Multi-Radiologist User Study for Artificial Intelligence-Guided Grading of COVID-19 Lung Disease Severity on Chest Radiographs

RATIONALE AND OBJECTIVES

Radiographic findings of COVID-19 pneumonia can be used for patient risk stratification; however, radiologist reporting of disease severity is inconsistent on chest radiographs (CXRs). We aimed to see if an artificial intelligence (AI) system could help improve radiologist interrater agreement.

MATERIALS AND METHODS

We performed a retrospective multi-radiologist user study to evaluate the impact of an AI system, the PXS score model, on the grading of categorical COVID-19 lung disease severity on 154 chest radiographs into four ordinal grades (normal/minimal, mild, moderate, and severe). Four radiologists (two thoracic and two emergency radiologists) independently interpreted 154 CXRs from 154 unique patients with COVID-19 hospitalized at a large academic center, before and after using the AI system (median washout time interval was 16 days). Three different thoracic radiologists assessed the same 154 CXRs using an updated version of the AI system trained on more imaging data. Radiologist interrater agreement was evaluated using Cohen and Fleiss kappa where appropriate. The lung disease severity categories were associated with clinical outcomes using a previously published outcomes dataset using Fisher's exact test and Chi-square test for trend.

RESULTS

Use of the AI system improved radiologist interrater agreement (Fleiss κ = 0.40 to 0.66, before and after use of the system). The Fleiss κ for three radiologists using the updated AI system was 0.74. Severity categories were significantly associated with subsequent intubation or death within 3 days.

CONCLUSION

An AI system used at the time of CXR study interpretation can improve the interrater agreement of radiologists.

Prognostic value of the chest X-ray in patients hospitalised for heart failure

Background

Patients admitted to hospital with heart failure will have had a chest X-ray (CXR), but little is known about their prognostic significance. We aimed to report the prevalence and prognostic value of the initial chest radiograph findings in patients admitted to hospital with heart failure (acute heart failure, AHF).

Methods

The erect CXRs of all patients admitted with AHF between October 2012 and November 2016 were reviewed for pulmonary venous congestion, Kerley B lines, pleural effusions and alveolar oedema. Film projection (whether anterior–posterior [AP] or posterior–anterior [PA]) and cardiothoracic ratio (CTR) were also recorded. Trial registration: ISRCTN96643197

Results

Of 1145 patients enrolled, 975 [median (interquartile range) age 77 (68–83) years, 61% with moderate, or worse, left ventricular systolic dysfunction, and median NT-proBNP 5047 (2337–10,945) ng/l] had an adequate initial radiograph, of which 691 (71%) were AP. The median CTR was 0.57 (IQR 0.53–0.61) in PA films and 0.60 (0.55–0.64) in AP films. Pulmonary venous congestion was present in 756 (78%) of films, Kerley B lines in 688 (71%), pleural effusions in 649 (67%) and alveolar oedema in 622 (64%).

A CXR score was constructed using the above features. Increasing score was associated with increasing age, urea, NT-proBNP, and decreasing systolic blood pressure, haemoglobin and albumin; and with all-cause mortality on multivariable analysis (hazard ratio 1.10, 95% confidence intervals 1.07–1.13, p < 0.001).

Conclusions

Radiographic evidence of congestion on a CXR is very common in patients with AHF and is associated with other clinical measures of worse prognosis.

Graphic abstract

Decoding the neonatal chest radiograph: An insight into neonatal respiratory distress

Respiratory distress is one of the leading causes of neonatal morbidity and mortality. Factors such as gestational age at birth, pulmonary maturity, and congenital factors are peculiar to this demographic. Clinical evaluation accompanied by chest radiography is the standard protocol for evaluating the underlying causative factors. Knowledge of the radiographic appearances of various pathologies and associations with certain congenital factors is quintessential for radiologists and primary neonatal care providers to steer the management in the right direction.

Prescription practice and clinical utility of chest radiographs in a pediatric intensive care unit: a prospective observational study

Background

Chest radiograph (CXR) prescribing pattern and practice vary widely among pediatric intensive care units (PICU). ‘On demand’ approach is increasingly recommended as against daily ‘routine’ CXRs; however, the real-world practice is largely unknown.

Methods

This was a prospective observational study performed in children younger than 12 years admitted to PICU of a tertiary care teaching hospital in India. Data were collected on all consecutive CXRs performed between December 2016 and April 2017. The primary outcome was to assess the factors that were associated with higher chest radiograph prescriptions in PICU. Secondary outcomes were to study the indications, association with mechanical ventilation, image quality and avoidable radiation exposure.

