German CheXpert Chest X-ray Radiology Report Labeler

PURPOSE

 The aim of this study was to develop an algorithm to automatically extract annotations from German thoracic radiology reports to train deep learning-based chest X-ray classification models.

MATERIALS AND METHODS

 An automatic label extraction model for German thoracic radiology reports was designed based on the CheXpert architecture. The algorithm can extract labels for twelve common chest pathologies, the presence of support devices, and "no finding". For iterative improvements and to generate a ground truth, a web-based multi-reader annotation interface was created. With the proposed annotation interface, a radiologist annotated 1086 retrospectively collected radiology reports from 2020-2021 (data set 1). The effect of automatically extracted labels on chest radiograph classification performance was evaluated on an additional, in-house pneumothorax data set (data set 2), containing 6434 chest radiographs with corresponding reports, by comparing a DenseNet-121 model trained on extracted labels from the associated reports, image-based pneumothorax labels, and publicly available data, respectively.

RESULTS

 Comparing automated to manual labeling on data set 1: "mention extraction" class-wise F1 scores ranged from 0.8 to 0.995, the "negation detection" F1 scores from 0.624 to 0.981, and F1 scores for "uncertainty detection" from 0.353 to 0.725. Extracted pneumothorax labels on data set 2 had a sensitivity of 0.997 [95 % CI: 0.994, 0.999] and specificity of 0.991 [95 % CI: 0.988, 0.994]. The model trained on publicly available data achieved an area under the receiver operating curve (AUC) for pneumothorax classification of 0.728 [95 % CI: 0.694, 0.760], while the models trained on automatically extracted labels and on manual annotations achieved values of 0.858 [95 % CI: 0.832, 0.882] and 0.934 [95 % CI: 0.918, 0.949], respectively.

CONCLUSION

 Automatic label extraction from German thoracic radiology reports is a promising substitute for manual labeling. By reducing the time required for data annotation, larger training data sets can be created, resulting in improved overall modeling performance. Our results demonstrated that a pneumothorax classifier trained on automatically extracted labels strongly outperformed the model trained on publicly available data, without the need for additional annotation time and performed competitively compared to manually labeled data.

KEY POINTS

  · An algorithm for automatic German thoracic radiology report annotation was developed.. · Automatic label extraction is a promising substitute for manual labeling.. · The classifier trained on extracted labels outperformed the model trained on publicly available data..

ZITIERWEISE

· Wollek A, Hyska S, Sedlmeyr T et al. German CheXpert Chest X-ray Radiology Report Labeler. Fortschr Röntgenstr 2024; 196: 956 - 965.

Assessing the performance of AI-assisted technicians in liver segmentation, Couinaud division, and lesion detection: a pilot study

BACKGROUND

In patients with primary and secondary liver cancer, the number and sizes of lesions, their locations within the Couinaud segments, and the volume and health status of the future liver remnant are key for informing treatment planning. Currently this is performed manually, generally by trained radiologists, who are seeing an inexorable growth in their workload. Integrating artificial intelligence (AI) and non-radiologist personnel into the workflow potentially addresses the increasing workload without sacrificing accuracy. This study evaluated the accuracy of non-radiologist technicians in liver cancer imaging compared with radiologists, both assisted by AI.

METHODS

Non-contrast T1-weighted MRI data from 18 colorectal liver metastasis patients were analyzed using an AI-enabled decision support tool that enables non-radiology trained technicians to perform key liver measurements. Three non-radiologist, experienced operators and three radiologists performed whole liver segmentation, Couinaud segment segmentation, and the detection and measurements of lesions aided by AI-generated delineations. Agreement between radiologists and non-radiologists was assessed using the intraclass correlation coefficient (ICC). Two additional radiologists adjudicated any lesion detection discrepancies.

RESULTS

Whole liver volume showed high levels of agreement between the non-radiologist and radiologist groups (ICC = 0.99). The Couinaud segment volumetry ICC range was 0.77-0.96. Both groups identified the same 41 lesions. As well, the non-radiologist group identified seven more structures which were also confirmed as lesions by the adjudicators. Lesion diameter categorization agreement was 90%, Couinaud localization 91.9%. Within-group variability was comparable for lesion measurements.

CONCLUSION

With AI assistance, non-radiologist experienced operators showed good agreement with radiologists for quantifying whole liver volume, Couinaud segment volume, and the detection and measurement of lesions in patients with known liver cancer. This AI-assisted non-radiologist approach has potential to reduce the stress on radiologists without compromising accuracy.

Worldwide differences in surgeon intraoperative practices for cochlear implantation

OBJECTIVE

To characterize practice patterns of intraoperative imaging and/or functional confirmation of cochlear implant electrode location worldwide.

METHODS

A cross-sectional survey of otolaryngologists performing cochlear implantation was conducted between March 1 and May 6, 2023. Participants were recruited worldwide using an international otologic society membership email list and at professional meetings. Ninety-seven of the 125 invited participants (78%) completed the survey. Participants were categorized by continent.

RESULTS

North American surgeons use intraoperative X-rays more frequently than surgeons in Europe and Asia (p < 0.001). Otolaryngologists in Europe and Asia more frequently use no intraoperative imaging (p = 0.02). There is no regional difference between the intraoperative use of electrophysiologic instruments. European and Asian surgeons implant MED-EL devices (p = 0.012) more frequently than North American surgeons, who more frequently use Cochlear Corporation devices (p = 0.003). MED-EL use is related to less frequent intraoperative X-ray use (p = 0.02). Advanced Bionics use is related to more frequent intraoperative CT use (p = 0.03). No significant association existed between years of practice, number of cochlear implantation surgeries performed yearly, volume of pediatric CI practice, and use of intraoperative tools.

