Background and aims Overall obesity has recently been established as an independent risk factor for critical illness in patients with coronavirus disease 2019 (COVID-19). The role of fat distribution and especially that of visceral fat, which is often associated with metabolic syndrome, remains unclear. Therefore, this study aims at investigating the association between fat distribution and COVID-19 severity. Methods Thirty patients with COVID-19 and a mean age of 65.6 ± 13.1 years from a level-one medical center in Berlin, Germany, were included in the present cross-sectional analysis. COVID-19 was confirmed by polymerase chain reaction (PCR) from nasal and throat swabs. A severe clinical course of COVID-19 was defined by hospitalization in the intensive care unit (ICU) and/or invasive mechanical ventilation. Fat was measured at the level of the first lumbar vertebra on routinely acquired low-dose chest computed tomography (CT). Results An increase in visceral fat area (VFA) by ten square centimeters was associated with a 1.37-fold higher likelihood of ICU treatment and a 1.32-fold higher likelihood of mechanical ventilation (adjusted for age and sex). For upper abdominal circumference, each additional centimeter of circumference was associated with a 1.13-fold higher likelihood of ICU treatment and a 1.25-fold higher likelihood of mechanical ventilation. Conclusions Our proof-of-concept study suggests that visceral adipose tissue and upper abdominal circumference specifically increase the likelihood of COVID-19 severity. CT-based quantification of visceral adipose tissue and upper abdominal circumference in routine chest CTs may therefore be a simple tool for risk assessment in COVID-19 patients.

Head and neck squamous cell carcinoma (HNSC) patients are at risk of suffering from both pulmonary metastases or a second squamous cell carcinoma of the lung (LUSC). Differentiating pulmonary metastases from primary lung cancers is of high clinical importance, but not possible in most cases with current diagnostics. To address this, we performed DNA methylation profiling of primary tumors and trained three different machine learning methods to distinguish metastatic HNSC from primary LUSC. We developed an artificial neural network that correctly classified 96.4% of the cases in a validation cohort of 279 patients with HNSC and LUSC as well as normal lung controls, outperforming support vector machines (95.7%) and random forests (87.8%). Prediction accuracies of more than 99% were achieved for 92.1% (neural network), 90% (support vector machine), and 43% (random forest) of these cases by applying thresholds to the resulting probability scores and excluding samples with low confidence. As independent clinical validation of the approach, we analyzed a series of 51 patients with a history of HNSC and a second lung tumor, demonstrating the correct classifications based on clinicopathological properties. In summary, our approach may facilitate the reliable diagnostic differentiation of pulmonary metastases of HNSC from primary LUSC to guide therapeutic decisions.

Chest radiographs are among the most frequently acquired images in radiology and are often the subject of computer vision research. However, most of the models used to classify chest radiographs are derived from openly available deep neural networks, trained on large image datasets. These datasets differ from chest radiographs in that they are mostly color images and have substantially more labels. Therefore, very deep convolutional neural networks (CNN) designed for ImageNet and often representing more complex relationships, might not be required for the comparably simpler task of classifying medical image data. Sixteen different architectures of CNN were compared regarding the classification performance on two openly available datasets, the CheXpert and COVID-19 Image Data Collection. Areas under the receiver operating characteristics curves (AUROC) between 0.83 and 0.89 could be achieved on the CheXpert dataset. On the COVID-19 Image Data Collection, all models showed an excellent ability to detect COVID-19 and non-COVID pneumonia with AUROC values between 0.983 and 0.998. It could be observed, that more shallow networks may achieve results comparable to their deeper and more complex counterparts with shorter training times, enabling classification performances on medical image data close to the state-of-the-art methods even when using limited hardware. Chest radiographs are among the most frequently used imaging procedures in radiology. They have been widely employed in the field of computer vision, as chest radiographs are a standardized technique and, if compared to other radiological examinations such as computed tomography or magnetic resonance imaging, contain a smaller group of relevant pathologies. Although many artificial neural networks for the classification of chest radiographs have been developed, it is still subject to intensive research. Only a few groups design their own networks from scratch, while most use already established architectures, such as ResNet-50 or DenseNet-121 (with 50 and 121 representing the number of layers within the respective neural network) 1-6. These neural networks have often been trained on large, openly available datasets, such as ImageNet, and are therefore already able to recognize numerous image features. When training a model for a new task, such as the classification of chest radiographs, the use of pre-trained networks may improve the training speed and accuracy of the new model, since important image features that have already been learned can be transferred to the new task and do not have to be learned again. However, the feature space of freely available data sets such as ImageNet differs from chest radiographs as they contain color images and more categories. The ImageNet Challenge includes 1,000 possible categories per image, while CheXpert, a large freely available data set of chest radiographs, only distinguishes between 14 categories (or classes) 7 , and the COVID-19 Image Data Collection only differentiates between three classes 8. Although the ImageNet...

