mric.2020.09.002

    View details for PubMedID 33237013

  • MRI pulse sequence integration for deep-learning-based brain metastases segmentation. Medical physics Yi, D., Grøvik, E., Tong, E., Iv, M., Emblem, K. E., Nilsen, L. B., Saxhaug, C., Latysheva, A., Jacobsen, K. D., Helland, Å., Zaharchuk, G., Rubin, D. 2021

    Abstract

    Magnetic resonance (MR) imaging is an essential diagnostic tool in clinical medicine. Recently, a variety of deep-learning methods have been applied to segmentation tasks in medical images, with promising results for computer-aided diagnosis. For MR images, effectively integrating different pulse sequences is important to optimize performance. However, the best way to integrate different pulse sequences remains unclear. In addition, networks trained with a certain subset of pulse sequences as input are unable to perform when given a subset of those pulse sequences. In this study, we evaluate multiple architectural features and characterize their effects in the task of metastasis segmentation while creating a method to robustly train a network to be able to work given any strict subset of the pulse sequences available during training.We use a 2.5D DeepLabv3 segmentation network to segment metastases lesions on brain MR's with four pulse sequence inputs. To study how we can best integrate MR pulse sequences for this task, we consider (1) different pulse sequence integration schemas, combining our features at early, middle, and late points within a deep network, (2) different modes of weight sharing for parallel network branches, and (3) a novel integration level dropout layer, which will allow the networks to be robust to performing inference on input with only a subset of pulse sequences available at the training.We find that levels of integration and modes of weight sharing that favor low variance work best in our regime of small amounts of training data (n = 100). By adding an input-level dropout layer, we could preserve the overall performance of these networks while allowing for inference on inputs with missing pulse sequences. We illustrate not only the generalizability of the network but also the utility of this robustness when applying the trained model to data from a different center, which does not use the same pulse sequences. Finally, we apply network visualization methods to better understand which input features are most important for network performance.Together, these results provide a framework for building networks with enhanced robustness to missing data while maintaining comparable performance in medical imaging applications.

    View details for DOI 10.1002/mp.15136

    View details for PubMedID 34405896

  • Non-contrast dual-energy CT virtual ischemia maps accurately estimate ischemic core size in large-vessel occlusive stroke. Scientific reports Wolman, D. N., van Ommen, F. n., Tong, E. n., Kauw, F. n., Dankbaar, J. W., Bennink, E. n., de Jong, H. W., Molvin, L. n., Wintermark, M. n., Heit, J. J. 2021; 11 (1): 6745

    Abstract

    Dual-energy CT (DECT) material decomposition techniques may better detect edema within cerebral infarcts than conventional non-contrast CT (NCCT). This study compared if Virtual Ischemia Maps (VIM) derived from non-contrast DECT of patients with acute ischemic stroke due to large-vessel occlusion (AIS-LVO) are superior to NCCT for ischemic core estimation, compared against reference-standard DWI-MRI. Only patients whose baseline ischemic core was most likely to remain stable on follow-up MRI were included, defined as those with excellent post-thrombectomy revascularization or no perfusion mismatch. Twenty-four consecutive AIS-LVO patients with baseline non-contrast DECT, CT perfusion (CTP), and DWI-MRI were analyzed. The primary outcome measure was agreement between volumetric manually segmented VIM, NCCT, and automatically segmented CTP estimates of the ischemic core relative to manually segmented DWI volumes. Volume agreement was assessed using Bland-Altman plots and comparison of CT to DWI volume ratios. DWI volumes were better approximated by VIM than NCCT (VIM/DWI ratio 0.68 ± 0.35 vs. NCCT/DWI ratio 0.34 ± 0.35; P < 0.001) or CTP (CTP/DWI ratio 0.45 ± 0.67; P < 0.001), and VIM best correlated with DWI (rVIM = 0.90; rNCCT = 0.75; rCTP = 0.77; P < 0.001). Bland-Altman analyses indicated significantly greater agreement between DWI and VIM than NCCT core volumes (mean bias 0.60 [95%AI 0.39-0.82] vs. 0.20 [95%AI 0.11-0.30]). We conclude that DECT VIM estimates the ischemic core in AIS-LVO patients more accurately than NCCT.

