Deep Neural Network Augments Performance of Junior Residents in Diagnosing COVID-19 Pneumonia on Chest Radiographs.

Chest X-rays (CXRs) are essential in the preliminary radiographic assessment of patients affected by COVID-19. Junior residents, as the first point-of-contact in the diagnostic process, are expected to interpret these CXRs accurately. We aimed to assess the effectiveness of a deep neural network in distinguishing COVID-19 from other types of pneumonia, and to determine its potential contribution to improving the diagnostic precision of less experienced residents. A total of 5051 CXRs were utilized to develop and assess an artificial intelligence (AI) model capable of performing three-class classification, namely non-pneumonia, non-COVID-19 pneumonia, and COVID-19 pneumonia. Additionally, an external dataset comprising 500 distinct CXRs was examined by three junior residents with differing levels of training. The CXRs were evaluated both with and without AI assistance. The AI model demonstrated impressive performance, with an Area under the ROC Curve (AUC) of 0.9518 on the internal test set and 0.8594 on the external test set, which improves the AUC score of the current state-of-the-art algorithms by 1.25% and 4.26%, respectively. When assisted by the AI model, the performance of the junior residents improved in a manner that was inversely proportional to their level of training. Among the three junior residents, two showed significant improvement with the assistance of AI. This research highlights the novel development of an AI model for three-class CXR classification and its potential to augment junior residents' diagnostic accuracy, with validation on external data to demonstrate real-world applicability. In practical use, the AI model effectively supported junior residents in interpreting CXRs, boosting their confidence in diagnosis. While the AI model improved junior residents' performance, a decline in performance was observed on the external test compared to the internal test set. This suggests a domain shift between the patient dataset and the external dataset, highlighting the need for future research on test-time training domain adaptation to address this issue.

Deep Supervised Domain Adaptation for Pneumonia Diagnosis From Chest X-Ray Images.

Pneumonia is one of the most common treatable causes of death, and early diagnosis allows for early intervention. Automated diagnosis of pneumonia can therefore improve outcomes. However, it is challenging to develop high-performance deep learning models due to the lack of well-annotated data for training. This paper proposes a novel method, called Deep Supervised Domain Adaptation (DSDA), to automatically diagnose pneumonia from chest X-ray images. Specifically, we propose to transfer the knowledge from a publicly available large-scale source dataset (ChestX-ray14) to a well-annotated but small-scale target dataset (the TTSH dataset). DSDA aligns the distributions of the source domain and the target domain according to the underlying semantics of the training samples. It includes two task-specific sub-networks for the source domain and the target domain, respectively. These two sub-networks share the feature extraction layers and are trained in an end-to-end manner. Unlike most existing domain adaptation approaches that perform the same tasks in the source domain and the target domain, we attempt to transfer the knowledge from a multi-label classification task in the source domain to a binary classification task in the target domain. To evaluate the effectiveness of our method, we compare it with several existing peer methods. The experimental results show that our method can achieve promising performance for automated pneumonia diagnosis.

Diagnostic Performance of a Deep Learning Model Deployed at a National COVID-19 Screening Facility for Detection of Pneumonia on Frontal Chest Radiographs.

