Dual center validation of deep learning for automated multi-label segmentation of thoracic anatomy in bedside chest radiographs.

BACKGROUND AND OBJECTIVES: Bedside chest radiographs (CXRs) are challenging to interpret but important for monitoring cardiothoracic disease and invasive therapy devices in critical care and emergency medicine. Taking surrounding anatomy into account is likely to improve the diagnostic accuracy of artificial intelligence and bring its performance closer to that of a radiologist. Therefore, we aimed to develop a deep convolutional neural network for efficient automatic anatomy segmentation of bedside CXRs. METHODS: To improve the efficiency of the segmentation process, we introduced a "human-in-the-loop" segmentation workflow with an active learning approach, looking at five major anatomical structures in the chest (heart, lungs, mediastinum, trachea, and clavicles). This allowed us to decrease the time needed for segmentation by 32% and select the most complex cases to utilize human expert annotators efficiently. After annotation of 2,000 CXRs from different Level 1 medical centers at Charité - University Hospital Berlin, there was no relevant improvement in model performance, and the annotation process was stopped. A 5-layer U-ResNet was trained for 150 epochs using a combined soft Dice similarity coefficient (DSC) and cross-entropy as a loss function. DSC, Jaccard index (JI), Hausdorff distance (HD) in mm, and average symmetric surface distance (ASSD) in mm were used to assess model performance. External validation was performed using an independent external test dataset from Aachen University Hospital (n = 20). RESULTS: The final training, validation, and testing dataset consisted of 1900/50/50 segmentation masks for each anatomical structure. Our model achieved a mean DSC/JI/HD/ASSD of 0.93/0.88/32.1/5.8 for the lung, 0.92/0.86/21.65/4.85 for the mediastinum, 0.91/0.84/11.83/1.35 for the clavicles, 0.9/0.85/9.6/2.19 for the trachea, and 0.88/0.8/31.74/8.73 for the heart. Validation using the external dataset showed an overall robust performance of our algorithm. CONCLUSIONS: Using an efficient computer-aided segmentation method with active learning, our anatomy-based model achieves comparable performance to state-of-the-art approaches. Instead of only segmenting the non-overlapping portions of the organs, as previous studies did, a closer approximation to actual anatomy is achieved by segmenting along the natural anatomical borders. This novel anatomy approach could be useful for developing pathology models for accurate and quantifiable diagnosis.

Dataset of prostate MRI annotated for anatomical zones and cancer.

In the present work, we present a publicly available, expert-segmented representative dataset of 158 3.0 Tesla biparametric MRIs [1]. There is an increasing number of studies investigating prostate and prostate carcinoma segmentation using deep learning (DL) with 3D architectures [2], [3], [4], [5], [6], [7]. The development of robust and data-driven DL models for prostate segmentation and assessment is currently limited by the availability of openly available expert-annotated datasets [8], [9], [10]. The dataset contains 3.0 Tesla MRI images of the prostate of patients with suspected prostate cancer. Patients over 50 years of age who had a 3.0 Tesla MRI scan of the prostate that met PI-RADS version 2.1 technical standards were included. All patients received a subsequent biopsy or surgery so that the MRI diagnosis could be verified/matched with the histopathologic diagnosis. For patients who had undergone multiple MRIs, the last MRI, which was less than six months before biopsy/surgery, was included. All patients were examined at a German university hospital (Charité Universitätsmedizin Berlin) between 02/2016 and 01/2020. All MRI were acquired with two 3.0 Tesla MRI scanners (Siemens VIDA and Skyra, Siemens Healthineers, Erlangen, Germany). Axial T2W sequences and axial diffusion-weighted sequences (DWI) with apparent diffusion coefficient maps (ADC) were included in the data set. T2W sequences and ADC maps were annotated by two board-certified radiologists with 6 and 8 years of experience, respectively. For T2W sequences, the central gland (central zone and transitional zone) and peripheral zone were segmented. If areas of suspected prostate cancer (PIRADS score of ≥ 4) were identified on examination, they were segmented in both the T2W sequences and ADC maps. Because restricted diffusion is best seen in DWI images with high b-values, only these images were selected and all images with low b-values were discarded. Data were then anonymized and converted to NIfTI (Neuroimaging Informatics Technology Initiative) format.