Results

Of 303 children admitted during the study period, 159 underwent a total of 524 CXRs in PICU. Median (IQR) age of the study cohort was 2 (0.6–5) years. More than two thirds [n = 115, 72.3%] were mechanically ventilated. Most CXRs (n = 449, 85.7%) were performed on mechanically ventilated patients, amounting to a median (IQR) of 3 (2–5) radiographs per ventilated patient. With increasing duration of ventilation, the number of CXRs proportionately increased in the first two weeks of mechanical ventilation. In non-ventilated children, about two thirds (68%) underwent only one CXR. Majority of the prescriptions were on demand (n = 461, 88%). Most common indications were peri-procedure prescriptions (37%) followed by evaluation for respiratory disease status (24%). About 40% CXRs resulted in interventions; adjustment in ventilator settings (13.5%) was the most frequent intervention. In 26% (n = 138) of radiographs, image quality required improvement. One or more additional body part exposure other than chest and upper abdomen were noted 336 (64%) images. Children with > 3 CXR had higher PRISM III score, more often mechanically ventilated, had higher number of indwelling devices [mean (SD) 2.6 (1.2) vs. 1.7 (1.0)] and stayed longer in PICU [median (IQR) 11(7.5–18.5) vs. 6 (3–9)].

Conclusion

On demand prescription was the prevalent practice in our PICU. Most non-ventilated children underwent only one CXR while duration of PICU stay and the number of devices determined the number of CXRs in mechanically ventilated children. Quality improvement strategies should concentrate on the process of acquisition of images and limiting the radiation exposure to unwanted body parts.

Deep metric learning-based image retrieval system for chest radiograph and its clinical applications in COVID-19

In recent years, deep learning-based image analysis methods have been widely applied in computer-aided detection, diagnosis and prognosis, and has shown its value during the public health crisis of the novel coronavirus disease 2019 (COVID-19) pandemic. Chest radiograph (CXR) has been playing a crucial role in COVID-19 patient triaging, diagnosing and monitoring, particularly in the United States. Considering the mixed and unspecific signals in CXR, an image retrieval model of CXR that provides both similar images and associated clinical information can be more clinically meaningful than a direct image diagnostic model. In this work we develop a novel CXR image retrieval model based on deep metric learning. Unlike traditional diagnostic models which aim at learning the direct mapping from images to labels, the proposed model aims at learning the optimized embedding space of images, where images with the same labels and similar contents are pulled together. The proposed model utilizes multi-similarity loss with hard-mining sampling strategy and attention mechanism to learn the optimized embedding space, and provides similar images, the visualizations of disease-related attention maps and useful clinical information to assist clinical decisions. The model is trained and validated on an international multi-site COVID-19 dataset collected from 3 different sources. Experimental results of COVID-19 image retrieval and diagnosis tasks show that the proposed model can serve as a robust solution for CXR analysis and patient management for COVID-19. The model is also tested on its transferability on a different clinical decision support task for COVID-19, where the pre-trained model is applied to extract image features from a new dataset without any further training. The extracted features are then combined with COVID-19 patient's vitals, lab tests and medical histories to predict the possibility of airway intubation in 72 hours, which is strongly associated with patient prognosis, and is crucial for patient care and hospital resource planning. These results demonstrate our deep metric learning based image retrieval model is highly efficient in the CXR retrieval, diagnosis and prognosis, and thus has great clinical value for the treatment and management of COVID-19 patients.

A review of chest radiographic patterns in mild to moderate novel corona virus disease 2019 at an urban hospital in Ghana

INTRODUCTION

The novel corona virus disease 2019 (COVID-19) was diagnosed in Wuhan, China in December 2019 and, in Ghana, in March 2020. As of 30th July 2020, Ghana had recorded 35,142 cases. COVID-19 which can be transmitted by both symptomatic and asymptomatic individuals usually manifest as pneumonia with symptoms like fever, cough, dyspnoea and fatigue. The current non-availability of a vaccine or drug for COVID-19 management calls for early detection and isolation of affected individuals. Chest imaging has become an integral part of patient management with chest radiography serving as a primary imaging modality in many centres.

METHODS

The study was a retrospective study conducted at Ga East Municipal Hospital (GEMH). Chest radiographs of patients with mild to moderate disease managed at GEMH were evaluated. The age, gender, symptom status, comorbidities and chest x-ray findings of the patients were documented.

RESULTS

11.4 % of the patients had some form of respiratory abnormality on chest radiography with 88.9% showing COVID-19 pneumonia features. 93.8% showed ground glass opacities (GGO), with 3.1% each showing consolidation (CN) only and CN with GGO. There was a significant association between COVID-19 radiographic features and patient's age, symptom status and comorbidities but not with gender.

CONCLUSION

Most radiographs were normal with only 11% showing COVID-19-like abnormality. There was a significant association between age, symptom status and comorbidities with the presence of COVID-19 like features but not for gender. There was no association between the extent of the lung changes and patient characteristics.

FUNDING

None declared.