(1) Background: Chest radiographs are the mainstay of initial radiological investigation in this COVID-19 pandemic. A reliable and readily deployable artificial intelligence (AI) algorithm that detects pneumonia in COVID-19 suspects can be useful for screening or triage in a hospital setting. This study has a few objectives: first, to develop a model that accurately detects pneumonia in COVID-19 suspects; second, to assess its performance in a real-world clinical setting; and third, by integrating the model with the daily clinical workflow, to measure its impact on report turn-around time. (2) Methods: The model was developed from the NIH Chest-14 open-source dataset and fine-tuned using an internal dataset comprising more than 4000 CXRs acquired in our institution. Input from two senior radiologists provided the reference standard. The model was integrated into daily clinical workflow, prioritising abnormal CXRs for expedited reporting. Area under the receiver operating characteristic curve (AUC), F1 score, sensitivity, and specificity were calculated to characterise diagnostic performance. The average time taken by radiologists in reporting the CXRs was compared against the mean baseline time taken prior to implementation of the AI model. (3) Results: 9431 unique CXRs were included in the datasets, of which 1232 were ground truth-labelled positive for pneumonia. On the "live" dataset, the model achieved an AUC of 0.95 (95% confidence interval (CI): 0.92, 0.96) corresponding to a specificity of 97% (95% CI: 0.97, 0.98) and sensitivity of 79% (95% CI: 0.72, 0.84). No statistically significant degradation of diagnostic performance was encountered during clinical deployment, and report turn-around time was reduced by 22%. (4) Conclusion: In real-world clinical deployment, our model expedites reporting of pneumonia in COVID-19 suspects while preserving diagnostic performance without significant model drift.

Clinical features and predictors of severity in COVID-19 patients with critical illness in Singapore.

We aim to describe a case series of critically and non-critically ill COVID-19 patients in Singapore. This was a multicentered prospective study with clinical and laboratory details. Details for fifty uncomplicated COVID-19 patients and ten who required mechanical ventilation were collected. We compared clinical features between the groups, assessed predictors of intubation, and described ventilatory management in ICU patients. Ventilated patients were significantly older, reported more dyspnea, had elevated C-reactive protein and lactate dehydrogenase. A multivariable logistic regression model identified respiratory rate (aOR 2.83, 95% CI 1.24-6.47) and neutrophil count (aOR 2.39, 95% CI 1.34-4.26) on admission as independent predictors of intubation with area under receiver operating characteristic curve of 0.928 (95% CI 0.828-0.979). Median APACHE II score was 19 (IQR 17-22) and PaO2/FiO2 ratio before intubation was 104 (IQR 89-129). Median peak FiO2 was 0.75 (IQR 0.6-1.0), positive end-expiratory pressure 12 (IQR 10-14) and plateau pressure 22 (IQR 18-26) in the first 24 h of ventilation. Median duration of ventilation was 6.5 days (IQR 5.5-13). There were no fatalities. Most COVID-19 patients in Singapore who required mechanical ventilation because of ARDS were extubated with no mortality.

High-risk chest radiographic features associated with COVID-19 disease severity.

OBJECTIVES: High-risk CXR features in COVID-19 are not clearly defined. We aimed to identify CXR features that correlate with severe COVID-19. METHODS: All confirmed COVID-19 patients admitted within the study period were screened. Those with suboptimal baseline CXR were excluded. CXRs were reviewed by three independent radiologists and opacities recorded according to zones and laterality. The primary endpoint was defined as hypoxia requiring supplemental oxygen, and CXR features were assessed for association with this endpoint to identify high-risk features. These features were then used to define criteria for a high-risk CXR, and clinical features and outcomes of patients with and without baseline high-risk CXR were compared using logistic regression analysis. RESULTS: 109 patients were included. In the initial analysis of 40 patients (36.7%) with abnormal baseline CXR, presence of bilateral opacities, multifocal opacities, or any upper or middle zone opacity were associated with supplemental oxygen requirement. Of the entire cohort, 29 patients (26.6%) had a baseline CXR with at least one of these features. Having a high-risk baseline CXR was significantly associated with requiring supplemental oxygen in univariate (odds ratio 14.0, 95% confidence interval 3.90-55.60) and multivariate (adjusted odds ratio 8.38, 95% CI 2.43-28.97, P = 0.001) analyses. CONCLUSION: We identified several high-risk CXR features that are significantly associated with severe illness. The association of upper or middle zone opacities with severe illness has not been previously emphasized. Recognition of these specific high-risk CXR features is important to prioritize limited healthcare resources for sicker patients.

Radiographic features of COVID-19 based on an initial cohort of 96 patients in Singapore.