Changing the Patient's Position: Pitfalls and Benefits for Radiation Dose and Image Quality of Computed Tomography in Polytrauma.

For computed tomography (CT), representing the diagnostic standard for trauma patients, image quality is essential. The positioning of the patient's arms next to the abdomen causes artifacts and is also considered to increase radiation exposure. The aim of this study was to evaluate the effect of various positionings during different CT examination steps on the extent of artifacts as well as radiation dose using iterative reconstruction (IR). 354 trauma-CTs were analyzed retrospectively. All datasets were reconstructed using IR and three different examination protocols were applied. Arm elevation led to a significant improvement of the image quality across all examination protocols (p < 0.001). Variation in arm positioning during image acquisition did not lead to a reduction of radiation dose (p = 0.123). Only elevation during scout acquisition resulted in the reduction of radiation exposure (p < 0.001). To receive high-quality CT images, patients should be placed with elevated arms for the trunk scan, as artifacts remain even with the IR. Arm repositioning during the examination itself had no effect on the applied radiation dose because its modulation refers to the initial scout obtained. In order to achieve a dose effect by different positioning, a two-scout protocol (dual scout) should be used.

Valve embolization during transcatheter aortic valve implantation: Incidence, risk factors and follow-up by computed tomography.

Background: In most cases of transcatheter valve embolization and migration (TVEM), the embolized valve remains in the aorta after implantation of a second valve into the aortic root. There is little data on potential late complications such as valve thrombosis or aortic wall alterations by embolized valves. Aims: The aim of this study was to analyze the incidence of TVEM in a large cohort of patients undergoing transcatheter aortic valve implantation (TAVI) and to examine embolized valves by computed tomography (CT) late after TAVI. Methods: The patient database of our center was screened for cases of TVEM between July 2009 and July 2021. To identify risk factors, TVEM cases were compared to a cohort of 200 consecutive TAVI cases. Out of 35 surviving TVEM patients, ten patients underwent follow-up by echocardiography and CT. Results: 54 TVEM occurred in 3757 TAVI procedures, 46 cases were managed percutaneously. Horizontal aorta (odds ratio [OR] 7.51, 95% confidence interval [CI] 3.4-16.6, p < 0.001), implantation of a self-expanding valve (OR 4.63, 95% CI 2.2-9.7, p < 0.01) and a left ventricular ejection fraction < 40% (OR 2.94, 95% CI 1.1-7.3, p = 0.016) were identified as risk factors for TVEM. CT scans were performed on average 26.3 months after TAVI (range 2-84 months) and detected hypoattenuated leaflet thickening (HALT) in two patients as well as parts of the stent frame protruding into the aortic wall in three patients. Conclusion: TVEM represents a rare complication of TAVI. Follow up-CT detected no pathological findings requiring intervention.

Bail-Out Implantation of an Embolized Balloon-Expandable Transcatheter Heart Valve in the Aortic Arch Using Self-Expandable Stents.

Embolization of a balloon expandable valve during transcatheter aortic valve implantation (TAVR) is a rare complication which generally can be managed by implantation of the embolized valve into the aorta. We present a TAVR case where the combination of an ascending aortic aneurysm and a narrow aortic arch precluded implantation of an embolized balloon-expandable valve into either the ascending and descending aorta. As a bailout strategy, the embolized valve was secured in the aortic arch using two self-expandable stents. Six month after the procedure, computed tomography confirmed a stable valve position with unobstructed blood flow into the supra-aortic arteries.

Deep Learning Detects Changes Indicative of Axial Spondyloarthritis at MRI of Sacroiliac Joints.