Two cases of spontaneous pneumomediastinum with pneumothorax in patients with COVID-19 associated pneumonia

Coronavirus 2019 (COVID-19) is an infectious viral illness caused by severe acute respiratory syndrome virus coronavirus 2 (SARS-CoV-2). This disease mainly affects the lungs manifesting as acute lung injury, pneumonia, and acute respiratory distress syndrome. We describe two patients who developed concomitant spontaneous pneumothorax and pneumomediastinum in the setting of SARS-CoV-2 leading to acute hypoxic respiratory failure. This report adds to the increasing number of cases describing pulmonary complications of COVID-19 infection. Further studies are needed to ascertain the prognostic significance of these pulmonary complications in patients with SARS-CoV-2 infection.

•

A dangerous complication of COVID-19 is the development spontaneous pneumothorax and pneumomediastinum.

•

CT chest is warranted in patients with declining respiratory status in the setting of COVID-19 infection.

•

The risk of SPX and SPM may be increased with the addition of high flow nasal canula.

Differentiation of COVID-19 conditions in planar chest radiographs using optimized convolutional neural networks

In this study, an attempt has been made to differentiate Novel Coronavirus-2019 (COVID-19) conditions from healthy subjects in Chest radiographs using a simplified end-to-end Convolutional Neural Network (CNN) model and occlusion sensitivity maps. Early detection and faster automated screening of the COVID-19 patients is essential. For this, the images are considered from publicly available datasets. Significant biomarkers representing critical image features are extracted from CNN by experimentally investigating on cross-validation methods and hyperparameter settings. The performance of the network is evaluated using standard metrics. Perturbation based occlusion sensitivity maps are employed on the features obtained from the classification model to visualise the localization of abnormal areas. Results demonstrate that the simplified CNN model with optimised parameters is able to extract significant features with a sensitivity of 97.35% and F-measure of 96.71% to detect COVID-19 images. The algorithm achieves an Area Under the Curve-Receiver Operating Characteristic score of 99.4% with Matthews correlation coefficient of 0.93. High value of Diagnostic odds ratio is also obtained. Occlusion sensitivity maps provide precise localization of abnormal regions by identifying COVID-19 conditions. As early detection through chest radiographic images are useful for automated screening of the disease, this method appears to be clinically relevant in providing a visual diagnostic solution using a simplified and efficient model.

Is a chest radiograph after thoracostomy tube removal necessary? A cost-effective analysis

BACKGROUND

Following placement of tube thoracostomy (TT) for evacuation of traumatic hemopneumothorax (HPTX), controversy persists over the need for routine post-TT removal chest radiograph (CXR). Current research demonstrates routine CXR may offer no advantage over clinical observation alone while simultaneously increasing hospital resource utilization. As such, we hypothesized that in resolved traumatic HPTXs routine post-TT removal CXR to assess recurrent PTX compared to clinical observation is not cost-effective.

METHODS

We performed a decision-analytic model to evaluate the cost-effectiveness of routine CXR compared to clinical observation following TT removal. Our base case was a patient that sustained thoracic trauma with radiographic and clinical resolution of HPTX following TT evacuation. Cost, utility and probability estimates were generated from published literature, with costs represented in 2019 US dollars and utilities in Quality-Adjusted Life Years (QALYs). Deterministic and probabilistic sensitivity analyses were performed.

RESULTS

Decision-analytic model identified that clinical observation after TT removal was the dominant strategy with increased benefit at less cost, when compared to routine CXR, with a net cost of $194.92, QALYs of 0.44. In comparison, routine CXR demonstrated an increase of $821.42 in cost with 0.43 QALYs. On probabilistic sensitivity analysis the clinical observation strategy was found cost-effective in 99.5% of 10,000 iterations.

CONCLUSION

In trauma patients with clinical and radiographic evidence of a resolved HPTX, the adoption of clinical observation in lieu of post-TT removal CXR is cost-effective. Routine CXR following TT removal accrues more cost without additional benefit. The practice of routinely obtaining a CXR following TT removal should be scrutinized.

Pulmonary Edema: A Pictorial Review of Imaging Manifestations and Current Understanding of Mechanisms of Disease

Pulmonary edema is a common clinical entity caused by the extravascular movement of fluid into the pulmonary interstitium and alveoli. The four physiologic categories of edema include hydrostatic pressure edema, permeability edema with and without diffuse alveolar damage (DAD), and mixed edema where there is both an increase in hydrostatic pressure and membrane permeability. As radiographic manifestations and etiologies are varied, an appreciation for both the common and uncommon manifestations and causes of pulmonary edema is essential for accurate diagnosis.

Simple quantitative chest CT for pulmonary edema

•

Quantitative CT is a highly accurate but underutilized method for identifying pulmonary edema on CT.

•

There is a moderatelystrong correlation between CT HUs and CXR pulmonary edema grade in every lobe with correlation coefficients ranging from 0.585−0.685.

•

CT Hounsfield unit measurement yields excellent accuracy in differentiating no edema from mild to severe edema, with AUCs up to 0.995 in the LUL.

•

Qualitative CT with a % versus 84 % and specificity 95 % versus 78 Qualitative CT with a HU cut-off of -825 in the LUL is more sensitive (100% vs 84%) and specific (95% vs 78%) than qualitative CT.