INTRODUCTION: Chest radiographs (CXRs) are widely used for the screening and management of COVID-19. This article describes the radiographic features of COVID-19 based on an initial national cohort of patients. METHODS: This is a retrospective review of swab-positive patients with COVID-19 who were admitted to four different hospitals in Singapore between 22 January and 9 March 2020. Initial and follow-up CXRs were reviewed by three experienced radiologists to identify the predominant pattern and distribution of lung parenchymal abnormalities. RESULTS: In total, 347 CXRs of 96 patients were reviewed. Initial CXRs were abnormal in 41 (42.7%) out of 96 patients. The mean time from onset of symptoms to CXR abnormality was 5.3 ± 4.7 days. The predominant pattern of lung abnormality was ground-glass opacity on initial CXRs (51.2%) and consolidation on follow-up CXRs (51.0%). Multifocal bilateral abnormalities in mixed central and peripheral distribution were observed in 63.4% and 59.2% of abnormal initial and follow-up CXRs, respectively. The lower zones were involved in 90.2% of initial CXRs and 93.9% of follow-up CXRs. CONCLUSION: In a cohort of swab-positive patients, including those identified from contact tracing, we found a lower incidence of CXR abnormalities than was previously reported. The most common pattern was ground-glass opacity or consolidation, but mixed central and peripheral involvement was more common than peripheral involvement alone.

Age and chest radiography as possible parameters for rapid triage in COVID-19 outbreak surge (preprint)

BackgroundTo evaluate the utility of age and chest radiography(CXR) in triaging COVID-19 patients for hospitalization versus isolation in non-hospital facilities, we examined how age and CXR at diagnosis were associated with clinical needs from late-January to early-April. MethodsClinical status of all COVID-19 cases was monitored for national disease surveillance. Cases were isolated in hospitals until SARS-CoV-2 RNA was undetectable on PCR. Age and CXR results on admission were analysed for association with oxygen supplementation and mechanical ventilation, the outcomes of interest.ResultsTill 4 April 2020, there were 1,481 COVID-19 cases in Singapore. Overall, 11.4% required supplemental oxygen while 4.8% required mechanical ventilation and intensive care. The respective proportions increased to 40.9% and 16.5% for cases aged ≥70 years. As a predictor of subsequent mechanical ventilation, age had an area under the receiver operator characteristic curve(AUROC) of 0.772 (95%CI:0.699-0.845). A combined criterion of either an abnormal CXR or age≥55 years had a sensitivity of 86.7% and specificity of 58.0% for the same outcome. A similar performance was observed for predicting oxygen supplementation needs.ConclusionsAge and CXR at diagnosis may be valuable in excluding severe disease, allowing safe triage for isolation in non-hospital facilities. 

Atypical Chest Computed Tomography Finding of Predominant Interstitial Thickening in a Patient with Coronavirus Disease 2019 (COVID-19) Pneumonia.

BACKGROUND Coronavirus disease 2019 (COVID-19) is caused by a novel coronavirus, SARS-CoV-2, and is associated with severe respiratory disease. There are extensive publications on the chest computed tomography (CT) findings of COVID-19 pneumonia, with ground-glass opacities (GGO) and mixed GGO and consolidation being the most common findings. Those with interstitial thickening manifesting as reticular opacities typically show superimposed ground-glass opacities, giving a crazy-paving pattern. CASE REPORT We report the case of a 77-year-old man with a background of asthma-chronic obstructive pulmonary disease (COPD) overlap syndrome (ACOS) who presented with progressive cough and shortness of breath for 2 days. He was in close contact with a confirmed COVID-19 case. Reverse-transcription polymerase chain reaction analysis of a nasopharyngeal swab was positive for SARS-CoV-2. The initial chest radiograph was negative for lung consolidation and ground-glass opacities. During admission, he had worsening shortness of breath with desaturation, prompting a chest CT examination, which was performed on day 14 of illness. The chest CT revealed an atypical finding of predominant focal subpleural interstitial thickening in the right lower lobe. He was provided supportive treatment along with steroid and antibiotics. He recovered well and subsequently tested negative for 2 consecutive swabs. He was discharged after 34 days. CONCLUSIONS Interstitial thickening or reticular pattern on CT has been described in COVID-19 pneumonia, but largely in association with ground-glass opacity or consolidation. This case demonstrates an atypical predominance of interstitial thickening on chest CT in COVID-19 pneumonia on day 14 of illness, which is the expected time of greatest severity of the disease.