Background MRI is frequently used for early diagnosis of axial spondyloarthritis (axSpA). However, evaluation is time-consuming and requires profound expertise because noninflammatory degenerative changes can mimic axSpA, and early signs may therefore be missed. Deep neural networks could function as assistance for axSpA detection. Purpose To create a deep neural network to detect MRI changes in sacroiliac joints indicative of axSpA. Materials and Methods This retrospective multicenter study included MRI examinations of five cohorts of patients with clinical suspicion of axSpA collected at university and community hospitals between January 2006 and September 2020. Data from four cohorts were used as the training set, and data from one cohort as the external test set. Each MRI examination in the training and test sets was scored by six and seven raters, respectively, for inflammatory changes (bone marrow edema, enthesitis) and structural changes (erosions, sclerosis). A deep learning tool to detect changes indicative of axSpA was developed. First, a neural network to homogenize the images, then a classification network were trained. Performance was evaluated with use of area under the receiver operating characteristic curve (AUC), sensitivity, and specificity. P < .05 was considered indicative of statistically significant difference. Results Overall, 593 patients (mean age, 37 years ± 11 [SD]; 302 women) were studied. Inflammatory and structural changes were found in 197 of 477 patients (41%) and 244 of 477 (51%), respectively, in the training set and 25 of 116 patients (22%) and 26 of 116 (22%) in the test set. The AUCs were 0.94 (95% CI: 0.84, 0.97) for all inflammatory changes, 0.88 (95% CI: 0.80, 0.95) for inflammatory changes fulfilling the Assessment of SpondyloArthritis international Society definition, and 0.89 (95% CI: 0.81, 0.96) for structural changes indicative of axSpA. Sensitivity and specificity on the external test set were 22 of 25 patients (88%) and 65 of 91 patients (71%), respectively, for inflammatory changes and 22 of 26 patients (85%) and 70 of 90 patients (78%) for structural changes. Conclusion Deep neural networks can detect inflammatory or structural changes to the sacroiliac joint indicative of axial spondyloarthritis at MRI. © RSNA, 2022 Online supplemental material is available for this article.

Prostate158 - An expert-annotated 3T MRI dataset and algorithm for prostate cancer detection.

BACKGROUND: The development of deep learning (DL) models for prostate segmentation on magnetic resonance imaging (MRI) depends on expert-annotated data and reliable baselines, which are often not publicly available. This limits both reproducibility and comparability. METHODS: Prostate158 consists of 158 expert annotated biparametric 3T prostate MRIs comprising T2w sequences and diffusion-weighted sequences with apparent diffusion coefficient maps. Two U-ResNets trained for segmentation of anatomy (central gland, peripheral zone) and suspicious lesions for prostate cancer (PCa) with a PI-RADS score of ≥4 served as baseline algorithms. Segmentation performance was evaluated using the Dice similarity coefficient (DSC), the Hausdorff distance (HD), and the average surface distance (ASD). The Wilcoxon test with Bonferroni correction was used to evaluate differences in performance. The generalizability of the baseline model was assessed using the open datasets Medical Segmentation Decathlon and PROSTATEx. RESULTS: Compared to Reader 1, the models achieved a DSC/HD/ASD of 0.88/18.3/2.2 for the central gland, 0.75/22.8/1.9 for the peripheral zone, and 0.45/36.7/17.4 for PCa. Compared with Reader 2, the DSC/HD/ASD were 0.88/17.5/2.6 for the central gland, 0.73/33.2/1.9 for the peripheral zone, and 0.4/39.5/19.1 for PCa. Interrater agreement measured in DSC/HD/ASD was 0.87/11.1/1.0 for the central gland, 0.75/15.8/0.74 for the peripheral zone, and 0.6/18.8/5.5 for PCa. Segmentation performances on the Medical Segmentation Decathlon and PROSTATEx were 0.82/22.5/3.4; 0.86/18.6/2.5 for the central gland, and 0.64/29.2/4.7; 0.71/26.3/2.2 for the peripheral zone. CONCLUSIONS: We provide an openly accessible, expert-annotated 3T dataset of prostate MRI and a reproducible benchmark to foster the development of prostate segmentation algorithms.

Cooling Effects Occur in Hepatic Microwave Ablation At Low Vascular Flow Rates and in Close Proximity to Liver Vessels - Ex Vivo.