Purpose

To determine the accuracy of quantitative CT to diagnose pulmonary edema compared to qualitative CT and CXR and to determine a threshold Hounsfield unit (HU) measurement for pulmonary edema on CT examinations.

Method

Electronic medical records were searched for patients with a billing diagnosis of heart failure and a Chest CT and CXR performed within three hours between 1/1/2016 to 10/1/2016, yielding 100 patients. CXR and CT examinations were scored for the presence and severity of edema, using a 0–5 scale, and CT HU measurements were obtained in each lobe. Polyserial correlation coefficients evaluated the association between CT HUs and CXR scores, and receiver operating characteristic (ROC) curve analysis determined a cutoff CT HU value for identification of pulmonary edema.

Results

Correlation between CT HU and CXR score was moderately strong (r = 0.585−0.685) with CT HU measurements demonstrating good to excellent accuracy in differentiating between no edema (grade 0) and mild to severe edema (grades 1–5) in every lobe, with AUCs ranging between 0.869 and 0.995. The left upper lobe demonstrated the highest accuracy, using a cutoff value of -825 HU (AUC of 0.995, sensitivity = 100 % and specificity = 95.1 %). Additionally, qualitative CT evaluation was less sensitive (84 %) than portable CXR in identifying pulmonary edema. However, quantitative CT evaluation was as sensitive as portable CXR (100 %) and highly specific (95 %).

Conclusions

Quantitative CT enables the identification of pulmonary edema with high accuracy and demonstrates a greater sensitivity than qualitative CT in assessment of pulmonary edema.

To repeat or not to repeat: Radiologists demonstrated more decisiveness than their fellow radiographers in reducing the repeat rate during mobile chest radiography

INTRODUCTION

Radiologists and radiographers play a complementary role in providing an optimal image quality with decrease radiation dose and proper diagnosis during chest radiographs. We aim Investigate years of experience among radiologists and radiographers on perception of image quality and its impact on repeat rate when evaluating portable pediatric chest radiographs.

METHODS

IRB approved retrospective study consisted of randomly selected images (n = 131) of pediatric portable chest radiographs. Images were blindly assessed by four radiologists and four radiographers. Readers were asked to assess qualitative and quantitative image quality by rating: image quality, decision to repeat and image technique. All data was compared employing Pearson's Correlation, Visual grading characteristic (VGC) and Cohens' kappa analyses.

RESULTS

Image quality: Radiologists (88.4%) rated images as excellent significantly more than radiographers (11.6%), and radiographers (90.1%) as poor significantly more than radiologists (9.9%) (p < 0.05). Repeat: Radiologists (57%) decided not to repeat images significantly more than radiographers (43%) (p < 0.05). Image technique: Radiologists rated images as acceptable (65%) and excellent (97.7%) significantly more than radiographers (35% and 2.3% respectively) (p < 0.05), whereas radiographers (84%) assessed image technique as poor significantly more than radiologists (16%) (p < 0.05). VGC: radiographers had slightly better qualitative evaluation of image quality than radiologists. An association between image quality (p < 0.002) and repeat decision (p < 0.044) with years of experience was established when comparing years of experience with image assessment rubric, while no association was noted with image technique (p < 0.9).

CONCLUSION

Radiologists demonstrated more decisiveness than their fellow radiographers in reducing the repeat rate of portable pediatric chest radiographs. Interestingly, years of experience only seem to affect image technique and image quality assessment among radiologists.

IMPLICATIONS FOR PRACTICE

Continuous education of radiographers and close collaboration with radiologists is crucial to achieve optimal image quality and low radiation doses.

ACR Appropriateness Criteria® Rib Fractures

Rib fractures are the most common thoracic injury after minor blunt trauma. Although rib fractures can produce significant morbidity, the diagnosis of injuries to underlying organs is arguably more important as these complications are likely to have the most significant clinical impact. Isolated rib fractures have a relatively low morbidity and mortality and treatment is generally conservative. As such, evaluation with standard chest radiographs is usually sufficient for the diagnosis of rib fractures, and further imaging is generally not appropriate as there is little data that undiagnosed isolated rib fractures after minor blunt trauma affect management or outcomes. Cardiopulmonary resuscitation frequently results in anterior rib fractures and chest radiographs are usually appropriate (and sufficient) as the initial imaging modality in these patients. In patients with suspected pathologic fractures, chest CT or Tc-99m bone scans are usually appropriate and complementary modalities to chest radiography based on the clinical scenario. The American College of Radiology Appropriateness Criteria are evidence-based guidelines for specific clinical conditions that are reviewed annually by a multidisciplinary expert panel. The guideline development and revision include an extensive analysis of current medical literature from peer reviewed journals and the application of well-established methodologies (RAND/UCLA Appropriateness Method and Grading of Recommendations Assessment, Development, and Evaluation or GRADE) to rate the appropriateness of imaging and treatment procedures for specific clinical scenarios. In those instances where evidence is lacking or equivocal, expert opinion may supplement the available evidence to recommend imaging or treatment.