CONCLUSION

Intraoperative practice for radiologic and functional verification of cochlear implant electrode positioning varies worldwide. Practice guidelines may help establish a standard of care for cochlear implantation.

Out-of-distribution detection with in-distribution voting using the medical example of chest x-ray classification

BACKGROUND

Deep learning models are being applied to more and more use cases with astonishing success stories, but how do they perform in the real world? Models are typically tested on specific cleaned data sets, but when deployed in the real world, the model will encounter unexpected, out-of-distribution (OOD) data.

PURPOSE

To investigate the impact of OOD radiographs on existing chest x-ray classification models and to increase their robustness against OOD data.

METHODS

The study employed the commonly used chest x-ray classification model, CheXnet, trained on the chest x-ray 14 data set, and tested its robustness against OOD data using three public radiography data sets: IRMA, Bone Age, and MURA, and the ImageNet data set. To detect OOD data for multi-label classification, we proposed in-distribution voting (IDV). The OOD detection performance is measured across data sets using the area under the receiver operating characteristic curve (AUC) analysis and compared with Mahalanobis-based OOD detection, MaxLogit, MaxEnergy, self-supervised OOD detection (SS OOD), and CutMix.

RESULTS

Without additional OOD detection, the chest x-ray classifier failed to discard any OOD images, with an AUC of 0.5. The proposed IDV approach trained on ID (chest x-ray 14) and OOD data (IRMA and ImageNet) achieved, on average, 0.999 OOD AUC across the three data sets, surpassing all other OOD detection methods. Mahalanobis-based OOD detection achieved an average OOD detection AUC of 0.982. IDV trained solely with a few thousand ImageNet images had an AUC 0.913, which was considerably higher than MaxLogit (0.726), MaxEnergy (0.724), SS OOD (0.476), and CutMix (0.376).

CONCLUSIONS

The performance of all tested OOD detection methods did not translate well to radiography data sets, except Mahalanobis-based OOD detection and the proposed IDV method. Consequently, training solely on ID data led to incorrect classification of OOD images as ID, resulting in increased false positive rates. IDV substantially improved the model's ID classification performance, even when trained with data that will not occur in the intended use case or test set (ImageNet), without additional inference overhead or performance decrease in the target classification. The corresponding code is available at https://gitlab.lrz.de/IP/a-knee-cannot-have-lung-disease.

Demands for medical imaging and workforce Size: A nationwide population-based Study, 2000-2020

PURPOSE

The aim of this study was to investigate associations between workforce and workload among radiologists in Taiwan.

MATERIALS AND METHODS

Data for the period 2000-2020 describing the demand for imaging services and radiologists have been obtained from databases and statistical reports of the Ministry of Health and Welfare. The future demand for radiologists was based on Taiwanese people aged 40 and over.

RESULTS

The workforce of Taiwan's radiologists has increased by 6 % annually over the past 20 years (from 450 to 993), performing 2125, 3202 and 3620 monthly examinations (mainly conventional radiography and CT) in medical centers, regional hospitals and district hospitals. Between 2000 and 2020, the use of CT and MRI increased by more than 3.5 times. Demand for interventional radiology also increased by 1.77 times, 2.25 times, and 5 times, respectively. To maintain this volume of services in 2040, at least 1168 radiologists are needed, about 1.18 times more in 2020.

CONCLUSION

Taiwan has 2.4 to 2.9 times fewer radiologists than the United States and 3 times fewer than Europe, while the annual workload is approximately 2 to 3.4 times greater than that of the United States and 1.4 to 2.5 times greater than that of the United Kingdom. This report may serve as a reference for policy makers who address the challenges of the growing workload among radiologists in countries of similar situations.

Systematic review of artificial intelligence development and evaluation for MRI diagnosis of knee ligament or meniscus tears

OBJECTIVE

The purpose of this systematic review was to summarize the results of original research studies evaluating the characteristics and performance of deep learning models for detection of knee ligament and meniscus tears on MRI.

MATERIALS AND METHODS

We searched PubMed for studies published as of February 2, 2022 for original studies evaluating development and evaluation of deep learning models for MRI diagnosis of knee ligament or meniscus tears. We summarized study details according to multiple criteria including baseline article details, model creation, deep learning details, and model evaluation.

RESULTS

19 studies were included with radiology departments leading the publications in deep learning development and implementation for detecting knee injuries via MRI. Among the studies, there was a lack of standard reporting and inconsistently described development details. However, all included studies reported consistently high model performance that significantly supplemented human reader performance.

CONCLUSION

From our review, we found radiology departments have been leading deep learning development for injury detection on knee MRIs. Although studies inconsistently described DL model development details, all reported high model performance, indicating great promise for DL in knee MRI analysis.

Framework for cyber risk loss distribution of hospital infrastructure: Bond percolation on mixed random graphs approach

Networks like those of healthcare infrastructure have been a primary target of cyberattacks for over a decade. From just a single cyberattack, a healthcare facility would expect to see millions of dollars in losses from legal fines, business interruption, and loss of revenue. As more medical devices become interconnected, more cyber vulnerabilities emerge, resulting in more potential exploitation that may disrupt patient care and give rise to catastrophic financial losses. In this paper, we propose a structural model of an aggregate loss distribution across multiple cyberattacks on a prototypical hospital network. Modeled as a mixed random graph, the hospital network consists of various patient-monitoring devices and medical imaging equipment as random nodes to account for the variable occupancy of patient rooms and availability of imaging equipment that are connected by bidirectional edges to fixed hospital and radiological information systems. Our framework accounts for the documented cyber vulnerabilities of a hospital's trusted internal network of its major medical assets. To our knowledge, there exist no other models of an aggregate loss distribution for cyber risk in this setting. We contextualize the problem in the probabilistic graph-theoretical framework using a percolation model and combinatorial techniques to compute the mean and variance of the loss distribution for a mixed random network with associated random costs that can be useful for healthcare administrators and cybersecurity professionals to improve cybersecurity management strategies. By characterizing this distribution, we allow for the further utility of pricing cyber risk.