    View details for DOI 10.1038/s41598-021-85143-3

    View details for PubMedID 33762589

  • Handling missing MRI sequences in deep learning segmentation of brain metastases: a multicenter study. NPJ digital medicine Grøvik, E. n., Yi, D. n., Iv, M. n., Tong, E. n., Nilsen, L. B., Latysheva, A. n., Saxhaug, C. n., Jacobsen, K. D., Helland, Å. n., Emblem, K. E., Rubin, D. L., Zaharchuk, G. n. 2021; 4 (1): 33

    Abstract

    The purpose of this study was to assess the clinical value of a deep learning (DL) model for automatic detection and segmentation of brain metastases, in which a neural network is trained on four distinct MRI sequences using an input-level dropout layer, thus simulating the scenario of missing MRI sequences by training on the full set and all possible subsets of the input data. This retrospective, multicenter study, evaluated 165 patients with brain metastases. The proposed input-level dropout (ILD) model was trained on multisequence MRI from 100 patients and validated/tested on 10/55 patients, in which the test set was missing one of the four MRI sequences used for training. The segmentation results were compared with the performance of a state-of-the-art DeepLab V3 model. The MR sequences in the training set included pre-gadolinium and post-gadolinium (Gd) T1-weighted 3D fast spin echo, post-Gd T1-weighted inversion recovery (IR) prepped fast spoiled gradient echo, and 3D fluid attenuated inversion recovery (FLAIR), whereas the test set did not include the IR prepped image-series. The ground truth segmentations were established by experienced neuroradiologists. The results were evaluated using precision, recall, Intersection over union (IoU)-score and Dice score, and receiver operating characteristics (ROC) curve statistics, while the Wilcoxon rank sum test was used to compare the performance of the two neural networks. The area under the ROC curve (AUC), averaged across all test cases, was 0.989 ± 0.029 for the ILD-model and 0.989 ± 0.023 for the DeepLab V3 model (p = 0.62). The ILD-model showed a significantly higher Dice score (0.795 ± 0.104 vs. 0.774 ± 0.104, p = 0.017), and IoU-score (0.561 ± 0.225 vs. 0.492 ± 0.186, p < 0.001) compared to the DeepLab V3 model, and a significantly lower average false positive rate of 3.6/patient vs. 7.0/patient (p < 0.001) using a 10 mm3 lesion-size limit. The ILD-model, trained on all possible combinations of four MRI sequences, may facilitate accurate detection and segmentation of brain metastases on a multicenter basis, even when the test cohort is missing input MRI sequences.

    View details for DOI 10.1038/s41746-021-00398-4

    View details for PubMedID 33619361

  • GENOME ASSOCIATIONS WITH NEUROCOGNITIVE OUTCOMES, CEREBRAL MICROBLEEDS (CMBS), AND BRAIN VOLUME AND WHITE MATTER (WM) CHANGES IN PEDIATRIC BRAIN TUMOR SURVIVORS Kline, C., Stoller, S., Byer, L., Edwards, C., Prasad, R., Torkildson, J., Gauvain, K., Samuel, D., Lupo, J., Morrison, M., Tong, E., Savchuk, S., Valencia, C., Rauschecker, A., Rudie, J., Hoffman, T., Dubal, D., Fullerton, H., Mueller, S. OXFORD UNIV PRESS INC. 2020: 440
  • GENOME ASSOCIATIONS WITH NEUROCOGNITIVE OUTCOMES, CEREBRAL MICROBLEEDS (CMBS), AND BRAIN VOLUME AND WHITE MATTER (WM) CHANGES IN PEDIATRIC BRAIN TUMOR SURVIVORS Kline, C., Stoller, S., Byer, L., Edwards, C., Prasad, R., Torkildson, J., Gauvain, K., Samuel, D., Lupo, J., Morrison, M., Tong, E., Savchuk, S., Valencia, C., Rauschecker, A., Rudie, J., Hoffmann, T., Dubal, D., Fullerton, H., Mueller, S. OXFORD UNIV PRESS INC. 2020: 145
  • Synthesize High-Quality Multi-Contrast Magnetic Resonance Imaging From Multi-Echo Acquisition Using Multi-Task Deep Generative Model IEEE TRANSACTIONS ON MEDICAL IMAGING Wang, G., Gong, E., Banerjee, S., Martin, D., Tong, E., Choi, J., Chen, H., Wintermark, M., Pauly, J. M., Zaharchuk, G. 2020; 39 (10): 3089–99

