Grading COVID-19 Severity with CT: CAD and AI

The pandemic of COVID-19 has presented an unprecedented challenge to modern global healthcare. Today everybody understands that diagnostic imaging has critical importance in detection, grading, and follow-up of this potentially fatal disease. Computed tomography (CT) has proven itself as the best imaging modality for the detection of the most important and typical imaging patterns of lung parenchyma changes in COVID-19 pneumonia.

Guidelines of the most influential national and international radiological societies recommend against the wide use of chest CT for screening, diagnosis, and surveillance of COVID-19 patients. They advise to use it for the differential diagnosis, detection of complications, and cases when results of CT may influence patient management and prognosis.

On the other hand, the typical patterns of COVID-19 pneumonia are well known (multiple peripheral ground-glass opacities (GGO) with or without areas of consolidation and some other manifestations). In cases of COVID-19, and the high occurrence of the disease, CT has a high sensitivity for the detection of patients with COVID-19 pneumonia. For this reason, in most hospitals in Russia, the standard approach is to perform chest CT for all patients admitted with suspected COVID-19 pneumonia. Structured reports for such patients are very important. The common practice is to use reporting guidelines from RSNA/ACR/STI defining imaging appearances as typical, indeterminate, or atypical for COVID-19.

A less clear situation exists in the grading of disease severity. There are recommendations to use simple subjective scales of the severity of changes in the lungs such as mild-moderate-severe or mild-moderate-severe-critical. Several more complicated grading scales have been proposed. But practical experience shows that the visual scoring of multiple areas of GGO or opacifications in the lungs is very subjective and varies from one radiologist to another. It looks like, for this purpose, the software for computer-aided diagnosis or artificial intelligence (AI) capabilities to detect, delineate, and analyse GGO/consolidations in the lungs allow for much more accurate and reproducible results. These technologies have proven themselves very useful in follow-up studies. Today there are plenty of Computer Aided Detection or AI solutions for COVID-19 from different manufacturers and vendors. Probably they can give different results from the same datasets. But in the setting of the COVID-19 pandemic, these differences are not so significant, and I believe that they are rather small. They are not time-consuming and help the radiologist to be more objective in grading the disease severity and its progression or regression.

I believe that the epidemy of COVID-19 will accelerate the development and performance of such computer-based assistance for radiologists. Besides the obvious practical value, they are going to be the very useful tools for research of COVID-19 outcomes, patient prognosis, and efficacy of the drugs and other types of treatment in upcoming clinical trials on COVID-19.