Evaluating Health Disparities in Radiology Practices in New Jersey: Exploring Radiologist Geographical Distribution

OBJECTIVE

This study aimed to determine if a disproportionate number of radiologists practice in high-income versus low-income counties in New Jersey (NJ), identify which vulnerable populations are most in need of more radiologists, and discuss how these relative differences can ultimately influence health outcomes.

METHODS

The NJ Health Care Profile, a database overseen and maintained by the Division of Consumer Affairs, was queried for all actively practicing radiologists within the state of NJ. These results were grouped into diagnostic and interventional radiologists followed by further stratification of physicians based on the counties where they currently practice. The median household income and population size of each county for 2021 were obtained from the US Census database. The ratio of the population size of each county over the number of radiologists in that county was used as a surrogate marker for disparities in patient care within the state and was compared between counties grouped by levels of income.

RESULTS

Of the 1,186 board-certified radiologists actively practicing within the state of NJ, 86% are solely diagnostic radiologists and 14% are interventional radiologists. About 44% of radiologists practice within counties that are within the top one-third of median household income in NJ, 25% practice within counties in the middle one-third, and 31% practice within counties in the bottom one-third.

CONCLUSIONS

There is a disproportionate number of radiologists practicing in high-income counties as opposed to lower-income counties. A contradiction to this trend was noted in three low-income counties: Essex, Camden, and Atlantic County, all of which exhibited low numbers of individuals per radiologist that rivaled those of higher-income counties. This finding is a concrete measure of successful radiologist recruitment efforts within these counties during the past few years to combat the increased prevalence of disease and associated complications that historically marginalized communities tend to disproportionately exhibit. Other low-income counties should look to what Essex, Camden, and Atlantic County have done to increase radiologist recruitment to levels that rival those of high-income areas.

Effect of image resolution on automated classification of chest X-rays

Purpose

Deep learning (DL) models have received much attention lately for their ability to achieve expert-level performance on the accurate automated analysis of chest X-rays (CXRs). Recently available public CXR datasets include high resolution images, but state-of-the-art models are trained on reduced size images due to limitations on graphics processing unit memory and training time. As computing hardware continues to advance, it has become feasible to train deep convolutional neural networks on high-resolution images without sacrificing detail by downscaling. This study examines the effect of increased resolution on CXR classification performance.

Approach

We used the publicly available MIMIC-CXR-JPG dataset, comprising 377,110 high resolution CXR images for this study. We applied image downscaling from native resolution to 2048 × 2048    pixels , 1024 × 1024    pixels , 512 × 512    pixels , and 256 × 256    pixels and then we used the DenseNet121 and EfficientNet-B4 DL models to evaluate clinical task performance using these four downscaled image resolutions.

Results

We find that while some clinical findings are more reliably labeled using high resolutions, many other findings are actually labeled better using downscaled inputs. We qualitatively verify that tasks requiring a large receptive field are better suited to downscaled low resolution input images, by inspecting effective receptive fields and class activation maps of trained models. Finally, we show that stacking an ensemble across resolutions outperforms each individual learner at all input resolutions while providing interpretable scale weights, indicating that diverse information is extracted across resolutions.

Conclusions

This study suggests that instead of focusing solely on the finest image resolutions, multi-scale features should be emphasized for information extraction from high-resolution CXRs.

Machine Learning Augmented Interpretation of Chest X-rays: A Systematic Review

Limitations of the chest X-ray (CXR) have resulted in attempts to create machine learning systems to assist clinicians and improve interpretation accuracy. An understanding of the capabilities and limitations of modern machine learning systems is necessary for clinicians as these tools begin to permeate practice. This systematic review aimed to provide an overview of machine learning applications designed to facilitate CXR interpretation. A systematic search strategy was executed to identify research into machine learning algorithms capable of detecting >2 radiographic findings on CXRs published between January 2020 and September 2022. Model details and study characteristics, including risk of bias and quality, were summarized. Initially, 2248 articles were retrieved, with 46 included in the final review. Published models demonstrated strong standalone performance and were typically as accurate, or more accurate, than radiologists or non-radiologist clinicians. Multiple studies demonstrated an improvement in the clinical finding classification performance of clinicians when models acted as a diagnostic assistance device. Device performance was compared with that of clinicians in 30% of studies, while effects on clinical perception and diagnosis were evaluated in 19%. Only one study was prospectively run. On average, 128,662 images were used to train and validate models. Most classified less than eight clinical findings, while the three most comprehensive models classified 54, 72, and 124 findings. This review suggests that machine learning devices designed to facilitate CXR interpretation perform strongly, improve the detection performance of clinicians, and improve the efficiency of radiology workflow. Several limitations were identified, and clinician involvement and expertise will be key to driving the safe implementation of quality CXR machine learning systems.