Background. The impact of vascular cooling effects in hepatic microwave ablation (MWA) is controversially discussed. The objective of this study was a systematic assessment of vascular cooling effects in hepatic MWA ex vivo. Methods. Microwave ablations were performed in fresh porcine liver ex vivo with a temperature-controlled MWA generator (902-928 MHz) and a non-cooled 14-G-antenna. Energy input was set to 9.0 kJ. Hepatic vessels were simulated by glass tubes. Three different vessel diameters (3.0, 5.0, 8.0 mm) and vessel to antenna distances (5, 10, 20 mm) were examined. Vessels were perfused with saline solution at nine different flow rates (0-500 mL/min). Vascular cooling effects were assessed at the largest cross-sectional ablation area. A quantitative and semi-quantitative/morphologic analysis was carried out. Results. 228 ablations were performed. Vascular cooling effects were observed at close (5 mm) and medium (10 mm) antenna to vessel distances (P < .05). Vascular cooling effects occurred around vessels with flow rates ≥1.0 mL/min (P < .05) and a vessel diameter ≥3 mm (P < .05). Higher flow rates did not result in more distinct cooling effects (P > .05). No cooling effects were measured at large (20 mm) antenna to vessel distances (P > .05). Conclusion. Vascular cooling effects occur in hepatic MWA and should be considered in treatment planning. The vascular cooling effect was mainly affected by antenna to vessel distance. Vessel diameter and vascular flow rate played a minor role in vascular cooling effects.

Do submillisievert-chest CT protocols impact diagnostic quality in suspected COVID-19 patients?

BACKGROUND: During the ongoing global SARS-CoV-2 pandemic, there is a high demand for quick and reliable methods for early identification of infected patients. Due to its widespread availability, chest-CT is commonly used to detect early pulmonary manifestations and for follow-ups. PURPOSE: This study aims to analyze image quality and reproducibility of readings of scans using low-dose chest CT protocols in patients suspected of SARS-CoV-2 infection. MATERIALS AND METHODS: Two radiologists retrospectively analyzed 100 low-dose chest CT scans of patients suspected of SARS-CoV-2 infection using two protocols on devices from two vendors regarding image quality based on a Likert scale. After 3 weeks, quality ratings were repeated to allow for analysis of intra-reader in addition to the inter-reader agreement. Furthermore, radiation dose and presence as well as distribution of radiological features were noted. RESULTS: The exams' effective radiation doses were in median in the submillisievert range (median of 0.53 mSv, IQR: 0.35 mSv). While most scans were rated as being of optimal quality, 38% of scans were scored as suboptimal, yet only one scan was non-diagnostic. Inter-reader and intra-reader reliability showed almost perfect agreement with Cohen's kappa of 0.82 and 0.87. CONCLUSION: Overall, in this study, we present two protocols for submillisievert low-dose chest CT demonstrating appropriate or better image quality with almost perfect inter-reader and intra-reader agreement in patients suspected of SARS-CoV-2 infection.

Deep learning for accurately recognizing common causes of shoulder pain on radiographs.

OBJECTIVE: Training a convolutional neural network (CNN) to detect the most common causes of shoulder pain on plain radiographs and to assess its potential value in serving as an assistive device to physicians. MATERIALS AND METHODS: We used a CNN of the ResNet-50 architecture which was trained on 2700 shoulder radiographs from clinical practice of multiple institutions. All radiographs were reviewed and labeled for six findings: proximal humeral fractures, joint dislocation, periarticular calcification, osteoarthritis, osteosynthesis, and joint endoprosthesis. The trained model was then evaluated on a separate test dataset, which was previously annotated by three independent expert radiologists. Both the training and the test datasets included radiographs of highly variable image quality to reflect the clinical situation and to foster robustness of the CNN. Performance of the model was evaluated using receiver operating characteristic (ROC) curves, the thereof derived AUC as well as sensitivity and specificity. RESULTS: The developed CNN demonstrated a high accuracy with an area under the curve (AUC) of 0.871 for detecting fractures, 0.896 for joint dislocation, 0.945 for osteoarthritis, and 0.800 for periarticular calcifications. It also detected osteosynthesis and endoprosthesis with near perfect accuracy (AUC 0.998 and 1.0, respectively). Sensitivity and specificity were 0.75 and 0.86 for fractures, 0.95 and 0.65 for joint dislocation, 0.90 and 0.86 for osteoarthrosis, and 0.60 and 0.89 for calcification. CONCLUSION: CNNs have the potential to serve as an assistive device by providing clinicians a means to prioritize worklists or providing additional safety in situations of increased workload.

