PENet: Continuous-Valued Pulmonary Edema Severity Prediction On Chest X-ray Using Siamese Convolutional Networks.

For physicians to make rapid clinical decisions for patients with congestive heart failure, the assessment of pulmonary edema severity in chest radiographs is vital. Although deep learning has shown promise in detecting the presence or absence or discrete grades of severity, of such edema, prediction of continuous-valued severity yet remains a challenge. Here, we propose PENet: Siamese convolutional neural networks to assess the continuous spectrum of severity of lung edema from chest radiographs. We present different modes of implementing this network and demonstrate that our best model outperforms that of earlier work (mean AUC of 0.91 over 0.87), while using only 1/16-th the dimension of input images and 1/69-th the size of training data, thus also saving expensive computation.

Impact of Follow-Up Imaging Recommendation Specificity on Adherence.

Radiology reports often contain follow-up imaging recommendations, but failure to comply with them in a timely manner can lead to delayed treatment, poor patient outcomes, complications, and legal liability. Using a dataset containing 2,972,164 exams for over 7 years, in this study we explored the association between recommendation specificity on follow-up rates. Our results suggest that explicitly mentioning the follow-up interval as part of a follow-up imaging recommendation has a significant impact on adherence making these recommendations 3 times more likely (95% CI: 2.95 - 3.05) to be followed-up, while explicit mentioning of the follow-up modality did not have a significant impact. Our findings can be incorporated into routine dictation macros so that the follow-up duration is explicitly mentioned whenever clinically applicable, and/or used as the basis for a quality improvement project focussed on improving adherence to follow-up imaging recommendations.

Deep Learning to Quantify Pulmonary Edema in Chest Radiographs.

PURPOSE: To develop a machine learning model to classify the severity grades of pulmonary edema on chest radiographs. MATERIALS AND METHODS: In this retrospective study, 369 071 chest radiographs and associated radiology reports from 64 581 patients (mean age, 51.71 years; 54.51% women) from the MIMIC-CXR chest radiograph dataset were included. This dataset was split into patients with and without congestive heart failure (CHF). Pulmonary edema severity labels from the associated radiology reports were extracted from patients with CHF as four different ordinal levels: 0, no edema; 1, vascular congestion; 2, interstitial edema; and 3, alveolar edema. Deep learning models were developed using two approaches: a semisupervised model using a variational autoencoder and a pretrained supervised learning model using a dense neural network. Receiver operating characteristic curve analysis was performed on both models. RESULTS: The area under the receiver operating characteristic curve (AUC) for differentiating alveolar edema from no edema was 0.99 for the semisupervised model and 0.87 for the pretrained models. Performance of the algorithm was inversely related to the difficulty in categorizing milder states of pulmonary edema (shown as AUCs for semisupervised model and pretrained model, respectively): 2 versus 0, 0.88 and 0.81; 1 versus 0, 0.79 and 0.66; 3 versus 1, 0.93 and 0.82; 2 versus 1, 0.69 and 0.73; and 3 versus 2, 0.88 and 0.63. CONCLUSION: Deep learning models were trained on a large chest radiograph dataset and could grade the severity of pulmonary edema on chest radiographs with high performance.Supplemental material is available for this article.See also the commentary by Auffermann in this issue.© RSNA, 2021.

Lung-RADS Version 1.0 versus Lung-RADS Version 1.1: Comparison of Categories Using Nodules from the National Lung Screening Trial.