Interventional Radiology Workforce Shortages Affecting Small and Rural Practices: A Report of the SIR/ACR Joint Task Force on Recruitment and Retention of Interventional Radiologists to Small and Rural Practices

Radiology practices characterized as small and rural are challenged to recruit and retain interventional radiologists. Lack of access to interventional radiologic services results in a failure to meet the needs of patients, hospitals, and other community stakeholders. Acknowledging this challenge, the ACR's Commission on General, Small, Emergency and/or Rural Practice and Commission on Interventional and Cardiovascular Imaging and the Society of Interventional Radiology partnered to establish a joint task force to study this issue and identify strategies the ACR and the Society of Interventional Radiology should take to improve small and rural practice recruitment and retention of interventional radiologists. This report describes the deliberations and recommendations of the task force.

Effective Utilization of Multiple Convolutional Neural Networks for Chest X-Ray Classification

Out of the numerous types of Medical Imaging modalities available, radiography stands out a bit more than others due to its capabilities of diagnosing diseases and conditions, including life-threatening conditions. Its affordability is another main reason for its prevalence. Chest Radiography holds even higher importance, as it focuses a critical area of the human body. However, interpreting a Chest Radiography image can be challenging and usually done by an experienced Radiologist for accurate results. There are two main issues related to this. One is that in some countries, experienced Radiologists are scarce. The other issue is that the inevitability of human errors in diagnoses. Researchers attempt to use Artificial Intelligence to address these two issues. Most of the existing work incorporates Convolutional Neural Networks for this purpose. This paper presents a novel way of parallelizing multiple architectures of Convolutional Neural Networks focusing on Chest X-ray classification. The paper further presents a comprehensive evaluation of the existing architectures with the parallelized results of them using our method. We used four large-scale datasets, including a non-medical one, for the evaluation of our models. We managed to achieve better accuracy for 9 out 13 and 11 out of 14 labels on our two main evaluation datasets. The paper concludes by presenting the limitations and future improvements possible for the system.

Geographic Variations in Growth of Radiologists and Medicare Enrollees From 2012 to 2019

OBJECTIVE

Analyze changes in the number of Medicare-serving radiologists and Medicare enrollees nationwide and by geographic region and state from 2012 to 2019 to understand variations in allocation of imaging health care services over the past decade.

METHODS

The number of radiologists submitting claims to Medicare was extracted from the CMS Physician and Other Supplier Public Use File Database. The number of Medicare enrollees by state was obtained from the Kaiser Family Foundation. National-, regional-, and state-level changes in rates of growth of radiologists, Medicare enrollees, and radiologists per 100,000 Medicare enrollees from 2012 to 2019 were tabulated.

RESULTS

The overall number of radiologists per 100,000 Medicare enrollees was 79.7 in 2012, increasing to 79.9 in 2019. In 2012, the number of radiologists per 100,000 enrollees was lower than the national average in the South (66.9; 16% lower) and Midwest (79.1; 0.7% lower) and higher in the Northeast (98.3; 23% higher) and West (88.8; 11% higher). In 2019, the number of radiologists per 100,000 enrollees was lower than the national average in the South (69.8; 12% lower) only and was higher in the Midwest (81.4; 1.9% higher), Northeast (99.3; 24% higher), and West (80.2; 0.4% higher). By state, there was a 4.2-fold variation in the number of radiologists per 100,000 Medicare enrollees, ranging from 38.8 in Wyoming to 161.4 in Minnesota (200.5 in Washington, DC).

DISCUSSION

The growth of Medicare-serving radiologists and Medicare enrollees was stable nationally and demonstrated tremendous variations by US region and state. These variations bring to light potential implications for patient access to care and distribution of health care resources.

Temporal Trends in the South African Diagnostic Radiology Workforce (2002-2019)

Background

To facilitate imaging resource planning and address key health targets of the United Nations (UN) 2030 Sustainable Development Goals, accurate data are required on imaging personnel at the country level. Such data are currently limited.

Objectives

This study aims to analyze trends in the number, geographical distribution, and demographics of South African (SA) diagnostic imaging personnel between 2002 and 2019.

Method

A retrospective analysis of the Health Professions Council of South Africa (HPCSA) database of imaging personnel from 2002 to 2019 was done. The total number of personnel and personnel per million people were calculated for the country and for each professional group (radiologist, diagnostic radiographer, and sonographer) by calendar year, province, and demographic profile. Population data were provided by Statistics SA.

Results

The total imaging personnel, number per million people, and national population increased by 283% (3,095 versus 8,753), 119% (68 versus 149/10⁶), and 29% (45.45 versus 58.77/10⁶), respectively.

Diagnostic radiographers constituted more than 80% of the workforce throughout the review period, increasing by 185% (2,540 versus 7,242). Sonographers, the smallest cohort, recorded the highest (49 versus 503; 906%) and radiologists (506 versus 1,007; 99%) the lowest proportional growth.

Although radiologists showed persistent male predominance, the male proportion decreased from 82% to 69%, while that of females increased from 18% to 31%. The average annual percentage increase in female radiologists (14%) was more than three times that of males (4%).

Diagnostic radiographers showed female predominance, but the proportion decreased from 90% to 83%, while that of males increased from 10% to 17%. Sonographers showed overwhelming female predominance (94% versus 92%). The average annual percentage increase in male diagnostic radiographers (21%) was more than double that of females (9%).

In 2002, 48% (n = 1,475) of imaging personnel identified as White, and 15% (n = 467) identified as Black African. By 2019, those identifying as White and Black African were 36% (n = 3,122) and 35% (n = 3,045), respectively.

