A novel abnormality annotation database for covid-19 affected frontal lung x-rays

A Novel Abnormality Annotation Database For Covid-19 Affected Frontal Lung X-Rays



To advance the usage of CXRs as a viable solution for efficient COVID-19 diagnostics by providing large-scale annotations of the abnormalities in frontal CXRs in BIMCV-COVID19+ database, and to provide a robust evaluation mechanism to facilitate its usage.

Materials and Methods:

We provide the abnormality annotations in frontal CXRs by creating bounding boxes. The frontal CXRs are a part of the existing BIMCV-COVID19+ database. We also define four different protocols for robust evaluation of semantic segmentation and classification algorithms. Finally, we benchmark the defined protocols and report the results using popular deep learning models as a part of this study.


For semantic segmentation, Mask-RCNN performs the best among all the models with a DICE score of 0.43 ± 0.01. For classification, we observe that MobileNetv2 yields the best results for 2-class and 3-class classification. We also observe that deep models report a lower performance for classifying other classes apart from the COVID class.


By making the annotated data and protocols available to the scientific community, we aim to advance the usage of CXRs as a viable solution for efficient COVID-19 diagnostics. This large-scale data will be useful for ML algorithms and can be used for learning radiological patterns observed in COVID-19 patients. Further, the protocols will facilitate ML practitioners for unified large-scale evaluation of their algorithms.

For full paper: http://https://www.medrxiv.org/content/10.1101/2021.01.07.21249323v2

A Systematic Meta-Analysis of CT Features of COVID-19: Lessons from Radiology

Vasantha Kumar Venugopal, Vidur Mahajan, Sriram Rajan, Vikash Kumar Agarwal, Ruchika Rajan, Salsabeel Syed, Harsh Mahajan

This article is a preprint and has not been peer-reviewed [what does this mean?]. It reports new medical research that has yet to be evaluated and so should not be used to guide clinical practice.


Several studies have been published in the past few months describing the CT features of Coronavirus Disease 2019 (COVID-19). There is a great degree of heterogeneity in the study designs, lesion descriptors used and conclusions derived. In our systematic analysis and meta-review, we have attempted to homogenize the reported features and provide a comprehensive view of the disease pattern and progression in different clinical stages. After an extensive literature search, we short-listed and reviewed 49 studies including over 4145 patients with 3615 RT-PCR positive cases of COVID-19 disease. We have found that there is a good agreement among these studies that diffuse bilateral ground-glass opacities (GGOs) is the most common finding at all stages of the disease followed by consolidations and mixed density lesions. 78% of patients with RT-PCR confirmed COVID-19 infections had either ground-glass opacities, consolidation or both. Inter-lobular septal thickening was also found to be a common feature in many patients in advanced stages. The progression of these initial patchy ground-glass opacities and consolidations to diffuse lesions with septal thickening, air bronchograms in the advanced stages, to either diffuse white-out lungs needing ICU admissions or finally resolving completely without or with residual fibrotic strips was also found to be congruent among multiple studies. Prominent juxta-lesional pulmonary vessels, pleural effusion and lymphadenopathy in RT-PCR proven cases were found to have poor clinical prognosis. Additionally, we noted wide variation in terminology used to describe lesions across studies and suggest the use of standardized lexicons to describe findings related to diseases of vital importance.

For Full Paper: https://www.medrxiv.org/content/10.1101/2020.04.04.20052241v1

Unboxing AI – Radiological Insights Into a Deep Neural Network for Lung Nodule Characterization

Rationale and Objectives: To explain predictions of a deep residual convolutional network for characterization of lung nodule by analyzing heat maps Materials and Methods A 20-layer deep residual CNN was trained on 1245 Chest CTs from NLST trial to predict the malignancy risk of a nodule. We used occlusion to systematically block regions of a nodule and map drops in malignancy risk score to generate clinical attribution heatmaps on 103 nodules from LIDC-IDRI dataset, which were analyzed by a thoracic radiologist. The features were described as heat inside nodule (IH)-bright areas inside nodule, peripheral heat (PH)-continuous/interrupted bright areas along nodule contours, heat in adjacent plane (AH)-brightness in scan planes juxtaposed with the nodule, satellite heat (SH)- a smaller bright spot in proximity to nodule in the same scan plane, heat map larger than nodule (LH)-bright areas corresponding to the shape of the nodule seen outside the nodule margins and heat in calcification (CH) Results These six features were assigned binary values. This feature vector was fed into a standard J48 decision tree with 10-fold cross-validation, which gave an 85 % weighted classification accuracy with a 77.8 %TP rate, 8% FP rate for benign cases and 91.8% TP and 22.2 %FP rates for malignant cases. IH was more frequently observed in nodules classified as malignant whereas PH, AH, and SH were more commonly seen in nodules classified as benign. Conclusion We discuss the potential ability of a radiologist to visually parse the deep learning algorithm-generated \’heat map\’ to identify features aiding classification

For full paper: http://vixra.org/abs/1909.0398?ref=10792316

The Algorithmic Audit: Working with Vendors to Validate Radiology-AI Algorithms – How We Do It


There is a plethora of Artificial Intelligence (AI) tools that are being developed around the world aiming at either speeding up or improving the accuracy of radiologists. It is essential for radiologists to work with the developers of such algorithms to determine true clinical utility and risks associated with these algorithms. We present a framework, called an Algorithmic Audit, for working with the developers of such algorithms to test and improve the performance of the algorithms. The framework includes concepts of true independent validation on data that the algorithm has not seen before, curating datasets for such testing, deep examination of false positives and false negatives (to examine implications of such errors) and real-world deployment and testing of algorithms.

Link – http://vixra.org/abs/1909.0104