Background The American College of Radiology updated Lung Imaging Reporting and Data System (Lung-RADS) version 1.0 to version 1.1 in May 2019, with the two key changes involving perifissural nodules (PFNs) and ground-glass nodules (GGNs) now designated as a negative screening result. This study examines the effects of these changes using National Lung Screening Trial (NLST) data. Purpose To determine the frequency of PFNs and GGNs reclassified from category 3 or 4A to the more benign category 2 in the updated Lung-RADS version 1.1, as compared with Lung-RADS version 1.0, using CT scans from the NLST. Materials and Methods In this secondary analysis of the NLST, the authors studied all noncalcified nodules (NCNs) found on the incident scan. Nodules were evaluated using criteria from both Lung-RADS version 1.0 and version 1.1, which were compared to determine changes in the number of nodules deemed benign. A McNemar test was used to assess statistical significance. Results A total of 2813 patients (mean age ± standard deviation, 62 years ± 5; 1717 men) with 4408 NCNs were studied. Of the largest 1092 solid NCNs measuring at least 6 mm but less than 10 mm, 216 (19.8%) were deemed PFNs (category 2) using Lung-RADS version 1.1. Eleven of the 1092 solid NCNs (1.0%) were malignant, but none were PFNs. Of 161 GGNs, three (1.9%) were category 3 according to Lung-RADS version 1.0, of which two (66.7%) were down-classified to category 2 with version 1.1. One of the three down-categorized GGNs (version 1.1) proved to be malignant (false-negative finding). Statistically significant improvement for Lung-RADS version 1.1 was found for total nodules (P < .01) and PFNs (P < .01), but not GGNs (P = .48). Conclusion This secondary analysis of National Lung Screening Trial data shows that Lung Imaging Reporting and Data System version 1.1 decreased the number of false-positive results. This was related to the down-classification of perifissural nodules in the range of 6 up to 10 mm. The increase in allowable nodule size for ground-glass nodules in category 2 from 20 mm (version 1.0) to 30 mm (version 1.1) showed no benefit. © RSNA, 2021 See also the editorial by Mayo and Lam in this issue.

Determining Follow-Up Imaging Study Using Radiology Reports.

Radiology reports often contain follow-up imaging recommendations. Failure to comply with these recommendations in a timely manner can lead to delayed treatment, poor patient outcomes, complications, unnecessary testing, lost revenue, and legal liability. The objective of this study was to develop a scalable approach to automatically identify the completion of a follow-up imaging study recommended by a radiologist in a preceding report. We selected imaging-reports containing 559 follow-up imaging recommendations and all subsequent reports from a multi-hospital academic practice. Three radiologists identified appropriate follow-up examinations among the subsequent reports for the same patient, if any, to establish a ground-truth dataset. We then trained an Extremely Randomized Trees that uses recommendation attributes, study meta-data and text similarity of the radiology reports to determine the most likely follow-up examination for a preceding recommendation. Pairwise inter-annotator F-score ranged from 0.853 to 0.868; the corresponding F-score of the classifier in identifying follow-up exams was 0.807. Our study describes a methodology to automatically determine the most likely follow-up exam after a follow-up imaging recommendation. The accuracy of the algorithm suggests that automated methods can be integrated into a follow-up management application to improve adherence to follow-up imaging recommendations. Radiology administrators could use such a system to monitor follow-up compliance rates and proactively send reminders to primary care providers and/or patients to improve adherence.

Automated Tracking of Follow-Up Imaging Recommendations.

OBJECTIVE. Radiology reports often contain follow-up imaging recommendations. Failure to comply with these recommendations in a timely manner can lead to poor patient outcomes, complications, and legal liability. As such, the primary objective of this research was to determine adherence rates to follow-up recommendations. MATERIALS AND METHODS. Radiology-related examination data, including report text, for examinations performed between June 1, 2015, and July 31, 2017, were extracted from the radiology departments at the University of Washington (UW) and Lahey Hospital and Medical Center (LHMC). The UW dataset contained 923,885 examinations, and the LHMC dataset contained 763,059 examinations. A 1-year period was used for detection of imaging recommendations and up to 14-months for the follow-up examination to be performed. RESULTS. On the basis of an algorithm with 97.9% detection accuracy, the follow-up imaging recommendation rate was 11.4% at UW and 20.9% at LHMC. Excluding mammography examinations, the overall follow-up imaging adherence rate was 51.9% at UW (range, 44.4% for nuclear medicine to 63.0% for MRI) and 52.0% at LHMC (range, 30.1% for fluoroscopy to 63.2% for ultrasound) using a matcher algorithm with 76.5% accuracy. CONCLUSION. This study suggests that follow-up imaging adherence rates vary by modality and between sites. Adherence rates can be influenced by various legitimate factors. Having the capability to identify patients who can benefit from patient engagement initiatives is important to improve overall adherence rates. Monitoring of follow-up adherence rates over time and critical evaluation of variation in recommendation patterns across the practice can inform measures to standardize and help mitigate risk.