Chest Radiography in Coronavirus Disease 2019 (COVID-19): Correlation with Clinical Course

Coronavirus disease 2019 (COVID-19) is caused by the severe acute respiratory syndrome coronavirus 2 and was declared a global pandemic by the World Health Organization on 11 March 2020. A definitive diagnosis of COVID-19 is made after a positive result is obtained on reverse transcription-polymerase chain reaction assay. In Singapore, rigorous contact tracing was practised to contain the spread of the virus. Nasal swabs and chest radiographs (CXR) were also taken from individuals who were suspected to be infected by COVID-19 upon their arrival at a centralised screening centre. From our experience, about 40% of patients who tested positive for COVID-19 had initial CXR that appeared "normal". In this case series, we described the temporal evolution of COVID-19 in patients with an initial "normal" CXR. Since CXR has limited sensitivity and specificity in COVID-19, it is not suitable as a first-line diagnostic tool. However, when CXR changes become unequivocally abnormal, close monitoring is recommended to manage potentially severe COVID-19 pneumonia.

Reversible platypnea-orthodeoxia in COVID-19 acute respiratory distress syndrome survivors.

Platypnea-orthodeoxia syndrome (POS) is a rare clinical syndrome characterized by orthostatic oxygen desaturation and positional dyspnea from supine to an upright position. We observed POS in 5 of 20 cases of severe 2019 novel coronavirus (COVID-19) pneumonia, which demonstrated persistently elevated shunt fraction even after liberation from mechanical ventilation. POS was first observed during physiotherapy sessions; median oxygen desaturation was 8 % (range: 8-12 %). Affected individuals were older (median 64 vs 53 years old, p = 0.05) and had lower body mass index (median 24.7 vs 27.6 kg/m2, p = 0.03) compared to those without POS. While POS caused alarm and reduced tolerance to therapy, this phenomenon resolved over a median of 17 days with improvement of parenchymal disease. The mechanisms of POS are likely due to gravitational redistribution of pulmonary blood flow resulting in increased basal physiological shunting and upper zone dead space ventilation due to the predominantly basal distribution of consolidative change and reported vasculoplegia and microthrombi in severe COVID-19 disease.

Clinical utility of chest radiography for severe COVID-19.

BACKGROUND: Chest radiography (CXR) is performed more widely and readily than CT for the management of coronavirus disease (COVID-19), but there remains little data on its clinical utility. This study aims to assess the diagnostic performance of CXR, with emphasis on its predictive value, for severe COVID-19 disease. METHODS: A retrospective cohort study was conducted, 358 chest radiographs were performed on 109 COVID-19 patients (median age 44.4 years, 58 males and 30 with comorbidities) admitted between 22 January 2020 and 15 March 2020. Each CXR was reviewed and scored by three radiologists in consensus using a 72-point COVID-19 Radiographic Score (CRS). Disease severity was determined by the need for supplemental oxygen and mechanical ventilation. RESULTS: Patients who needed supplemental oxygen (n=19, 17.4%) were significantly older (P<0.001) and significantly more of them had co-morbidities (P=0.011). They also had higher C-reactive protein (CRP) (P<0.001), higher lactate dehydrogenase (LDH) (P<0.001), lower lymphocyte count (P<0.001) and lower hemoglobin (Hb) (P=0.001). Their initial (CRSinitial) and maximal CRS (CRSmax) were higher (P<0.001). Adjusting for age and baseline hemoglobin, the AUROC of CRSmax (0.983) was as high as CRPmax (0.987) and higher than the AUROC for lymphocyte countmin (0.897), and LDHmax (0.900). The AUROC for CRSinitial was slightly lower (0.930). CRSinitial ≥5 had a sensitivity of 63% and specificity of 92% in predicting the need for oxygen, and 73% sensitivity and 88% specificity in predicting the need for mechanical ventilation. CRS between the 6th and 10th day from the onset of symptoms (CRSD6-10) ≥5 had a sensitivity of 89% and specificity of 95% in predicting the need for oxygen, and 100% sensitivity and 86% specificity in predicting the need for mechanical ventilation. CONCLUSIONS: Adjusting for key confounders of age and baseline Hb, CRSmax performed comparable to or better than laboratory markers in the diagnosis of severe disease. CXR performed between the 6th and 10th days from symptom onset was a better predictor of severe disease than CXR performed earlier at presentation. A benign clinical course was seen in CXR that were normal or had very mild abnormalities.

