Deep learning algorithms, in particular convolutional networks, have rapidly become a methodology of choice for analyzing medical images. This paper reviews the major deep learning concepts pertinent to medical image analysis and summarizes over 300 contributions to the field, most of which appeared in the last year. We survey the use of deep learning for image classification, object detection, segmentation, registration, and other tasks. Concise overviews are provided of studies per application area: neuro, retinal, pulmonary, digital pathology, breast, cardiac, abdominal, musculoskeletal. We end with a summary of the current state-of-the-art, a critical discussion of open challenges and directions for future research.

A method is presented for automated segmentation of vessels in two-dimensional color images of the retina. This method can be used in computer analyses of retinal images, e.g., in automated screening for diabetic retinopathy. The system is based on extraction of image ridges, which coincide approximately with vessel centerlines. The ridges are used to compose primitives in the form of line elements. With the line elements an image is partitioned into patches by assigning each image pixel to the closest line element. Every line element constitutes a local coordinate frame for its corresponding patch. For every pixel, feature vectors are computed that make use of properties of the patches and the line elements. The feature vectors are classified using a kappaNN-classifier and sequential forward feature selection. The algorithm was tested on a database consisting of 40 manually labeled images. The method achieves an area under the receiver operating characteristic curve of 0.952. The method is compared with two recently published rule-based methods of Hoover et al. and Jiang et al. The results show that our method is significantly better than the two rule-based methods (p < 0.01). The accuracy of our method is 0.944 versus 0.947 for a second observer.

statement Reporting and Data System, is a categorical assessment scheme for chest CT in patients suspected of COVID-19, representing the level of suspicion for pulmonary involvement. The substantial agreement among observers and its discriminatory value make it well-suited for use in clinical practice. Key results• CO-RADS, for COVID-19 Reporting and Data System, provides a standardized assessment scheme that simplifies reporting with a five-point scale of suspicion for pulmonary involvement of COVID-19 on chest CT • CO-RADS has a moderate to substantial agreement among observers with an overall Fleiss' kappa of 0.47 (95% CI 0.45-0.49).• The discriminatory power of CO-RADS for diagnosing COVID-19 was high, with a mean area under the ROC curve of 0.91 (95% CI 0.85-0.97) for positive RT-PCR results. AbbreviationsRT-PCR: reverse transcriptase-polymerase chain reaction. ROC: receiver operating characteristics. AUC: area under the ROC curve. CI: confidence interval. CT: computed tomography. IQR: interquartile range. Abstract PurposeTo introduce the COVID-19 Reporting and Data System (CO-RADS) for standardized assessment of pulmonary involvement of COVID-19 on non-enhanced chest CT and report its initial interobserver agreement and performance. MethodsThe Dutch Radiological Society (NVvR) developed CO-RADS based on other efforts for standardization, such as Lung-RADS or BI-RADS. CO-RADS assesses the suspicion for pulmonary involvement of COVID-19 on a scale from 1 (very low) to 5 (very high). The system is meant to be used in patients presenting with moderate to severe symptoms of COVID-19. The system was evaluated using 105 chest CTs of patients admitted to the hospital with clinical suspicion of COVID-19 in whom RT-PCR was performed (62 +/-16 years, 61 men, 53 with positive RT-PCR). Eight observers assessed the scans using CO-RADS. Fleiss' kappa was calculated, and scores of individual observers were compared to the median of the remaining seven observers. The resulting area under the receiver operating characteristics curve (AUC) was compared to results from RT-PCR and clinical diagnosis of COVID-19. ResultsThere was absolute agreement among observers in 573 (68.2%) of 840 observations. Fleiss' kappa was 0.47 (95% confidence interval (CI) 0.45-0.47), with the highest kappa for CO-RADS categories 1 (0.58, 95% CI 0.54-0.62) and 5 (0.68, 95% CI 0.65-0.72). The average AUC was 0.91 (95% CI 0.85-0.97) for predicting RT-PCR outcome and 0.95 (95% CI 0.91-0.99) for clinical diagnosis. The false negative rate for CO-RADS 1 was 9/161 (5.6%, 95% CI 1.0-10%), and the false positive rate for CO-RADS 5 was 1/286 (0.3%, 95% CI 0-1.0%). ConclusionsCO-RADS is a categorical assessment scheme for pulmonary involvement of COVID-19 on non-enhanced chest CT providing very good performance for predicting COVID-19 in patients with moderate to severe symptoms and has a substantial interobserver agreement, especially for categories 1 and 5.

In the setting of a challenge competition, some deep learning algorithms achieved better diagnostic performance than a panel of 11 pathologists participating in a simulation exercise designed to mimic routine pathology workflow; algorithm performance was comparable with an expert pathologist interpreting whole-slide images without time constraints. Whether this approach has clinical utility will require evaluation in a clinical setting.

No abstract