Projected lung areas using dynamic X-ray (DXR)

•

The right projected lung area (PLA) was significantly larger than left one.

•

PLA had correlation with height, weight, BMI, vital capacity (VC), and forced expiratory volume in one second (FEV₁).

•

Multivariate analysis showed that body mass index (BMI), sex and VC were considered independent correlation factors, respectively.

Background

Dynamic X-ray (DXR) provides images of multiple phases of breath with less radiation exposure than CT. The exact images at end-inspiratory or end-expiratory phases can be chosen accurately.

Purpose

To investigate the correlation of the projected lung area (PLA) by dynamic chest X-ray with pulmonary functions.

Material and Methods

One hundred sixty-two healthy volunteers who received medical check-ups for health screening were included in this study. All subjects underwent DXR in both posteroanterior (PA) and lateral views and pulmonary function tests on the same day. All the volunteers took several tidal breaths before one forced breath as instructed. The outlines of lungs were contoured manually on the workstation with reference to the motion of diaphragm and the graph of pixel values. The PLAs were calculated automatically, and correlations with pulmonary functions and demographic data were analyzed statistically.

Results

The PLAs have correlation with physical characteristics, including height, weight and BMI, and pulmonary functions such as vital capacity (VC) and forced expiratory volume in one second (FEV₁). VC and FEV₁ revealed moderate correlation with the PLAs of PA view in forced inspiratory phase (VC: right, r = 0.65; left, r = 0.69. FEV1: right, r = 0.54; left, r = 0.59). Multivariate analysis showed that body mass index (BMI), sex and VC were considered independent correlation factors, respectively.

Conclusion

PLA showed statistically significant correlation with pulmonary functions. Our results indicate DXR has a possibility to serve as an alternate method for pulmonary function tests in subjects requiring contact inhibition including patients with suspected or confirmed covid-19.

Radiologists are increasingly recommending follow-up of chest radiographs: a 10-year review

PURPOSE

To assess for changes in trends of GP chest radiograph reporting over a 10-year period and to assess if there has been a change in recommendations for follow-up.

METHODS

Retrospective study of an Irish tertiary referral center. The total number of GP-referred chest x-rays performed per year from 2007 to 2017 are recorded. One-hundred male/100 female GP-referred chest x-rays are chosen at random from NIMIS data for each of 2007, 2010, 2013, and 2017. Reports are analyzed with regard to abnormal findings, recommendation for follow-up, and yield of follow-up imaging.

RESULTS

There were 4917 GP CXRs performed in 2007, 4856 in 2010, 5561 in 2013, and 6492 in 2017. Follow-up was recommended in 17 studies(8.5%) in 2007, 19 studies(9.5%) in 2010, 22 studies(11%) in 2013, and 27 studies(13.5%) in 2017. Indications for follow-up recommendation were largely to ensure resolution of infection (52%) or for nodule surveillance (43%). There has been a notable increase in lung nodule follow-up, with suggested follow-ups increasing from 6 in 2007, to 7 in 2010, 9 in 2013, and 14 in 2017, an increase of 58%.

CONCLUSION

Along with the increase in the quantity of GP-referred chest radiographs over the past 10 years, suggestions for follow-up have increased, particularly for nodule surveillance. Reasons for this increase may include lack of availability of CT to GPs for lung cancer screening, insensitivity of plain radiographs to early cancer detection, and possible fear of litigation for missing lesions, making radiologists more cautious.

The role of 22q11.2 deletion syndrome in the relationship between congenital heart disease and scoliosis

BACKGROUND CONTEXT

For over four decades, clinicians and researchers have suggested a relationship between congenital heart disease (CHD) and scoliosis, attributed to either the disease itself or to the long-term effects of cardiac surgery on the immature thoracic cage. However, no study has yet accounted for 22q11.2 deletion syndrome (22q11.2DS), the second most common cause of CHD after Down syndrome. 22q11.2DS has a scoliosis risk of 50%, but within 22q11.2DS a previous report found no significant association between scoliosis and CHD. We, therefore, hypothesized that scoliosis within a CHD cohort would be related to an underlying 22q11.2 deletion.

PURPOSE

To determine the prevalence of scoliosis in CHD patients with and without 22q11.2DS.

STUDY DESIGN/SETTING

Cross-sectional.

PATIENT SAMPLE

A well-characterized existing database of 315 adults with CHD (primarily tetralogy of Fallot), with (n=86) and without (n=229) 22q11.2DS, matched by sex and CHD severity, and excluding other known syndromic diagnoses. We compared the scoliosis prevalence of patients with 22q11.2DS and CHD patients to the prevalence of scoliosis in a cohort of adults with 22q11.2DS without CHD based on medical records.

OUTCOME MEASURES

Presence of scoliosis (Cobb angle ≥10°).

METHODS

We systematically determined the presence of scoliosis in all included patients using chest radiographs, blind to genetic diagnosis. Besides 22q11.2DS, we analyzed other suspected risk factors for scoliosis using a regression model: thoracotomy before the age of 12 years, severe CHD type and sex.

