Faculty Bibliography

OBJECTIVE. The objective of this article is to show how artificial intelligence (AI) has impacted different components of the imaging value chain thus far as well as to describe its potential future uses. CONCLUSION. The use of AI has the potential to greatly enhance every component of the imaging value chain. From assessing the appropriateness of imaging orders to helping predict patients at risk for fracture, AI can increase the value that musculoskeletal imagers provide to their patients and to referring clinicians by improving image quality, patient centricity, imaging efficiency, and diagnostic accuracy.

OBJECTIVE:The purpose of this study is to increase the value of MRI by reengineering the MRI workflow at a new imaging center to shorten the interval (i.e., turnaround time) between each patient examination by at least 5 minutes. MATERIALS AND METHODS/METHODS:The elements of the MRI workflow that were optimized included the use of dockable tables, the location of patient preparation rooms, the number of doors per scanning room, and the storage location and duplication of coils. Turnaround times at the new center and at two existing centers were measured both for all patients and for situations when the next patient was ready to be brought into the scanner room after the previous patient's examination was completed. RESULTS:Workflow optimizations included the use of dockable tables, dedicated patient preparation rooms, two doors in each MRI room, positioning the scanner to provide the most direct path to the scanner, and coil storage in the preparation rooms, with duplication of the most frequently used coils. At the new imaging center, the median and mean (± SD) turnaround times for situations in which patients were ready for scanning were 115 seconds (95% CI, 113-117 seconds) and 132 ± 72 seconds (95% CI, 129-135 seconds), respectively, and the median and mean turnaround times for all situations were 141 seconds (95% CI, 137-146 seconds) and 272 ± 270 seconds (95% CI, 263-282 seconds), respectively. For existing imaging centers, the median and mean turnaround times for situations in which patients were ready for scanning were 430 seconds (95% CI, 424-434 seconds) and 460 ± 156 seconds (95% CI, 455-465 seconds), respectively, and the median and mean turnaround times for all situations were 481 seconds (95% CI, 474-486 seconds) and 537 ± 219 seconds (95% CI, 532-543 seconds), respectively. CONCLUSION/CONCLUSIONS:The optimized MRI workflow resulted in a mean time savings of 5 minutes 28 seconds per patient.

PURPOSE/OBJECTIVE:Although deep learning has shown great promise for MR image reconstruction, an open question regarding the success of this approach is the robustness in the case of deviations between training and test data. The goal of this study is to assess the influence of image contrast, SNR, and image content on the generalization of learned image reconstruction, and to demonstrate the potential for transfer learning. METHODS:Reconstructions were trained from undersampled data using data sets with varying SNR, sampling pattern, image contrast, and synthetic data generated from a public image database. The performance of the trained reconstructions was evaluated on 10 in vivo patient knee MRI acquisitions from 2 different pulse sequences that were not used during training. Transfer learning was evaluated by fine-tuning baseline trainings from synthetic data with a small subset of in vivo MR training data. RESULTS:Deviations in SNR between training and testing led to substantial decreases in reconstruction image quality, whereas image contrast was less relevant. Trainings from heterogeneous training data generalized well toward the test data with a range of acquisition parameters. Trainings from synthetic, non-MR image data showed residual aliasing artifacts, which could be removed by transfer learning-inspired fine-tuning. CONCLUSION/CONCLUSIONS:This study presents insights into the generalization ability of learned image reconstruction with respect to deviations in the acquisition settings between training and testing. It also provides an outlook for the potential of transfer learning to fine-tune trainings to a particular target application using only a small number of training cases.

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

Provided methodologic training, more imagers can contribute to the evidence basis on improved health outcomes and value in diagnostic imaging. The Value of Imaging Through Comparative Effectiveness Research Program was developed to provide hands-on, practical training in five core areas for comparative effectiveness and big biomedical data research: decision analysis, cost-effectiveness analysis, evidence synthesis, big data principles, and applications of big data analytics. The program's mixed format consists of web-based modules for asynchronous learning as well as in-person sessions for practical skills and group discussion. Seven diagnostic radiology subspecialties and cardiology are represented in the first group of program participants, showing the collective potential for greater depth of comparative effectiveness research in the imaging community.