The Western Cape Province (WCP) maintained the highest overall number of imaging personnel per million people (165 versus 233/10⁶) and Limpopo the lowest (12 versus 54/10⁶). However, Limpopo recorded the highest proportional growth in imaging personnel/10⁶ people (368%) and the WCP the lowest (41%). The differential between the best- and least-resourced provinces thus decreased from 14:1 in 2002 to 4:1 in 2019.

Conclusion

In the review period, the SA imaging workforce has shown substantial expansion and transformation and has assumed a more equitable distribution.

Yield of pelvic CT in emergency department patients undergoing CT torso for generalized or multiple complaints

PURPOSE

To evaluate the utility of pelvic computed tomography (CT) in emergency department (ED) patients undergoing chest CT angiogram (CTA) for chest pain or suspected pulmonary embolism (PE) followed by abdominopelvic CT in the same session for additional multisystem or generalized complaints.

METHODS

This retrospective study included consecutive adult ED patients from January 2017 to December 2019 who underwent CTA for suspected PE followed by portovenous abdominopelvic CT for multisystem or generalized complaints. Patient demographics, vitals, laboratory values, exam indication, malignancy history, and recent surgery/intervention were recorded. CT reports were reviewed for acute chest, abdomen, and/or pelvic pathology.

RESULTS

There were 400 patients with 243 (61%) women and mean age of 59.8 years. Acute pelvic findings were seen in 11% (45/400). In 53% (24/45) of these, pelvic pathology could be diagnosed based on the abdominal portion of the CT. Five percent (21/400) of patients demonstrated isolated acute pelvic findings with 86% of these (18/21) clinically suspected prior to imaging. Acute pelvic pathology was associated with female gender (p = 0.015) and elevated white blood cell count (WBC) (p = 0.03). Specific pelvic CT indications and female gender were significantly associated with (p = 0.02 each) and independent predictors of isolated acute pelvic pathology.

CONCLUSION

In ED patients undergoing chest CTA for chest pain or suspected PE combined with abdominopelvic CT, the presence of acute pelvic-related pathology not visualized on abdominal CT is low. For this ED patient cohort, pelvic CT may not be necessary in men with normal WBC and a low pre-imaging clinical suspicion for acute pelvic pathology.

Nationwide Analysis of Integrated and Independent Interventional Radiology Residency Websites During the COVID-19 Pandemic

OBJECTIVES

To analyze current interventional radiology residency program websites based on validated criteria and highlight areas for improvement during the COVID-19 pandemic.

MATERIALS AND METHODS

ACGME-accredited interventional radiology residency programs were identified from the Society of Interventional Radiology (SIR) public database, including 91 independent and 89 integrated programs. Program Webpages were then evaluated during September and October 2020 based on the presence of 48 criteria, organized into seven main categories including visibility & communication, program information, curriculum information, faculty description, research, recruitment, and salary and benefits. Programs were also evaluated based on region and research ranking. Additionally, 166 programs with accreditation for Early Specialization in Interventional Radiology (ESIR) were assessed for the presence or absence of ESIR pathway acknowledgement on program webpages.

RESULTS

The online search yielded information on all integrated programs (89/89, 100%) and 74 independent programs (74/91, 80.3%). For the ESIR programs, the online search for accreditation acknowledgement yielded 108 programs (108/166, 65%) approved for this pathway. Only seven of the 89 integrated programs met at least 75% of the criteria. Of the 91 independent programs, only one met at least 75% of the criteria. On average, integrated programs met more criteria (25, 52%) than independent programs (17, 36%). When comparing programs based on national rank, the visibility & communication category met more criteria on average than the lower ranked programs (integrated =73% vs. 64%, p = 0.01), (independent = 73% vs. 45%, p = 0.01). When comparing programs regionally, statistical significance was found only in the research category (p = 0.01). When comparing the integrated programs with the independent programs for averages in the 7 categories and the total criteria, statistical significance was found in all categories except facility description: visibility & communication (67.5% vs. 53. 7%, p = 0.01), program information (75.7% vs. 58%, p = 0.01), curriculum information (54.8% vs. 31.4, p = 0.01), research (42.5% vs. 27.5%, p = 0.01), recruitment (42.6% vs. 26.8%, p = 0.01), salary & benefits (47.8% vs. 26.8%, p = 0.01), and total criteria (52% vs. 35.8%, p = 0.01.

CONCLUSION

IR residency programs across the country are proficient in providing curricular, and logistical information online. However, improvement is needed in providing nonacademic highlights unique to programs that can aid in maximizing applicant match and compatibility. The information provided by online resources has the potential to influence residency applicant's program ranking and chosen pathway, particularly during the COVID19 pandemic.

Evolving Radiologist Participation in Medicare Shared Savings Program Accountable Care Organizations

PURPOSE

The aim of this study was to temporally characterize radiologist participation in Medicare Shared Savings Program (MSSP) accountable care organizations (ACOs).

METHODS

Using CMS Physician and Other Supplier Public Use Files, ACO provider-level Research Identifiable Files, and Shared Savings Program ACO Public-Use Files for 2013 through 2018, characteristics of radiologist ACO participation were assessed over time.

RESULTS

Between 2013 and 2018, the percentage of Medicare-participating radiologists affiliated with MSSP ACOs increased from 10.4% to 34.9%. During that time, the share of large ACOs (>20,000 beneficiaries) with participating radiologists averaged 87.0%, and the shares of medium ACOs (10,000-20,000) and small ACOs (<10,000) with participating radiologists rose from 62.5% to 66.0% and from 26.3% to 51.6%, respectively. The number of physicians in MSSP ACOs with radiologists was substantially larger than those without radiologists (mean range across years, 573-945 versus 107-179). Primary care physicians constituted a larger percentage of the physician population for ACOs without radiologists (average across years, 66.3% versus 38.5%), and ACOs with radiologists had a higher rate of specialist representation (56.0% versus 33.7%). Beneficiary age, race, and sex demographics were similar among radiologist-participating versus nonparticipating ACOs.