RESULTS

The prevalence of scoliosis in adults with CHD and 22q11.2DS (n=46, 53.5%) was significantly greater than in those without 22q11.2DS (n=18, 7.9%, p<.0001). The presence of a 22q11.2 deletion (odds ratio [OR] 25.4, 95% confidence interval [95% CI] 11.2-57.4, p<.0001), a history of thoracotomy before the age of 12 years (OR 3.5, 95% CI 1.6-8.1, p=.0027) and most complex CHD class (OR 2.3, 95% CI 1.1-4.7, p=.0196), but not sex, were significant independent predictors of scoliosis. In the 22q11.2DS group, a right-sided aortic arch was associated with a left thoracic scoliotic curve (p=.036).

CONCLUSIONS

The prevalence of scoliosis in those with CHD but without a 22q11.2 deletion approximates that of the general population. However, in the CHD population with a 22q11.2 deletion, the prevalence of scoliosis approximates that of others with 22q11.2DS. The pediatric surgical approach and severity of CHD were weaker independent contributors as compared to the 22q11.2 deletion. The results support the importance of a genetic diagnosis of 22q11.2DS to the risk of developing scoliosis in individuals with CHD. The 22q11.2 deletion may represent a common etiopathogenetic pathway for both CHD and scoliosis, possibly involving early laterality mechanisms.

Prognostic relevance and inter-observer reliability of chest-imaging in pediatric ARDS: a pediatric acute respiratory distress incidence and epidemiology (PARDIE) study

Purpose

Definitions of acute respiratory distress syndrome (ARDS) include radiographic criteria, but there are concerns about reliability and prognostic relevance. This study aimed to evaluate the independent relationship between chest imaging and mortality and examine the inter-rater variability of interpretations of chest radiographs (CXR) in pediatric ARDS (PARDS).

Methods

Prospective, international observational study in children meeting Pediatric Acute Lung Injury Consensus Conference (PALICC) criteria for PARDS, which requires new infiltrate(s) consistent with pulmonary parenchymal disease, without mandating bilateral infiltrates. Mortality analysis focused on the entire cohort, whereas inter-observer variability used a subset of patients with blinded, simultaneous interpretation of CXRs by intensivists and radiologists.

Results

Bilateral infiltrates and four quadrants of alveolar consolidation were associated with mortality on a univariable basis, using CXRs from 708 patients with PARDS. For patients on either invasive (IMV) or non-invasive ventilation (NIV) with PaO₂/FiO₂ (PF) ratios (or SpO₂/FiO₂ (SF) ratio equivalent) > 100, neither bilateral infiltrates (OR 1.3 (95% CI 0.68, 2.5), p = 0.43), nor 4 quadrants of alveolar consolidation (OR 1.6 (0.85, 3), p = 0.14) were associated with mortality. For patients with PF ≤ 100, bilateral infiltrates (OR 3.6 (1.4, 9.4), p = 0.01) and four quadrants of consolidation (OR 2.0 (1.14, 3.5), p = 0.02) were associated with higher mortality. A subset of 702 CXRs from 233 patients had simultaneous interpretations. Interobserver agreement for bilateral infiltrates and quadrants was “slight” (kappa 0.31 and 0.33). Subgroup analysis showed agreement did not differ when stratified by PARDS severity but was slightly higher for children with chronic respiratory support (kappa 0.62), NIV at PARDS diagnosis (kappa 0.53), age > 10 years (kappa 0.43) and fluid balance > 40 ml/kg (kappa 0.48).

Conclusion

Bilateral infiltrates and quadrants of alveolar consolidation are associated with mortality only for those with PF ratio ≤ 100, although there is high- inter-rater variability in these chest-x ray parameters.

Electronic supplementary material

The online version of this article (10.1007/s00134-020-06074-7) contains supplementary material, which is available to authorized users.

Clinical Implementation of Deep Learning in Thoracic Radiology: Potential Applications and Challenges

Chest X-ray radiography and computed tomography, the two mainstay modalities in thoracic radiology, are under active investigation with deep learning technology, which has shown promising performance in various tasks, including detection, classification, segmentation, and image synthesis, outperforming conventional methods and suggesting its potential for clinical implementation. However, the implementation of deep learning in daily clinical practice is in its infancy and facing several challenges, such as its limited ability to explain the output results, uncertain benefits regarding patient outcomes, and incomplete integration in daily workflow. In this review article, we will introduce the potential clinical applications of deep learning technology in thoracic radiology and discuss several challenges for its implementation in daily clinical practice.

