Faculty Bibliography

During the COVID-19 pandemic, rapid and accurate triage of patients at the emergency department is critical to inform decision-making. We propose a data-driven approach for automatic prediction of deterioration risk using a deep neural network that learns from chest X-ray images, and a gradient boosting model that learns from routine clinical variables. Our AI prognosis system, trained using data from 3,661 patients, achieves an AUC of 0.786 (95% CI: 0.742-0.827) when predicting deterioration within 96 hours. The deep neural network extracts informative areas of chest X-ray images to assist clinicians in interpreting the predictions, and performs comparably to two radiologists in a reader study. In order to verify performance in a real clinical setting, we silently deployed a preliminary version of the deep neural network at NYU Langone Health during the first wave of the pandemic, which produced accurate predictions in real-time. In summary, our findings demonstrate the potential of the proposed system for assisting front-line physicians in the triage of COVID-19 patients.

The thalamus consists of several functionally distinct nuclei, some of which serve as targets for functional neurosurgery. Visualization of such nuclei is a major challenge due to their low signal contrast on conventional imaging. We introduce MR susceptibility imaging with a short TE, leveraging susceptibility differences among thalamic nuclei, to automatically delineate 15 thalamic subregions. The technique has the potential to enable direct targeting of thalamic nuclei for functional neurosurgical guidance.

INTRODUCTION/BACKGROUND:In addition to neurogenic orthostatic hypotension (nOH), patients with synucleinopathies frequently have hypertension when supine. The long-term consequences of both abnormalities are difficult to disentangle. We aimed to determine if supine hypertension is associated with target organ damage and worse survival in patients with nOH. METHODS:Patients with nOH due to multiple system atrophy (MSA), Parkinson disease (PD), or pure autonomic failure (PAF) were classified into those with or without supine hypertension (systolic BP of at least 140 mmHg or diastolic BP of at least 90 mmHg). Organ damage was assessed by measuring cerebral white matter hyperintensities (WMH), left ventricular hypertrophy (LVH), and renal function. We prospectively followed patients for 30 months (range: 12-66 months) and recorded incident cardiovascular events and all-cause mortality. RESULTS:Fifty-seven patients (35 with probable MSA, 14 with PD and 8 with PAF) completed all evaluations. In addition to nOH (average fall 35 ± 21/17 ± 14 mmHg, systolic/diastolic, mean ± SD), 38 patients (67%) had supine hypertension (systolic BP > 140 mmHg). Compared to those without hypertension, patients with hypertension had higher blood urea nitrogen levels (P = 0.005), lower estimated glomerular filtration rate (P = 0.008), higher prevalence of LVH (P = 0.040), and higher WMH volume (P = 0.019). Longitudinal follow-up of patients for over 2 years (27.1 ± 14.5 months) showed that supine hypertension was independently associated with earlier incidence of cardiovascular events and death (HR = 0.25; P = 0.039). CONCLUSIONS:Supine hypertension in patients with nOH was associated with an increased risk for target organ damage, cardiovascular events, and premature death. Defining management strategies and safe blood pressure ranges in patients with nOH remains an important research question.

Coronavirus disease 2019 (COVID-19) has been reported in association with a variety of brain imaging findings such as ischemic infarct, hemorrhage, and acute hemorrhagic necrotizing encephalopathy. Here, we report brain imaging features in 11 critically ill COVID-19 patients with persistently depressed mental status who underwent MRI between April 5-25, 2020 at our institution. These features include, 1) Confluent T2 hyperintensity and mild restricted diffusion in bilateral supratentorial deep and subcortical white matter (in 10 of 11 patients), and 2) multiple punctate microhemorrhages in juxtacortical and callosal white matter (in 7 of 11 patients). We also discuss potential pathogeneses.

A publicly available dataset containing k-space data as well as Digital Imaging and Communications in Medicine image data of knee images for accelerated MR image reconstruction using machine learning is presented.