Evaluation of potential tissue heating during percutaneous drill-assisted bone sampling in an in vivo porcine study.

BACKGROUND: Minimally invasive, battery-powered drilling systems have become the preferred tool for obtaining representative samples from bone lesions. However, the heat generated during battery-powered bone drilling for bone biopsies has not yet been sufficiently investigated. Thermal necrosis can occur if the bone temperature exceeds a critical threshold for a certain period of time. PURPOSE: To investigate heat production as a function of femur temperature during and after battery-powered percutaneous bone drilling in a porcine in vivo model. METHODS: We performed 16 femur drillings in 13 domestic pigs with an average age of 22 weeks and an average body temperature of 39.7 °C, using a battery-powered drilling system and an intraosseous temperature monitoring device. The standardized duration of the drilling procedure was 20 s. The bone core specimens obtained were embedded in 4% formalin, stained with haematoxylin and eosin (H&E) and sent for pathological analysis of tissue quality and signs of thermal damage. RESULTS: No significant changes in the pigs' local temperature were observed after bone drilling with a battery-powered drill device. Across all measurements, the median change in temperature between the initial measurement and the temperature measured after drilling (at 20 s) was 0.1 °C. Histological examination of the bone core specimens revealed no signs of mechanical or thermal damage. CONCLUSION: Overall, this preliminary study shows that battery-powered, drill-assisted harvesting of bone core specimens does not appear to cause mechanical or thermal damage.

Detectability of Head and Neck Cancer via New Computed Tomography Reconstruction Tools including Iterative Reconstruction and Metal Artifact Reduction.

State-of-the-art technology in Computed Tomography (CT) includes iterative reconstruction algorithms (IR) and metal artefact reduction (MAR) techniques. The objective of the study is to show the benefits of this technology for the detection of primary and recurrent head and neck cancer. A total of 131 patients underwent contrast-enhanced CT for diagnosis of primary and recurrent Head and Neck cancer; 110 patients were included. All scans were reconstructed using iterative reconstruction, and metal artifact reduction was applied when indicated. Tumor detectability was evaluated dichotomously. Histopathological findings were used as a standard of reference. Data were analyzed retrospectively, statistics was performed through diagnostic test characteristics. State-of-the-art Head and Neck CT showed a sensitivity of 0.83 (95% CI; 0.61-0.95) with 0.93 specificity (95% CI; 0.84-0.98) for primary tumor detection. Recurrent tumors were identified with a 0.94 sensitivity (95% CI; 0.71-0.99) and 0.93 specificity (95% CI; 0.84-0.98) in this study. Conclusion: State-of-the-art reconstruction tools improve the diagnostic quality of Head and Neck CT, especially for recurrent tumor detection, compared with data published for standard CT. IR and MAR are easily implemented in routine clinical settings and improve image evaluation by reducing artifacts and image noise while lowering radiation exposure.

CT diagnostics of pulmonary embolism: Does iodine delivery rate still affect image quality in iterative reconstruction?