Care of the pregnant woman with coronavirus disease 2019 in labor and delivery: anesthesia, emergency cesarean delivery, differential diagnosis in the acutely ill parturient, care of the newborn, and protection of the healthcare personnel.

Coronavirus disease 2019, caused by the severe acute respiratory syndrome coronavirus 2, has been declared a pandemic by the World Health Organization. As the pandemic evolves rapidly, there are data emerging to suggest that pregnant women diagnosed as having coronavirus disease 2019 can have severe morbidities (up to 9%). This is in contrast to earlier data that showed good maternal and neonatal outcomes. Clinical manifestations of coronavirus disease 2019 include features of acute respiratory illnesses. Typical radiologic findings consists of patchy infiltrates on chest radiograph and ground glass opacities on computed tomography scan of the chest. Patients who are pregnant may present with atypical features such as the absence of fever as well as leukocytosis. Confirmation of coronavirus disease 2019 is by reverse transcriptase-polymerized chain reaction from upper airway swabs. When the reverse transcriptase-polymerized chain reaction test result is negative in suspect cases, chest imaging should be considered. A pregnant woman with coronavirus disease 2019 is at the greatest risk when she is in labor, especially if she is acutely ill. We present an algorithm of care for the acutely ill parturient and guidelines for the protection of the healthcare team who is caring for the patient. Key decisions are made based on the presence of maternal and/or fetal compromise, adequacy of maternal oxygenation (SpO2 >93%) and stability of maternal blood pressure. Although vertical transmission is unlikely, there must be measures in place to prevent neonatal infections. Routine birth processes such as delayed cord clamping and skin-to-skin bonding between mother and newborn need to be revised. Considerations can be made to allow the use of screened donated breast milk from mothers who are free of coronavirus disease 2019. We present management strategies derived from best available evidence to provide guidance in caring for the high-risk and acutely ill parturient. These include protection of the healthcare workers caring for the coronavirus disease 2019 gravida, establishing a diagnosis in symptomatic cases, deciding between reverse transcriptase-polymerized chain reaction and chest imaging, and management of the unwell parturient.

Variable computed tomography appearances of COVID-19.

The coronavirus disease 2019 (COVID-19) is typically diagnosed by specific assays that detect viral nucleic acid from the upper respiratory tract; however, this may miss infections involving only the lower airways. Computed tomography (CT) has been described as a diagnostic modality in the COVID-19 diagnosis and treatment plan. We present a case series with virologically confirmed COVID-19 pneumonia. Variable CT features were observed: consolidation with ground-glass opacities, ground-glass opacities with subpleural reticular bands, and an anterior-posterior gradient of lung abnormalities resembling that of acute respiratory distress syndrome. Evolution of CT findings was observed in one patient, where there was interval resolution of bilateral lung consolidation with development of bronchiolectasis and subpleural fibrotic bands. While sensitive for detecting lung parenchymal abnormalities in COVID-19 pneumonia, the use of CT for initial diagnosis is discouraged and should be reserved for specific clinical indications. Interpretation of chest CT findings should be correlated with duration of symptoms to better determine the disease stage and aid in patient management.