CONCLUSIONS

In recent years, radiologist participation in MSSP ACOs has increased substantially. ACOs with radiologist participation are large and more diverse in their physician specialty composition. Nonparticipating radiologists should prepare accordingly.

Direct communication between radiologists and patients improves the quality of imaging reports

OBJECTIVES

We investigate in what percentage of cases and to what extent radiological reports change when radiologists directly communicate with patients after imaging examinations.

METHODS

One hundred twenty-two consecutive outpatients undergoing MRI examinations at a single center were prospectively included. Radiological reports of the patients were drafted by two radiologists in consensus using only the clinical information that was made available by the referring physicians. Thereafter, one radiologist talked directly with the patient and recorded the duration of the conversation. Afterwards, the additional information from the patient was used to reevaluate the imaging studies in consensus. The radiologists determined whether the radiological report changed based on additional information and, if yes, to what extent. The degree of change was graded on a 4-point Likert scale (1, non-relevant findings, to 4, highly relevant findings).

RESULTS

Following direct communication (duration 170.9 ± 53.9 s), the radiological reports of 52 patients (42.6%) were changed. Of the 52 patients, the degree of change was classified as grade 1 for 8 patients (15.4 %), grade 2 for 27 patients (51.9%), grade 3 for 13 patients (25%), and grade 4 for 4 patients (7.7%). The reasons leading to changes were missing clinical information in 50 cases (96.2%) and the lack of additional external imaging in 2 cases (3.8%).

CONCLUSIONS

Radiologists should be aware that a lack of accurate information from the clinician can lead to incorrect radiological reports or diagnosis. Radiologists should communicate directly with patients, especially when the provided information is unclear, as it may significantly alter the radiological report.

KEY POINTS

• Direct communication between radiologists and patients for an average of 170's resulted in a change in the radiological reports of 52 patients (42.6%). • Of the 42.6% of cases where the reports were changed, the alterations were highly relevant (grades 3 and 4) in 32.7%, indicating major changes with significant impact towards patient management.

Temporal and geospatial variations among the interventional radiology physician workforce in the United States

OBJECTIVE

To analyze the temporal trends and state-wide geospatial variations in Vascular and Interventional Radiology (VIR) workforce in the United States.

METHODS

The State Physician Workforce Data from the AAMC website was accessed for years 2015, 2017, and 2019. The variables collected for each state included total number of active physicians, total number of physicians per specialty and total number of female physicians in VIR. Comparative data was obtained for vascular surgery (VS), diagnostic radiology (DR), and radiation oncology (RO). The annual growth rate for total physicians and sub-analysis of female physicians in each state was computed for each specialty.

RESULTS

From 2015 to 2019, the total number of active physicians in the United States grew by 1.8% per year. Growth of active physicians in VIR grew by 8.3%, DR 0.06%, VS 4.4%, and RO 1.9% per year. Colorado and Minnesota had the highest growth rate for VIR physicians (15%). VIR physicians per 100,000 people increased from 0.84 (2015) to 1.10 (2019) in the US. In comparison, VS physicians increased from 0.99 (2015) to 1.14 (2019), DR physicians decreased from 8.61 (2015) to 8.43 (2019), and RO physicians grew from 1.48 (2015) to 1.56 (2019). Women represented 6.8% of the VIR workforce in the US in 2019 and increased by a rate of 16% annually in the US from 2015 to 2019. In comparison, the number of women in VS has grown by 21%, DR by 2%, and RO by 2.4% during the same period. The state of Maryland has the highest proportion of women in VIR at 18%.

CONCLUSION

The number of VIR physicians is increasing at a higher rate than the national overall physician growth, and while female VIR physicians makeup a small fraction of the VIR workforce, their numbers have increased at a faster rate than overall VIR physicians.

Radiology residency training in China: results from the first retrospective nationwide survey

Objectives

This was the first study to systematically landscape and examine China’s nationwide standardized residency training in radiology.

Methods

In this retrospective cross-sectional study, we used data from the 2019 national survey of the first two cohorts of 3679 radiology residents who completed training in 2017 and 2018 across all 31 provinces in China. A total of 1163 (32%) residents participated in the survey. Multivariable logistic regression was used to examine the implementation frequency of 24 identified training tasks (categorized into six competencies) by region, demographics, and other residency information.

Results

Among the 1163 respondents, 592 (51%) were trained in the more developed eastern region. Of the 24 identified training tasks, 15 were implemented significantly differently across regions, while the frequency of the most frequently conducted tasks (e.g., CT, MR, and radiograph interpretation and reporting) was consistent. The top 10 tasks all fell into the patient care and medical knowledge competency domains, while other competencies tended to be neglected. We found region and marital status were the most influential factors of training task implementation frequencies. Respondents trained in the northeast and the west were more likely to report, for instance, radiological examination recommendation (OR = 1.91, 95%CI = 1.27–2.88), as “very frequent.” Married respondents were more likely to report first-line night shift as “very frequent” (OR = 1.71, 95%CI = 1.29–2.26).

Conclusions

Despite the fast-win achievements of developing a national radiology residency training program, there is a gap to train quality and homogeneous radiologists across regions. Future improvement should be more tailored to residents’ personal characteristics and emphasize some “soft” competencies (e.g., communication skills).