Quantitative, volumetric analysis of Magnetic Resonance Imaging (MRI) is a fundamental way researchers study the brain in a host of neurological conditions including normal maturation and aging. Despite the availability of open-source brain segmentation software, widespread clinical adoption of volumetric analysis has been hindered due to processing times and reliance on manual corrections. Here, we extend the use of deep learning models from proof-of-concept, as previously reported, to present a comprehensive segmentation of cortical and deep gray matter brain structures matching the standard regions of aseg+ aparc included in the commonly used open-source tool, Freesurfer. The work presented here provides a real-life, rapid deep learning-based brain segmentation tool to enable clinical translation as well as research application of quantitative brain segmentation. The advantages of the presented tool include short (~ 1 minute) processing time and improved segmentation quality. This is the first study to perform quick and accurate segmentation of 102 brain regions based on the surface-based protocol (DMK protocol), widely used by experts in the field. This is also the first work to include an expert reader study to assess the quality of the segmentation obtained using a deep-learning-based model. We show the superior performance of our deep-learning-based models over the traditional segmentation tool, Freesurfer. We refer to the proposed deep learning-based tool as DARTS (DenseUnet-based Automatic Rapid Tool for brain Segmentation)

Accelerating Magnetic Resonance Imaging (MRI) by taking fewer measurements has the potential to reduce medical costs, minimize stress to patients and make MRI possible in applications where it is currently prohibitively slow or expensive. We introduce the fastMRI dataset, a large-scale collection of both raw MR measurements and clinical MR images, that can be used for training and evaluation of machine-learning approaches to MR image reconstruction. By introducing standardized evaluation criteria and a freely-accessible dataset, our goal is to help the community make rapid advances in the state of the art for MR image reconstruction. We also provide a self-contained introduction to MRI for machine learning researchers with no medical imaging background

We develop and evaluate a neural network-based method for Gibbs artifact and noise removal. A convolutional neural network (CNN) was designed for artifact removal in diffusion-weighted imaging data. Two implementations were considered: one for magnitude images and one for complex images. Both models were based on the same encoder-decoder structure and were trained by simulating MRI acquisitions on synthetic non-MRI images. Both machine learning methods were able to mitigate artifacts in diffusion-weighted images and diffusion parameter maps. The CNN for complex images was also able to reduce artifacts in partial Fourier acquisitions. The proposed CNNs extend the ability of artifact correction in diffusion MRI. The machine learning method described here can be applied on each imaging slice independently, allowing it to be used flexibly in clinical applications

Introduction: Although often characterized as a disease of white matter, gray matter (GM) pathology has been shown to play an important role in multiple sclerosis (MS).
Objective(s): We used diffusion kurtosis imaging (DKI), a clinically feasible extension of diffusion tensor imaging (DTI) to characterize pathology in cortical and subcortical GM regions in MS patients compared to controls and study how selected DKI parameters correlate with disease severity in comparison to traditional volumetric approaches.
Method(s): 36 MS patients and 24 age and gender matched controls were enrolled in the study. MS patients completed a Patient Determined Disease Steps Score (PDDS). All patients received MRI on a 3T MR Scanner (Siemens, Skyra, or Prisma), which included whole brain 3D magnetization-prepared rapid gradientecho (MPRAGE) (1 mm³ isotropic resolution) for extracting volumetrics and monopolar diffusion-weighted echo-planar imaging (EPI) (voxel size = 1.7 x 1.7 x 3 mm³, b=0, 250, 1000, and 2000 s/m² along 84 directions, TE/TR = 100/3500 ms, GRAPPA with acceleration 2, and multiband 2) for deriving diffusion metrics. Volume metrics from automatic segmentation from MPRAGE and diffusion metrics which included mean diffusivity (MD), mean kurtosis (MK), and fractional anisotropy (FA) were derived for 7 subcortical and 5 cortical GM regions. We determined the partial correlations between PDDS and either GM volume or diffusion parameters covarying for gender and age. We also determined the differences in volume and diffusion metrics between MS patients and controls using ANCOVA with age as the covariate.
Result(s): We observed statistically significant differences in volumes between MS patients and controls for the amygdala, caudate, putamen, nucleus accumbens, cingulate lobe, and subcortical gray volumes with p-values ranging from 0.001 to 0.044. Statistically significant group differences were observed in a majority of the ROI for FA, MD, and MK. Overall, FA was increased, MD was increased, and MK was decreased for most ROI in MS patients compared to controls. There was an increased number of significant partial correlations between PDDS and diffusion metrics compared to PDDS and volume metrics, specifically positive correlations for occipital lobe MD and FA and negative correlations for hippocampal FA.
Conclusion(s): Our results suggest that DKI metrics are sensitive to changes in GM and complimentary to GM volumetrics as an index of GM pathology