BackgroundRenal cell carcinoma (RCC) are accompanied by inferior vena cava (IVC) thrombus in up to 10% of the cases, with surgical resection remaining the only curative option. In case of IVC wall invasion, the operative procedure is more challenging and may even require IVC resection. This study aims to determine the diagnostic performance of contrast-enhanced magnetic resonance imaging (MRI) for the assessment of wall invasion by IVC thrombus in patients with RCC, validated with intraoperative findings.MethodsData were collected on 81 patients with RCC and IVC thrombus, who received a radical nephrectomy and vena cava thrombectomy between February 2008 and November 2017. Forty eight patients met the inclusion criteria. Sensitivity and specificity as well as the positive and negative predictive values were calculated for preoperative MRI, based on the assessments of the two readers for visual wall invasion. Furthermore, a logistic regression model was used to determine if there was an association between intraoperative wall adherence and IVC diameter.ResultsComplete occlusion of the IVC lumen or vessel breach could reliably assess IVC wall invasion with a sensitivity of 92.3% (95%-CI: 0.75–0.99) and a specificity of 86.4% (95%-CI: 0.65–0.97) (Fisher-test: p-value< 0.001). The positive predictive value (PPV) was 88.9% (95%-CI: 0.71–0.98) and the negative predictive value reached 90.5% (95%-CI: 0.70–0.99). There was an excellent interobserver agreement for determining IVC wall invasion with a kappa coefficient of 0.90 (95%CI: 0.79–1.00).ConclusionsThe present study indicates that standard preoperative MR imaging can be used to reliably assess IVC wall invasion, evaluating morphologic features such as the complete occlusion of the IVC lumen or vessel breach. Increases in IVC diameter are associated with a higher probability of IVC wall invasion.

Objectives The aims of this study were to identify higher-grade clear cell renal cell carcinoma (cRCC) with native T1 mapping and to histologically correlate the results with the collagen volume fraction. Materials and Methods For this institutional review board–approved, single-center prospective study, 68 consecutive patients received abdominal magnetic resonance imaging scans at 1.5 T between January 2017 and July 2018, using a Modified Look-Locker Inversion Recovery (MOLLI) sequence. Thirty patients with cRCC (20 men; mean age, 61.9 ± 13.1 years) who underwent partial or radical nephrectomy and histological grading according to the International Society of Urological Pathology (ISUP) classification and a separate healthy cohort of 30 individuals without renal malignancies or complex cysts (16 men; mean age, 59.7 ± 14.6 years) met the eligibility criteria. T1 values were quantitatively measured with region of interest measurements in T1 maps. Quantification of the collagen volume fraction was performed on histological sections (picrosirius red staining). Results Native T1 values were significantly lower for lower-grade cRCC (ISUP 1 and 2) compared with higher-grade cRCC (ISUP 3 and 4; P < 0.001). A cutoff value of 1101 milliseconds distinguished higher-grade from lower-grade tumors with a sensitivity of 100% (95% confidence interval [CI], 0.69–1.00), a specificity of 85% (95% CI, 0.62–0.97), and an accuracy of 90% (95% CI, 0.73–0.98). Native T1 values were significantly associated with the histological collagen volume fraction (P < 0.05). Furthermore, T1 times in the renal cortex, medulla, and tumor tissue showed an excellent interobserver agreement. Conclusions Native T1 mapping could represent an in vivo biomarker for the differentiation of lower- and higher-grade cRCCs, providing incremental diagnostic value beyond qualitative magnetic resonance imaging features.