BACKGROUND: Computed tomographic (CT) imaging in suspected pulmonary artery embolism represents the standard procedure. Studies without iterative reconstruction proved beneficial using increased iodine delivery rate (IDR). This study compares image quality in pulmonary arteries on iteratively reconstructed CT images of patients with suspected pulmonary embolism using different IDR. MATERIAL AND METHODS: 1065 patients were included in the study. Patients in group A (n = 493) received an iodine concentration of 40 g/100 ml (IDR 1.6 g/s) and patients in group B (n = 572) an iodine concentration of 35 g/100 ml (IDR 1.4 g/s) at a flow rate of 4 ml/s. A 80-detector spiral CT scanner with iterative reconstruction was used. We measured mean density values in truncus pulmonalis, both pulmonary arteries and segmental pulmonary arteries. Subjectively, the contrast of apical and basal pulmonary arteries was determined on a 4-point Likert scale. RESULTS: Radiodensity was significantly higher in all measured pulmonary arteries using the increased IDR (p < 0.001). TP: 483.0 HU vs. 393.4 HU; APD: 452.1 HU vs. 372.1 HU; APS: 448.2 HU vs. 374.4 HU; ASP: 443.9 vs. 374.4 HU. Subjectively assessed contrast enhancement in apical (p = 0.077) and basal (p = 0.429) lung sections showed no significant differences. CONCLUSION: Higher IDR improves objective image quality in all patients with significantly higher radiodensities by iterative reconstruction. Subjective contrast of apical and basal lung sections did not differ. The number of non-sufficient scans decreased with high IDR.

Computed Tomography Imaging in Simulated Ongoing Cardiopulmonary Resuscitation: No Need to Switch Off the Chest Compression Device during Image Acquisition.

Computed tomography (CT) represents the current standard for imaging of patients with acute life-threatening diseases. As some patients present with circulatory arrest, they require cardiopulmonary resuscitation. Automated chest compression devices are used to continue resuscitation during CT examinations, but tend to cause motion artifacts degrading diagnostic evaluation of the chest. The aim was to investigate and evaluate a CT protocol for motion-free imaging of thoracic structures during ongoing mechanical resuscitation. The standard CT trauma protocol and a CT protocol with ECG triggering using a simulated ECG were applied in an experimental setup to examine a compressible thorax phantom during resuscitation with two different compression devices. Twenty-eight phantom examinations were performed, 14 with AutoPulse® and 14 with corpuls cpr®. With each device, seven CT examinations were carried out with ECG triggering and seven without. Image quality improved significantly applying the ECG-triggered protocol (p < 0.001), which allowed almost artifact-free chest evaluation. With the investigated protocol, radiation exposure was 5.09% higher (15.51 mSv vs. 14.76 mSv), and average reconstruction time of CT scans increased from 45 to 76 s. Image acquisition using the proposed CT protocol prevents thoracic motion artifacts and facilitates diagnosis of acute life-threatening conditions during continuous automated chest compression.

De Novo Radiomics Approach Using Image Augmentation and Features From T1 Mapping to Predict Gleason Scores in Prostate Cancer.

OBJECTIVES: The aims of this study were to discriminate among prostate cancers (PCa's) with Gleason scores 6, 7, and ≥8 on biparametric magnetic resonance imaging (bpMRI) of the prostate using radiomics and to evaluate the added value of image augmentation and quantitative T1 mapping. MATERIALS AND METHODS: Eighty-five patients with subsequently histologically proven PCa underwent bpMRI at 3 T (T2-weighted imaging, diffusion-weighted imaging) with 66 patients undergoing additional T1 mapping at 3 T. The PCa lesions as well as the peripheral and transition zones were segmented pixel by pixel in multiple slices of the 3D MRI data sets (T2-weighted images, apparent diffusion coefficient, and T1 maps). To increase the size of the data set, images were augmented for contrast, brightness, noise, and perspective multiple times, effectively increasing the sample size 10-fold, and 322 different radiomics features were extracted before and after augmentation. Four different machine learning algorithms, including a random forest (RF), stochastic gradient boosting (SGB), support vector machine (SVM), and k-nearest neighbor, were trained with and without features from T1 maps to differentiate among 3 different Gleason groups (6, 7, and ≥8). RESULTS: Support vector machine showed the highest accuracy of 0.92 (95% confidence interval [CI], 0.62-1.00) for classifying the different Gleason scores, followed by RF (0.83; 95% CI, 0.52-0.98), SGB (0.75; 95% CI, 0.43-0.95), and k-nearest neighbor (0.50; 95% CI, 0.21-0.79). Image augmentation resulted in an average increase in accuracy between 0.08 (SGB) and 0.48 (SVM). Removing T1 mapping features led to a decline in accuracy for RF (-0.16) and SGB (-0.25) and a higher generalization error. CONCLUSIONS: When data are limited, image augmentations and features from quantitative T1 mapping sequences might help to achieve higher accuracy and lower generalization error for classification among different Gleason groups in bpMRI by using radiomics.