Radiologists are facing increasing workplace pressures that can lead to decreased job satisfaction and burnout. The increasing complexity and volumes of cases and increasing numbers of noninterpretive tasks, compounded by decreasing reimbursements and visibility in this digital age, have created a critical need to develop innovations that optimize workflow, increase radiologist engagement, and enhance patient care. During their workday, radiologists often must navigate through multiple software programs, including picture archiving and communication systems, electronic health records, and dictation software. Furthermore, additional noninterpretive duties can interrupt image review. Fragmented data and frequent task switching can create frustration and potentially affect patient care. Despite the current successful technological advancements across industries, radiology software systems often remain nonintegrated and not leveraged to their full potential. Each step of the imaging process can be enhanced with use of information technology (IT). Successful implementation of IT innovations requires a collaborative team of radiologists, IT professionals, and software programmers to develop customized solutions. This article includes a discussion of how IT tools are used to improve many steps of the imaging process, including examination protocoling, image interpretation, reporting, communication, and radiologist feedback. ^©RSNA, 2018.

PURPOSE: To allow fast and high-quality reconstruction of clinical accelerated multi-coil MR data by learning a variational network that combines the mathematical structure of variational models with deep learning. THEORY AND METHODS: Generalized compressed sensing reconstruction formulated as a variational model is embedded in an unrolled gradient descent scheme. All parameters of this formulation, including the prior model defined by filter kernels and activation functions as well as the data term weights, are learned during an offline training procedure. The learned model can then be applied online to previously unseen data. RESULTS: The variational network approach is evaluated on a clinical knee imaging protocol for different acceleration factors and sampling patterns using retrospectively and prospectively undersampled data. The variational network reconstructions outperform standard reconstruction algorithms, verified by quantitative error measures and a clinical reader study for regular sampling and acceleration factor 4. CONCLUSION: Variational network reconstructions preserve the natural appearance of MR images as well as pathologies that were not included in the training data set. Due to its high computational performance, that is, reconstruction time of 193 ms on a single graphics card, and the omission of parameter tuning once the network is trained, this new approach to image reconstruction can easily be integrated into clinical workflow. Magn Reson Med, 2017. (c) 2017 International Society for Magnetic Resonance in Medicine.

OBJECTIVES AND RATIONALE/UNASSIGNED:Radiology-pathology correlation is time-consuming and is not feasible in most clinical settings, with the notable exception of breast imaging. The purpose of this study was to determine if an automated radiology-pathology report pairing system could accurately match radiology and pathology reports, thus creating a feedback loop allowing for more frequent and timely radiology-pathology correlation. METHODS:An experienced radiologist created a matching matrix of radiology and pathology reports. These matching rules were then exported to a novel comprehensive radiology-pathology module. All distinct radiology-pathology pairings at our institution from January 1, 2016 to July 1, 2016 were included (n = 8999). The appropriateness of each radiology-pathology report pairing was scored as either "correlative" or "non-correlative." Pathology reports relating to anatomy imaged in the specific imaging study were deemed correlative, whereas pathology reports describing anatomy not imaged with the particular study were denoted non-correlative. RESULTS:Overall, there was 88.3% correlation (accuracy) of the radiology and pathology reports (n = 8999). Subset analysis demonstrated that computed tomography (CT) abdomen/pelvis, CT head/neck/face, CT chest, musculoskeletal CT (excluding spine), mammography, magnetic resonance imaging (MRI) abdomen/pelvis, MRI brain, musculoskeletal MRI (excluding spine), breast MRI, positron emission tomography (PET), breast ultrasound, and head/neck ultrasound all demonstrated greater than 91% correlation. When further stratified by imaging modality, CT, MRI, mammography, and PET demonstrated excellent correlation (greater than 96.3%). Ultrasound and non-PET nuclear medicine studies demonstrated poorer correlation (80%). CONCLUSION/CONCLUSIONS:There is excellent correlation of radiology imaging reports and appropriate pathology reports when matched by organ system. Rapid, appropriate radiology-pathology report pairings provide an excellent opportunity to close feedback loop to the interpreting radiologist.