Recalibration of deep learning models for abnormality detection in smartphone-captured chest radiograph

Image-based teleconsultation using smartphones has become increasingly popular. In parallel, deep learning algorithms have been developed to detect radiological findings in chest X-rays (CXRs). However, the feasibility of using smartphones to automate this process has yet to be evaluated. This study developed a recalibration method to build deep learning models to detect radiological findings on CXR photographs. Two publicly available databases (MIMIC-CXR and CheXpert) were used to build the models, and four derivative datasets containing 6453 CXR photographs were collected to evaluate model performance. After recalibration, the model achieved areas under the receiver operating characteristic curve of 0.80 (95% confidence interval: 0.78-0.82), 0.88 (0.86-0.90), 0.81 (0.79-0.84), 0.79 (0.77-0.81), 0.84 (0.80-0.88), and 0.90 (0.88-0.92), respectively, for detecting cardiomegaly, edema, consolidation, atelectasis, pneumothorax, and pleural effusion. The recalibration strategy, respectively, recovered 84.9%, 83.5%, 53.2%, 57.8%, 69.9%, and 83.0% of performance losses of the uncalibrated model. We conclude that the recalibration method can transfer models from digital CXRs to CXR photographs, which is expected to help physicians' clinical works.

Geographical and Temporal Distribution of Radiologists, Computed Tomography and Magnetic Resonance Scanners in Croatia

The aim of the study was to analyse the temporal and geographic distribution of radiologists, computed tomography and magnetic resonance scanners in Croatia. In this observational study we estimated radiologists’ number per 100,000 population for 1997, 2006, and 2017 and compared private and public CT and MR scanners between 2011 and 2018. We analyzed the availability of radiologists and scanners, and the relationship between the radiological workforce and economic strength among counties. The workforce increased significantly from 1997 to 2017 and was associated with economic strength categories in 2017. In 2018, there were more CT scanners in the public sector, while MR scanners were distributed evenly. In 2011, there was similar distribution of CT and MR between sectors, while in 2018 there were significantly more public CT scanners. Counties with a medical school had significantly more radiologists and MR scanners. The high-to-low ratios per CT and MR were 11 and 8.2, suggesting inequality of health care. Croatia significantly increased its radiological workforce; however, cross-county inequality remained. Counties with higher economic strength and medical schools have better availability of radiologists and equipment. To ensure the sustainable activity of the health care system, a precise estimate of supply and demand of radiology services is needed.

Practice Characteristics of the United States General Radiologist Workforce: Most Generalists Work as Multispecialists

RATIONALE AND OBJECTIVES

While subspecialty radiologists' practice patterns have received recent attention, little is known about the practice patterns of general radiologists. We aim to characterize this group (which represents most US radiologists).

MATERIALS AND METHODS

US radiologists' individual work efforts were assessed using the 2017 Medicare Provider and Other Supplier Public Use File and a previously validated wRVU-weighted claims-based classification system. Using prior criteria, radiologists without >50% work efforts in a single subspecialty were deemed generalists. For this study, a >25% subspecialty work effort threshold was deemed a subspecialty "focus area," and generalists with ≥2 subspecialty focus areas were deemed "multispecialists." Practice characteristics were summarized using various parameters.

RESULTS

Among 12,438 radiologists meeting existing claims-based criteria to be deemed generalists, 85.0% had ≥2 subspecialty focus areas of >25% work effort (i.e., multispecialists), 14.6% had one focus area, and 0.4% had no focus area. The fraction of generalists meeting multispecialist criteria was similar across radiologists' years in practice (range 84.7% to 85.4%), academic vs. nonacademic status (84.9% to 86.6%), and practice size (83.3% to 87.0%). Although general radiologist multispecialization varied geographically, a majority were multispecialists in all states (range 57.6% in VT to 93.9% in WY) and percentages were not associated with state-level population density (r = 0.013; p = 0.926).

CONCLUSION

The large majority of US general radiologists practice as multispecialists, and nearly all have at least one subspecialty focus area. The predominance of general radiologists' multispecialty focus across various practice types and locations supports their role in facilitating patient access to a range of radiologist subspecialties.

MIMIC-CXR, a de-identified publicly available database of chest radiographs with free-text reports

Chest radiography is an extremely powerful imaging modality, allowing for a detailed inspection of a patient's chest, but requires specialized training for proper interpretation. With the advent of high performance general purpose computer vision algorithms, the accurate automated analysis of chest radiographs is becoming increasingly of interest to researchers. Here we describe MIMIC-CXR, a large dataset of 227,835 imaging studies for 65,379 patients presenting to the Beth Israel Deaconess Medical Center Emergency Department between 2011-2016. Each imaging study can contain one or more images, usually a frontal view and a lateral view. A total of 377,110 images are available in the dataset. Studies are made available with a semi-structured free-text radiology report that describes the radiological findings of the images, written by a practicing radiologist contemporaneously during routine clinical care. All images and reports have been de-identified to protect patient privacy. The dataset is made freely available to facilitate and encourage a wide range of research in computer vision, natural language processing, and clinical data mining.

Evaluation of an AI-Based Detection Software for Acute Findings in Abdominal Computed Tomography Scans: Toward an Automated Work List Prioritization of Routine CT Examinations

OBJECTIVE

The aim of this study was to test the diagnostic performance of a deep learning-based triage system for the detection of acute findings in abdominal computed tomography (CT) examinations.