Crowdsourcing pneumothorax annotations using machine learning annotations on the NIH chest X-ray dataset

Pneumothorax is a potentially life-threatening condition that requires prompt recognition and often urgent intervention. In the ICU setting, large numbers of chest radiographs are performed and must be interpreted on a daily basis which may delay diagnosis of this entity. Development of artificial intelligence (AI) techniques to detect pneumothorax could help expedite detection as well as localize and potentially quantify pneumothorax. Open image analysis competitions are useful in advancing state-of-the art AI algorithms but generally require large expert annotated datasets. We have annotated and adjudicated a large dataset of chest radiographs to be made public with the goal of sparking innovation in this space. Because of the cumbersome and time-consuming nature of image labeling, we explored the value of using AI models to generate annotations for review. Utilization of this machine learning annotation (MLA) technique appeared to expedite our annotation process with relatively high sensitivity at the expense of specificity. Further research is required to confirm and better characterize the value of MLAs. Our adjudicated dataset is now available for public consumption in the form of a challenge.

Is routine pre-entry chest radiograph necessary in a high tuberculosis prevalence country?

CONTEXT

Chest radiographs have been used worldwide as a screening tool before employment and training, by various healthcare and other government and nongovernment institutions. Many studies done in the past have demonstrated a relatively low yield for tuberculosis detection and therefore, the authors have questioned this practice.

AIMS

To compare the value of the preadmission/employment chest radiograph in two groups, namely, those who have been previously exposed to a healthcare setting (post-exposure group) and those who have not been exposed (pre-exposure group) and to determine if there is a significant difference in tuberculosis detection between these two groups.

SETTINGS AND DESIGN

A retrospective review of the reports of the chest radiographs of all candidates appearing for admission to various undergraduate and postgraduate courses in our institute between 2014 and 2017 was performed.

MATERIALS AND METHODS

The various abnormalities detected were recorded and the findings in the two groups were compared.

STATISTICAL ANALYSIS USED

Chi-square test was used to compare between two group proportions.

RESULTS

Thirty out of 4333 (0.69%) candidates in the pre-exposure group and 53 out of 3379 (1.57%) candidates in the post-exposure group showed abnormalities on chest radiographs involving the lung parenchyma, mediastinum, heart, or pleura. In the pre-exposure group, six (0.14%) were found to have underlying cardiac disease and one (0.02%) had tuberculosis. Among the six candidates in the post-exposure group who underwent further investigations in our institute, five (0.15%) were diagnosed to have tuberculosis. Although there was no statistically significant difference in tuberculosis detection between the groups (P = 0.051), there is a trend towards higher detection of tuberculosis in the post-exposure group.

CONCLUSIONS

In a country where the prevalence of tuberculosis is high, the pre-employment chest radiograph may still have a role in detecting tuberculosis in the post-exposure group.

Detectability of pulmonary nodules on chest radiographs: bone suppression versus standard technique with single versus dual monitors for visualization

PURPOSE

To evaluate the diagnostic accuracy of bone suppression imaging (BSI) in the detection of pulmonary nodules on chest radiographs (CXRs) and the effect of visualization method (single or dual monitors) on diagnostic accuracy.

MATERIALS AND METHODS

Ten observers interpreted the CXRs of 100 patients: 50 with a T1 lung cancer nodule and 50 without nodules. Each standard CXR was first read alone and then in combination with the corresponding BSI. Two sessions of viewing were conducted: (1) the standard CXR and BSI were placed side by side on dual monitors and (2) both images were shown on the same monitor in alternation. The nodule location, confidence level, and interpretation time were recorded and analyzed statistically.

RESULTS

When BSI was added, the area under the receiver operating characteristic curve (AUC) improved with dual monitors and a single monitor. The AUC was not significantly different between the dual-monitor and single-monitor sessions; however, the specificity with BSI and dual monitors decreased. The total interpretation time was significantly shorter with a single monitor than with dual monitors.

CONCLUSIONS

The use of BSI improved detectability of T1 lung cancer nodules on CXRs; however, specificity and reading time were affected by the visualization method.

Automatic delineation of ribs and clavicles in chest radiographs using fully convolutional DenseNets

BACKGROUND AND OBJECTIVE

In chest radiographs (CXRs), all bones and soft tissues are overlapping with each other, which raises issues for radiologists to read and interpret CXRs. Delineating the ribs and clavicles is helpful for suppressing them from chest radiographs so that their effects can be reduced for chest radiography analysis. However, delineating ribs and clavicles automatically is difficult by methods without deep learning models. Moreover, few of methods without deep learning models can delineate the anterior ribs effectively due to their faint rib edges in the posterior-anterior (PA) CXRs.

METHODS

In this work, we present an effective deep learning method for delineating posterior ribs, anterior ribs and clavicles automatically using a fully convolutional DenseNet (FC-DenseNet) as pixel classifier. We consider a pixel-weighted loss function to mitigate the uncertainty issue during manually delineating for robust prediction.