    Abstract

    Multi-echo saturation recovery sequence can provide redundant information to synthesize multi-contrast magnetic resonance imaging. Traditional synthesis methods, such as GE's MAGiC platform, employ a model-fitting approach to generate parameter-weighted contrasts. However, models' over-simplification, as well as imperfections in the acquisition, can lead to undesirable reconstruction artifacts, especially in T2-FLAIR contrast. To improve the image quality, in this study, a multi-task deep learning model is developed to synthesize multi-contrast neuroimaging jointly using both signal relaxation relationships and spatial information. Compared with previous deep learning-based synthesis, the correlation between different destination contrast is utilized to enhance reconstruction quality. To improve model generalizability and evaluate clinical significance, the proposed model was trained and tested on a large multi-center dataset, including healthy subjects and patients with pathology. Results from both quantitative comparison and clinical reader study demonstrate that the multi-task formulation leads to more efficient and accurate contrast synthesis than previous methods.

    View details for DOI 10.1109/TMI.2020.2987026

    View details for Web of Science ID 000574745800010

    View details for PubMedID 32286966

  • A within-coil optical prospective motion-correction system for brain imaging at 7T. Magnetic resonance in medicine DiGiacomo, P. n., Maclaren, J. n., Aksoy, M. n., Tong, E. n., Carlson, M. n., Lanzman, B. n., Hashmi, S. n., Watkins, R. n., Rosenberg, J. n., Burns, B. n., Skloss, T. W., Rettmann, D. n., Rutt, B. n., Bammer, R. n., Zeineh, M. n. 2020

    Abstract

    Motion artifact limits the clinical translation of high-field MR. We present an optical prospective motion correction system for 7 Tesla MRI using a custom-built, within-coil camera to track an optical marker mounted on a subject.The camera was constructed to fit between the transmit-receive coils with direct line of sight to a forehead-mounted marker, improving upon prior mouthpiece work at 7 Tesla MRI. We validated the system by acquiring a 3D-IR-FSPGR on a phantom with deliberate motion applied. The same 3D-IR-FSPGR and a 2D gradient echo were then acquired on 7 volunteers, with/without deliberate motion and with/without motion correction. Three neuroradiologists blindly assessed image quality. In 1 subject, an ultrahigh-resolution 2D gradient echo with 4 averages was acquired with motion correction. Four single-average acquisitions were then acquired serially, with the subject allowed to move between acquisitions. A fifth single-average 2D gradient echo was acquired following subject removal and reentry.In both the phantom and human subjects, deliberate and involuntary motion were well corrected. Despite marked levels of motion, high-quality images were produced without spurious artifacts. The quantitative ratings confirmed significant improvements in image quality in the absence and presence of deliberate motion across both acquisitions (P < .001). The system enabled ultrahigh-resolution visualization of the hippocampus during a long scan and robust alignment of serially acquired scans with interspersed movement.We demonstrate the use of a within-coil camera to perform optical prospective motion correction and ultrahigh-resolution imaging at 7 Tesla MRI. The setup does not require a mouthpiece, which could improve accessibility of motion correction during 7 Tesla MRI exams.

    View details for DOI 10.1002/mrm.28211

    View details for PubMedID 32077521

  • Deep Learning Enables Automatic Detection and Segmentation of Brain Metastases on Multisequence MRI JOURNAL OF MAGNETIC RESONANCE IMAGING Grovik, E., Yi, D., Iv, M., Tong, E., Rubin, D., Zaharchuk, G. 2020; 51 (1): 175–82

    View details for DOI 10.1002/jmri.26766

    View details for Web of Science ID 000530627200017

  • Advanced Imaging of Brain Metastases: From Augmenting Visualization and Improving Diagnosis to Evaluating Treatment Response. Frontiers in neurology Tong, E. n., McCullagh, K. L., Iv, M. n. 2020; 11: 270

    Abstract

    Early detection of brain metastases and differentiation from other neuropathologies is crucial. Although biopsy is often required for definitive diagnosis, imaging can provide useful information. After treatment commences, imaging is also performed to assess the efficacy of treatment. Contrast-enhanced magnetic resonance imaging (MRI) is the traditional imaging method for the evaluation of brain metastases, as it provides information about lesion size, morphology, and macroscopic properties. Newer MRI sequences have been developed to increase the conspicuity of detecting enhancing metastases. Other advanced MRI techniques, that have the capability to probe beyond the anatomic structure, are available to characterize micro-structures, cellularity, physiology, perfusion, and metabolism. Artificial intelligence provides powerful computational tools for detection, segmentation, classification, prediction, and prognosis. We highlight and review a few advanced MRI techniques for the assessment of brain metastases-specifically for (1) diagnosis, including differentiating between malignancy types and (2) evaluation of treatment response, including the differentiation between radiation necrosis and disease progression.