Vancouver Risk Calculator Compared with ACR Lung-RADS in Predicting Malignancy: Analysis of the National Lung Screening Trial.

Purpose To compare the Vancouver risk calculator (VRC) with American College of Radiology (ACR) Lung Imaging Reporting and Data System (Lung-RADS) in predicting the risk of malignancy in the National Lung Screening Trial (NLST). Materials and Methods A total of 2813 patients with 4408 nodules (4078 solid, 330 subsolid) were available from the NLST for evaluation. Nodules were scored by using VRC with nine parameters (output was the percentage likelihood of malignancy; VRC threshold for malignancy likelihood set as greater than 5%) and Lung-RADS (output was category 2-4B; malignancy defined as category 4A or 4B; malignancy likelihood greater than 5%). Lung-RADS and VRC were compared for sensitivity, specificity, and accuracy for malignancy on a per-nodule and per-patient basis. Results Of 4408 total nodules, 100 of 4078 (2.5%) solid nodules were malignant and 10 of 330 (3%) subsolid nodules were malignant. On an overall per-nodule basis, the sensitivity, specificity, and accuracy for VRC and Lung-RADS were 93%, 90%, and 90% for VRC and 87%, 83%, and 83% for Lung-RADS, respectively (P = .077, P < .001, and P < .001, respectively). On a per-patient basis, the sensitivity, specificity, and accuracy for VRC and Lung-RADS were 93%, 85%, and 85% for VRC and 87%, 76%, and 76% for Lung-RADS, respectively (P = .077, P < .001, and P < .001, respectively). Conclusion The Vancouver risk calculator had superior overall accuracy than the Lung Imaging Reporting and Data System in predicting malignancy in the National Lung Screening Trial for total nodules, as well as on a per-patient basis. © RSNA, 2019 See also the editorial by Black in this issue.

Determining Adherence to Follow-up Imaging Recommendations.

PURPOSE: Radiology reports often contain follow-up imaging recommendations. However, these recommendations are not always followed up by referring physicians and patients. Failure to comply in a timely manner can lead to delayed treatment, poor patient outcomes, unnecessary testing, lost revenue, and legal liability. Therefore, the primary objective of this research was to determine adherence rates to follow-up recommendations. METHODS: We extracted radiology examination-related data, including report text, for examinations performed between January 1, 2010, and February 28, 2017, from the radiology information system at an academic institution. The data set contained 2,972,164 examinations. The first 6 years were used as the period during which a follow-up recommendation was to be detected, allowing for a maximum of 14 months for a follow-up examination to be performed. RESULTS: At least one recommendation for follow-up imaging was present in 10.6% of radiology reports. Overall, the follow-up imaging adherence rate was 58.14%. Mammography had the highest follow-up adherence rate at 69.03%, followed by MRI at 67.54%. Of the modalities, nuclear medicine had the lowest adherence rate at 37.93%. CONCLUSIONS: This study confirms that follow-up imaging adherence rates are inherently low and vary by modality and that appropriate interventions may be needed to improve compliance to follow-up imaging recommendations.

Extracting Follow-Up Recommendations and Associated Anatomy from Radiology Reports.

