Performance of Radiologists in the Evaluation of the Chest Radiography with the Use of a “new software score” in Coronavirus Disease 2019 Pneumonia Suspected Patients

Abstract

Objectives: The purpose of this study is to assess the performance of radiologists using a new software called “COVID-19 score” when performing chest radiography on patients potentially infected by coronavirus disease 2019 (COVID-19) pneumonia. Chest radiography (or chest X-ray, CXR) and CT are important for the imaging diagnosis of the coronavirus pneumonia (COVID-19). CXR mobile devices are efficient during epidemies, because allow to reduce the risk of contagion and are easy to sanitize. Material and Methods: From February–April 2020, 14 radiologists retrospectively evaluated a pool of 312 chest X-ray exams to test a new software function for lung imaging analysis based on radiological features and graded on a three-point scale. This tool automatically generates a cumulative score (0–18). The intra- rater agreement (evaluated with Fleiss’s method) and the average time for the compilation of the banner were calculated. Results: Fourteen radiologists evaluated 312 chest radiographs of COVID-19 pneumonia suspected patients (80 males and 38 females) with an average age of 64, 47 years. The inter-rater agreement showed a Fleiss’ kappa value of 0.53 and the intra-group agreement varied from Fleiss’ Kappa value between 0.49 and 0.59, indicating a moderate agreement (considering as “moderate” ranges 0.4–0.6). The years of work experience were irrelevant. The average time for obtaining the result with the automatic software was between 7 s (e.g., zero COVID-19 score) and 21 s (e.g., with COVID-19 score from 6 to 12). Conclusion: The use of automatic software for the generation of a CXR “COVID-19 score” has proven to be simple, fast, and replicable. Implementing this tool with scores weighed on the number of lung pathological areas, a useful parameter for clinical monitoring could be available.

Screening, Diagnostic and Prognostic Tests for COVID-19: A Comprehensive Review

“…These findings demonstrate that chest imaging modalities are sensitive in identifying COVID-19 with lower respiratory tract involvement, while their specificity is limited [80]. Interestingly, inter-rater agreement on the interpretation of chest x-rays is moderate [81,82].…”

Improving diagnosis accuracy with an intelligent image retrieval system for lung pathologies detection: a features extractor approach

“…Conversely, Mehta 8 proposed a CBMIR system for sub-images in high-resolution digital pathology images, utilizing scale-invariant feature extraction. Lowe 9 utilized Scale-Invariant Feature Transform (SIFT) to index sub-images and reported an 80% accuracy rate for the top 5 retrieved images. Lowe’s experiments were conducted on 50 ImmunohHistoChemistry (IHC) stained pathology images at eight different resolutions.…”

Application of machine learning in CT images and X-rays of COVID-19 pneumonia

“…Therefore, training is more difficult and time-consuming. Oktay et al [ 66 ] proposed a U-Net that can capture fine structures in medical images, which is suitable for the segmentation of lesions and lung nodules in COVID-19 applications. Training a segmentation network requires sufficient labeled data.…”