    View details for DOI 10.3389/fneur.2020.00270

    View details for PubMedID 32351445

    View details for PubMedCentralID PMC7174761

  • Artificial Intelligence and Stroke Imaging: A West Coast Perspective. Neuroimaging clinics of North America Zhu, G. n., Jiang, B. n., Chen, H. n., Tong, E. n., Xie, Y. n., Faizy, T. D., Heit, J. J., Zaharchuk, G. n., Wintermark, M. n. 2020; 30 (4): 479–92

    Abstract

    Artificial intelligence (AI) advancements have significant implications for medical imaging. Stroke is the leading cause of disability and the fifth leading cause of death in the United States. AI applications for stroke imaging are a topic of intense research. AI techniques are well-suited for dealing with vast amounts of stroke imaging data and a large number of multidisciplinary approaches used in classification, risk assessment, segmentation tasks, diagnosis, prognosis, and even prediction of therapy responses. This article addresses this topic and seeks to present an overview of machine learning and/or deep learning applied to stroke imaging.

    View details for DOI 10.1016/j.nic.2020.07.001

    View details for PubMedID 33038998

  • The Potential Utility of Arterial Spin Labeling in Detecting and Localizing Posterior Circulation Occlusions in Every Day Practice: A Clinical Report of Selected Cases. Journal of clinical imaging science Yedavalli, V., Tong, E. 2020; 10: 78

    Abstract

    Arterial spin labeling (ASL) is a non-contrast, non-invasive method used for the evaluation of cerebral perfusion, which is now increasingly utilized in everyday clinical practice. As a marker of cerebral blood flow at the capillary level, it has particular utility in stroke assessment. One rarer stroke subtype with non-specific symptomatology that can lead to significant morbidity is the posterior circulation (PC) infarct. As with the more common anterior circulation infarcts, ASL has shown benefit in PC infarcts as well, but has not been extensively explored in the literature nor been directly compared to bolus perfusion techniques. This clinical report of selected cases shows the utility of ASL in localization and detection of PC infarcts both in conjunction with and in the absence of bolus perfusion.

    View details for DOI 10.25259/JCIS_118_2020

    View details for PubMedID 33365200

  • Artificial intelligence in stroke imaging: Current and future perspectives. Clinical imaging Yedavalli, V. S., Tong, E. n., Martin, D. n., Yeom, K. W., Forkert, N. D. 2020; 69: 246–54

    Abstract

    Artificial intelligence (AI) is a fast-growing research area in computer science that aims to mimic cognitive processes through a number of techniques. Supervised machine learning, a subfield of AI, includes methods that can identify patterns in high-dimensional data using labeled 'ground truth' data and apply these learnt patterns to analyze, interpret, or make predictions on new datasets. Supervised machine learning has become a significant area of interest within the medical community. Radiology and neuroradiology in particular are especially well suited for application of machine learning due to the vast amount of data that is generated. One devastating disease for which neuroimaging plays a significant role in the clinical management is stroke. Within this context, AI techniques can play pivotal roles for image-based diagnosis and management of stroke. This overview focuses on the recent advances of artificial intelligence methods - particularly supervised machine learning and deep learning - with respect to workflow, image acquisition and reconstruction, and image interpretation in patients with acute stroke, while also discussing potential pitfalls and future applications.

    View details for DOI 10.1016/j.clinimag.2020.09.005

    View details for PubMedID 32980785

  • Deep learning enables automatic detection and segmentation of brain metastases on multisequence MRI. Journal of magnetic resonance imaging : JMRI Grøvik, E. n., Yi, D. n., Iv, M. n., Tong, E. n., Rubin, D. n., Zaharchuk, G. n. 2019

    Abstract

    Detecting and segmenting brain metastases is a tedious and time-consuming task for many radiologists, particularly with the growing use of multisequence 3D imaging.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).Retrospective.In all, 156 patients with brain metastases from several primary cancers were included.5T and 3T.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).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.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.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).A deep-learning approach using multisequence MRI can automatically detect and segment brain metastases with high accuracy.3 Technical Efficacy Stage: 2 J. Magn. Reson. Imaging 2019.