Deep-Learning-Based Diagnosis of Bedside Chest X-ray in Intensive Care and Emergency Medicine.

OBJECTIVES: Validation of deep learning models should separately consider bedside chest radiographs (CXRs) as they are the most challenging to interpret, while at the same time the resulting diagnoses are important for managing critically ill patients. Therefore, we aimed to develop and evaluate deep learning models for the identification of clinically relevant abnormalities in bedside CXRs, using reference standards established by computed tomography (CT) and multiple radiologists. MATERIALS AND METHODS: In this retrospective study, a dataset consisting of 18,361 bedside CXRs of patients treated at a level 1 medical center between January 2009 and March 2019 was used. All included CXRs occurred within 24 hours before or after a chest CT. A deep learning algorithm was developed to identify 8 findings on bedside CXRs (cardiac congestion, pleural effusion, air-space opacification, pneumothorax, central venous catheter, thoracic drain, gastric tube, and tracheal tube/cannula). For the training dataset, 17,275 combined labels were extracted from the CXR and CT reports by a deep learning natural language processing (NLP) tool. In case of a disagreement between CXR and CT, human-in-the-loop annotations were used. The test dataset consisted of 583 images, evaluated by 4 radiologists. Performance was assessed by area under the receiver operating characteristic curve analysis, sensitivity, specificity, and positive predictive value. RESULTS: Areas under the receiver operating characteristic curve for cardiac congestion, pleural effusion, air-space opacification, pneumothorax, central venous catheter, thoracic drain, gastric tube, and tracheal tube/cannula were 0.90 (95% confidence interval [CI], 0.87-0.93; 3 radiologists on the receiver operating characteristic [ROC] curve), 0.95 (95% CI, 0.93-0.96; 3 radiologists on the ROC curve), 0.85 (95% CI, 0.82-0.89; 1 radiologist on the ROC curve), 0.92 (95% CI, 0.89-0.95; 1 radiologist on the ROC curve), 0.99 (95% CI, 0.98-0.99), 0.99 (95% CI, 0.98-0.99), 0.98 (95% CI, 0.97-0.99), and 0.99 (95% CI, 0.98-1.00), respectively. CONCLUSIONS: A deep learning model used specifically for bedside CXRs showed similar performance to expert radiologists. It could therefore be used to detect clinically relevant findings during after-hours and help emergency and intensive care physicians to focus on patient care.

Improving CT accuracy in the diagnosis of COVID-19 in a hospital setting.

OBJECTIVE: This study aimed to improve the accuracy of CT for detection of COVID-19-associated pneumonia and to identify patient subgroups who might benefit most from CT imaging. METHODS: A total of 269 patients who underwent CT for suspected COVID-19 were included in this retrospective analysis. COVID-19 was confirmed by reverse-transcription-polymerase-chain-reaction. Basic demographics (age and sex) and initial vital parameters (O2-saturation, respiratory rate, and body temperature) were recorded. Generalized mixed models were used to calculate the accuracy of vital parameters for detection of COVID-19 and to evaluate the diagnostic accuracy of CT. A clinical score based on vital parameters, age, and sex was established to estimate the pretest probability of COVID-19 and used to define low, intermediate, and high risk groups. A p-value of <0.05 was considered statistically significant. RESULTS: The sole use of vital parameters for the prediction of COVID-19 was inferior to CT. After correction for confounders, such as age and sex, CT showed a sensitivity of 0.86, specificity of 0.78, and positive predictive value of 0.36. In the subgroup analysis based on pretest probability, positive predictive value and sensitivity increased to 0.53 and 0.89 in the high-risk group, while specificity was reduced to 0.68. In the low-risk group, sensitivity and positive predictive value decreased to 0.76 and 0.33 with a specificity of 0.83. The negative predictive value remained high (0.94 and 0.97) in both groups. CONCLUSIONS: The accuracy of CT for the detection of COVID-19 might be increased by selecting patients with a high-pretest probability of COVID-19.