Adherence rates for timely imaging follow-up are usually low due to low rates of diligence by referring physicians and/or patients with following recommendations for follow-up imaging. This can lead to delayed treatment, poor patient outcomes, unnecessary testing, and legal liability. Existing follow-up recommendation detection methods are often disease- and modality-specific. To address some of these limitations, we present a generic radiology report processing pipeline that can be used to extract follow-up imaging recommendations by anatomy using an ontology-based approach. Using a large dataset from three hospitals, we discuss our methodology in the context of identifying follow-up imaging recommendations that are related to lung, adrenal and/or thyroid conditions. The algorithm has 99% accuracy (95% CI: 95.8-99%). We also present an interactive dashboard that can be used to understand trends related to follow-up recommendations.

Real-time FDG PET guidance during biopsies and radiofrequency ablation using multimodality fusion with electromagnetic navigation.

PURPOSE: To assess the feasibility of combined electromagnetic device tracking and computed tomography (CT)/ultrasonography (US)/fluorine 18 fluorodeoxyglucose (FDG) positron emission tomography (PET) fusion for real-time feedback during percutaneous and intraoperative biopsies and hepatic radiofrequency (RF) ablation. MATERIALS AND METHODS: In this HIPAA-compliant, institutional review board-approved prospective study with written informed consent, 25 patients (17 men, eight women) underwent 33 percutaneous and three intraoperative biopsies of 36 FDG-avid targets between November 2007 and August 2010. One patient underwent biopsy and RF ablation of an FDG-avid hepatic focus. Targets demonstrated heterogeneous FDG uptake or were not well seen or were totally inapparent at conventional imaging. Preprocedural FDG PET scans were rigidly registered through a semiautomatic method to intraprocedural CT scans. Coaxial biopsy needle introducer tips and RF ablation electrode guider needle tips containing electromagnetic sensor coils were spatially tracked through an electromagnetic field generator. Real-time US scans were registered through a fiducial-based method, allowing US scans to be fused with intraprocedural CT and preacquired FDG PET scans. A visual display of US/CT image fusion with overlaid coregistered FDG PET targets was used for guidance; navigation software enabled real-time biopsy needle and needle electrode navigation and feedback. RESULTS: Successful fusion of real-time US to coregistered CT and FDG PET scans was achieved in all patients. Thirty-one of 36 biopsies were diagnostic (malignancy in 18 cases, benign processes in 13 cases). RF ablation resulted in resolution of targeted FDG avidity, with no local treatment failure during short follow-up (56 days). CONCLUSION: Combined electromagnetic device tracking and image fusion with real-time feedback may facilitate biopsies and ablations of focal FDG PET abnormalities that would be challenging with conventional image guidance.

Intra-operative volume imaging of the left atrium and pulmonary veins with rotational X-ray angiography.

Complex electrophysiology (EP) procedures, such as catheter-based ablation in the left atrium and pulmonary veins (LAPV) for treatment of atrial fibrillation, require knowledge of heart chamber anatomy. Electroanatomical mapping (EAM) is typically used to define cardiac structures by combining electromagnetic spatial catheter localization with surface models which interpolate the anatomy between EAM point locations in 3D. Recently, the incorporation of pre-operative volumetric CT or MR data sets has allowed for more detailed maps of LAPV anatomy to be used intra-operatively. Preoperative data sets are however a rough guide since they can be acquired several days to weeks prior to EP intervention. Due to positional and physiological changes, the intra-operative cardiac anatomy can be different from that depicted in the pre-operative data. We present a novel application of contrast-enhanced rotational X-ray imaging for CT-like reconstruction of 3D LAPV anatomy during the intervention itself. We perform two selective contrast-enhanced rotational acquisitions and reconstruct CT-like volumes with 3D filtered back projection. Two volumes depicting the left and right portions of the LAPV are registered and fused. The combined data sets are then visualized and segmented intra-procedurally to provide anatomical data and surface models for intervention guidance. Our results from animal and human experiments indicate that the anatomical information from intra-operative CT-like reconstructions compares favorably with pre-acquired CT data and can be of sufficient quality for intra-operative guidance.