    View details for PubMedID 31050074

  • Time-resolved CT assessment of collaterals as imaging biomarkers to predict clinical outcomes in acute ischemic stroke NEURORADIOLOGY Tong, E., Patrie, J., Tong, S., Evans, A., Michel, P., Eskandari, A., Wintermark, M. 2017; 59 (11): 1101-1109

    Abstract

    Collateral circulation plays a pivotal role in the pathophysiology of acute ischemic stroke and is increasingly recognized as a promising biomarker for predicting the clinical outcome. However, there is no single established grading system. We designed a novel machine-learning software that allows non-invasive, objective, and quantitative assessment of collaterals according to their vascular territories. Our goal is to investigate the prognostic and predictive value of this collateral score for the prediction of acute stroke outcome.This is a retrospective study of 135 patients with anterior circulation stroke treated with IV TPA. An equation using this collateral score (adjusting for age, baseline NIHSS, and recanalization) was derived to predict the clinical outcome (90-day mRS). The primary analyses focused on determining the prognostic value of our newly developed collateral scores. Secondary analyses examined the interrelationships between the collateral score and other variables.The collateral score emerged as a statistically significant prognostic biomarker for good clinical outcome (p < 0.033) among recanalized patients, but not among non-recanalized patients (p < 0.497). Our results also showed that collateral score was a predictive biomarker (p < 0.044). These results suggest that (1) patients with good collateral score derive more benefit from successful recanalization than patients with poor collateral score and (2) collateral status is inconsequential if recanalization is not achieved.Our data results reinforce the importance of careful patient selection for recanalization therapy to avoid futile recanalization. The paucity of collaterals predicts poor clinical outcome despite recanalization. On the other hand, robust collaterals warrant consideration for recanalization therapy given the better odds of good clinical outcome.

    View details for DOI 10.1007/s00234-017-1914-z

    View details for Web of Science ID 000412758900009

    View details for PubMedID 28864854

  • Understanding the Neurophysiology and Quantification of Brain Perfusion. Topics in magnetic resonance imaging Tong, E., Sugrue, L., Wintermark, M. 2017; 26 (2): 57-65

    Abstract

    Newer neuroimaging technology has moved beyond pure anatomical imaging and ventured into functional and physiological imaging. Perfusion magnetic resonance imaging (PWI), which depicts hemodynamic conditions of the brain at the microvascular level, has an increasingly important role in clinical central nervous system applications. This review provides an overview of the established role of PWI in brain tumor and cerebrovascular imaging, as well as some emerging applications in neuroimaging. PWI allows better characterization of brain tumors, grading, and monitoring. In acute stroke imaging, PWI is utilized to distinguish penumbra from infarcted tissue. PWI is a promising tool in the assessment of neurodegenerative and neuropsychiatric diseases, although its clinical role is not yet defined.

    View details for DOI 10.1097/RMR.0000000000000128

    View details for PubMedID 28277465

  • One-stop-shop stroke imaging with functional CT EUROPEAN JOURNAL OF RADIOLOGY Tong, E., Komlosi, P., Wintermark, M. 2015; 84 (12): 2425-2431

    Abstract

    Advanced imaging techniques have extended beyond traditional anatomic imaging and progressed to dynamic, physiologic and functional imaging. Neuroimaging is no longer a mere diagnostic tool. Multimodal functional CT, comprising of NCCT, PCT and CTA, provides a one-stop-shop for rapid stroke imaging. Integrating those imaging findings with pertinent clinical information can help guide subsequent treatment decisions, medical management and follow-up imaging selection. This review article will briefly discuss the indication and utility of each modality in acute stroke imaging.

    View details for DOI 10.1016/j.ejrad.2014.11.027

    View details for Web of Science ID 000367357700011

  • One-stop-shop stroke imaging with functional CT. European journal of radiology Tong, E., Komlosi, P., Wintermark, M. 2015; 84 (12): 2425-31

    Abstract

    Advanced imaging techniques have extended beyond traditional anatomic imaging and progressed to dynamic, physiologic and functional imaging. Neuroimaging is no longer a mere diagnostic tool. Multimodal functional CT, comprising of NCCT, PCT and CTA, provides a one-stop-shop for rapid stroke imaging. Integrating those imaging findings with pertinent clinical information can help guide subsequent treatment decisions, medical management and follow-up imaging selection. This review article will briefly discuss the indication and utility of each modality in acute stroke imaging.

    View details for DOI 10.1016/j.ejrad.2014.11.027

    View details for PubMedID 25554006