Background Correct staging and grading of patients with clear cell renal cell carcinoma (cRCC) is of clinical relevance for the prediction of operability and for individualized patient management. As partial or radial resection with postoperative tumor grading currently remain the methods of choice for the classification of cRCC, non-invasive preoperative alternatives to differentiate lower grade from higher grade cRCC would be beneficial. Methods This institutional-review-board approved cross-sectional study included twenty-seven patients (8 women, mean age ± SD, 61.3 ± 14.2) with histopathologically confirmed cRCC, graded according to the International Society of Urological Pathology (ISUP). A native, balanced steady-state free precession T2 mapping sequence (TrueFISP) was performed at 1.5 T. Quantitative T2 values were measured with circular 2D ROIs in the solid tumor portion and also in the normal renal parenchyma (cortex and medulla). To estimate the optimal cut-off T2 value for identifying lower grade cRCC, a Receiver Operating Characteristic Curve (ROC) analysis was performed and sensitivity and specificity were calculated. Students’ t-tests were used to evaluate the differences in mean values for continuous variables, while intergroup differences were tested for significance with two-tailed Mann-Whitney-U tests. Results There were significant differences between the T2 values for lower grade (ISUP 1–2) and higher grade (ISUP 3–4) cRCC ( p < 0.001), with higher T2 values for lower grade cRCC compared to higher grade cRCC. The sensitivity and specificity for the differentiation of lower grade from higher grade tumors were 83.3% (95% CI: 0.59–0.96) and 88.9% (95% CI: 0.52–1.00), respectively, using a threshold value of ≥110 ms. Intraobserver/interobserver agreement for T2 measurements was excellent/substantial. Conclusions Native T2 mapping based on a balanced steady-state free precession MR sequence might support an image-based distinction between lower and higher grade cRCC in a two-tier-system and could be a helpful addition to multiparametric imaging. Electronic supplementary material The online version of this article (10.1186/s40644-019-0222-8) contains supplementary material, which is available to authorized users.

Motivation The development of deep, bidirectional transformers such as BERT (Bidirectional Encoder Representations from Transformers) led to an outperformance of several Natural Language Processing (NLP) benchmarks. Especially in radiology, large amounts of free-text data are generated in the daily clinical workflow. These report texts could be of particular use for the generation of labels in machine learning, especially for image classification. However, as report text are mostly unstructured, advanced NLP methods are needed for text classification. While neural networks can be used for this purpose, they must first be trained on large amounts of manually labelled data in order to achieve good results. In contrast, BERT models can be pre-trained on unlabelledunlabeled data and then only require fine tuning on a small amount of manually labelled data to achieve even better results. Results By using BERT to identify the most important findings in intensive care chest x-ray reports, we achieve areas under the receiver operation characteristics curve of 0.98 for congestion, 0.97 for effusion, 0.97 for consolidation and 0.99 for pneumothorax, surpassing the accuracy of previous approaches with comparatively little annotation effort. Our approach could help to improve information extraction from free-text medical reports. Availability :

Background Radiographs of the sacroiliac joints are commonly used for the diagnosis and classification of axial spondyloarthritis. The aim of this study was to develop and validate an artificial neural network for the detection of definite radiographic sacroiliitis as a manifestation of axial spondyloarthritis (axSpA). Methods Conventional radiographs of the sacroiliac joints obtained in two independent studies of patients with axSpA were used. The first cohort comprised 1553 radiographs and was split into training (n = 1324) and validation (n = 229) sets. The second cohort comprised 458 radiographs and was used as an independent test dataset. All radiographs were assessed in a central reading session, and the final decision on the presence or absence of definite radiographic sacroiliitis was used as a reference. The performance of the neural network was evaluated by calculating areas under the receiver operating characteristic curves (AUCs) as well as sensitivity and specificity. Cohen’s kappa and the absolute agreement were used to assess the agreement between the neural network and the human readers. Results The neural network achieved an excellent performance in the detection of definite radiographic sacroiliitis with an AUC of 0.97 and 0.94 for the validation and test datasets, respectively. Sensitivity and specificity for the cut-off weighting both measurements equally were 88% and 95% for the validation and 92% and 81% for the test set. The Cohen’s kappa between the neural network and the reference judgements were 0.79 and 0.72 for the validation and test sets with an absolute agreement of 90% and 88%, respectively. Conclusion Deep artificial neural networks enable the accurate detection of definite radiographic sacroiliitis relevant for the diagnosis and classification of axSpA.