RESULTS

We conduct a comparative analysis with two other fully convolutional networks for edge detection and the state-of-the-art method without deep learning models. The proposed method significantly outperforms these methods in terms of quantitative evaluation metrics and visual perception. The average recall, precision and F-measure are 0.773 ± 0.030, 0.861 ± 0.043 and 0.814 ± 0.023 respectively, and the mean boundary distance (MBD) is 0.855 ± 0.642 pixels of the proposed method on the test dataset. The proposed method also performs well on JSRT and NIH Chest X-ray datasets, indicating its generalizability across multiple databases. Besides, a preliminary result of suppressing the bone components of CXRs has been produced by using our delineating system.

CONCLUSIONS

The proposed method can automatically delineate ribs and clavicles in CXRs and produce accurate edge maps.

Prognostic value of chest radiographs in patients with acute heart failure: the Radiology in Acute Heart Failure (RAD-ICA) study

OBJECTIVES

To determine whether chest radiographs can contribute to prognosis in patients with acute heart failure (AHF).

MATERIAL AND METHODS

Consecutive patients with AHF were enrolled by the participating emergency departments. Radiographic variables assessed were the presence or absence of evidence of cardiomegaly and pleural effusion and the pulmonary parenchymal pattern observed (vascular redistribution, interstitial edema, and/or alveolar edema). We gathered variables for the AHF episode and the patient's baseline state. Outcomes were in-hospital and 1-year mortality; hospital stay longer than 7 days, and a composite of events within 30 days of discharge (revisit, rehospitalization, and/or death). Crude and adjusted hazard ratios were calculated for the 3 categories of radiographic variables. The variables were also studied in combination.

RESULTS

A total of 2703 patients with a mean (SD) age of 81 (19) years were enrolled; 54.5% were women. Cardiomegaly was observed in 1711 cases (76.8%) and pleural effusion in 992 (36.7%). A pulmonary parenchymal pattern was observed in all cases, as follows: vascular redistribution in 1672 (61.9%), interstitial edema in 629 (23.3%) and alveolar edema in 402 (14.9%). The adjusted hazard ratios showed that cardiomegaly lacked prognostic value. However, the presence of pleural effusion was associated with a 23% (95% CI, 2%-49%) higher rate of the 30- day composite outcome; in-hospital mortality was 89% (30%-177%) higher in the presence of alveolar edema, and 1-year mortality was 38% (14%-67%) higher in association with vascular redistribution. The results for the variables in combination were consistent with the results for individual variables.

CONCLUSION

A diagnostic chest radiograph can also contribute to the prediction of adverse events. Pleural effusion is associated with a higher rate of events after discharge, and alveolar edema is associated with higher mortality.

Cardiomegaly on chest radiographs as a predictor of heart disease in the pediatric population

BACKGROUND

Cardiomegaly on chest radiographs (CXR) in pediatric patients leads to multiple tests. We aimed to determine the positive predictive value (PPV) of cardiomegaly on CXR in predicting subsequent heart disease and to assess the utility of obtaining a B-type Natriuretic Peptide level (BNP) and/or electrocardiogram (EKG) in such patients. We hypothesized that an echocardiogram may not be appropriate in all cases of cardiomegaly on CXR, particularly in a patient with a normal EKG and BNP level.

METHODS

We performed a retrospective cohort study of pediatric patients with cardiomegaly on their initial CXR between January 2015-December 2017. Patients without a subsequent echocardiogram or known congenital heart disease were excluded. A patient was deemed to have heart disease if they had structural abnormalities, functional abnormalities or a pericardial effusion on echocardiogram. The PPV of CXR and the PPV/NPV of the other tests (EKG, BNP) were calculated using contingency tables.

RESULTS

Four hundred and eighty nine patients met inclusion criteria. The PPV of cardiomegaly on CXR alone without any other diagnostic testing in predicting subsequent heart disease was 15%. The PPV increased if there was either an abnormal EKG or a BNP >100 pg/ml and further increased if both of these were present. The PPV values were higher in patients <1 year of age.

CONCLUSIONS

Cardiomegaly on CXR can often predict the presence of heart disease, particularly in infants. Further testing with EKG and BNP can better predict who may have heart disease, but it may not eliminate the need for echocardiography.

Hilar asymmetry in endobronchial tuberculosis patients: An often-overlooked clue

BACKGROUND

Endobronchial tuberculosis (EBTB) is a challenging diagnosis because of its varied clinical and radiological manifestations. Hilar asymmetry on chest radiograph (CXR) may be found in patient with EBTB but is often overlooked, which may lead to delayed diagnosis.

CASE REPORT

We present five cases with EBTB. Clinicians failed to identify unilateral hilar abnormalities on CXR, and these patients were treated initially for pharyngitis, bronchitis, or pneumonia with no improvement. Subsequently, they visited the pulmonary clinic and bronchoscopy revealed endobronchial lesions and microbial/histopathological evidence of tuberculous infection consistent with EBTB. Anti-tuberculosis therapy resulted in complete clinical resolution in four of the five patients; one patient had persistent bronchial stenosis.

CONCLUSION

Hilar asymmetry on CXR may occur with EBTB and may suggest this diagnosis in the appropriate clinical setting. Bronchoscopy has an important role in establishing the final diagnosis.