MATERIALS AND METHODS

Using a RIS/PACS (Radiology Information System/Picture Archiving and Communication System) search engine, we obtained 100 consecutive abdominal CTs with at least one of the following findings: free-gas, free-fluid, or fat-stranding and 100 control cases with absence of these findings. The CT data were analyzed using a convolutional neural network algorithm previously trained for detection of these findings on an independent sample. The validation of the results was performed on a Web-based feedback system by a radiologist with 1 year of experience in abdominal imaging without prior knowledge of image findings through both visual confirmation and comparison with the clinically approved, written report as the standard of reference. All cases were included in the final analysis, except those in which the whole dataset could not be processed by the detection software. Measures of diagnostic accuracy were then calculated.

RESULTS

A total of 194 cases were included in the analysis, 6 excluded because of technical problems during the extraction of the DICOM datasets from the local PACS. Overall, the algorithm achieved a 93% sensitivity (91/98, 7 false-negative) and 97% specificity (93/96, 3 false-positive) in the detection of acute abdominal findings. Intra-abdominal free gas was detected with a 92% sensitivity (54/59) and 93% specificity (39/42), free fluid with a 85% sensitivity (68/80) and 95% specificity (20/21), and fat stranding with a 81% sensitivity (42/50) and 98% specificity (48/49). False-positive results were due to streak artifacts, partial volume effects, and a misidentification of a diverticulum (each n = 1).

CONCLUSIONS

The algorithm's autonomous detection of acute pathological abdominal findings demonstrated a high diagnostic performance, enabling guidance of the radiology workflow toward prioritization of abdominal CT examinations with acute conditions.

The Physician Payment Sunshine Act: Evaluating Industrial Payments in Radiology

RATIONALE AND OBJECTIVES

The characterization of payments made to physicians by pharmaceutical companies, device manufacturers, and group purchasing organizations is crucial for assessing potential conflicts of interest and their impact on practice patterns. This study examines the compensation received by general radiologists (GR) in the United States, as well as radiologists in the following five subspecialties: body imaging, neuroradiology, pediatric radiology, nuclear radiology and radiological physics, and vascular and interventional radiology.

MATERIALS AND METHODS

Data were extracted from the Open Payments database for radiology subspecialists in the United States who received installments in calendar year 2015 from pharmaceutical and device manufacturing companies.

RESULTS

In 2015, a total of $43,685,052 was paid in 65,507 payments (mean $667/payment; median $32/payment) to radiologists, including 9826 GR, 362 body imaging radiologists, 479 neuroradiologists, 127 pediatric radiologists, 175 physicians in nuclear radiology and radiological physics, and 1584 vascular and interventional radiologists. Payments were unequally distributed across these six major subspecialties of radiology (p < 0.01), with GR receiving the largest number of total payments (44,695), and neuroradiologists receiving significantly higher median payments than any other subspecialty ($80 vs $32 for all radiologists; p < 0.01). Medtronic Neurovascular was the single largest payer to all radiologists combined.

CONCLUSION

Commercial entities make substantial payments to radiologists, with a significant variation in payments made to the different radiology subspecialties. While the largest number of total payments was made to GGR, the highest median payments were made to neuroradiologists, and significant dispersion in these payments was seen across different geographic regions. The impact of these payments on practice patterns remains to be elucidated.

Global and Japanese regional variations in radiologist potential workload for computed tomography and magnetic resonance imaging examinations

PURPOSE

To investigate the global variation in radiologist potential workload for CT and MRI examinations, and the regional variation in potential workload and extent of radiologists' involvement in CT and MRI examinations in Japan.

METHODS

"Radiologist potential workload" was defined as the annual number of CT plus MRI examinations divided by the total number of diagnostic radiologists. The extent of radiologists' involvement was measured as the proportion of CT and MRI examinations to which "Added-fees for Radiological Managements on Imaging-studies (ARMIs)" were applied among eligible examinations. Maximum variation was computed as the ratio of the highest-to-lowest values among the countries or Japanese prefectures.

RESULTS

The radiologist potential workload in Japan was 2.78-4.17 times higher than those in other countries. A maximum prefecture-to-prefecture variation was 3.88. The average percentage of CT plus MRI examinations with ARMI applied was 43.3%, with a maximum prefecture-to-prefecture variation of 3.97. Prefectures with more radiologists tended to have a higher extent of radiologists' involvement.

CONCLUSIONS

Japan had a far greater radiologist potential workload compared with other countries, with a large regional variation among prefectures. Prefectures with more radiologists tended to have a higher extent of radiologists' involvement in CT and MRI examinations.

Defining Quality in Cardiovascular Imaging: A Scientific Statement From the American Heart Association

The aims of the current statement are to refine the definition of quality in cardiovascular imaging and to propose novel methodological approaches to inform the demonstration of quality in imaging in future clinical trials and registries. We propose defining quality in cardiovascular imaging using an analytical framework put forth by the Institute of Medicine whereby quality was defined as testing being safe, effective, patient-centered, timely, equitable, and efficient. The implications of each of these components of quality health care are as essential for cardiovascular imaging as they are for other areas within health care. Our proposed statement may serve as the foundation for integrating these quality indicators into establishing designations of quality laboratory practices and developing standards for value-based payment reform for imaging services. We also include recommendations for future clinical research to fulfill quality aims within cardiovascular imaging, including clinical hypotheses of improving patient outcomes, the importance of health status as an end point, and deferred testing options. Future research should evolve to define novel methods optimized for the role of cardiovascular imaging for detecting disease and guiding treatment and to demonstrate the role of cardiovascular imaging in facilitating healthcare quality.