The role of artificial intelligence technology analysis of high-resolution computed tomography images in predicting the severity of COVID-19 pneumonia.

Abstract

High-resolution computed tomography (HRCT) is usually used only for qualitative analysis of COVID-19 pneumonia. However, when coupled with artificial intelligence (AI) it can also automatically provide quantitative data. The purpose of the study was to analyze the role of automatic assessment of COVID‑19 pneumonia severity on HRCT images by AI technology. We retrospectively studied medical records of consecutive patients admitted to the Krakow University Hospital due to COVID‑19. Of the 1729 patients, 804 underwent HRCT with automatic analysis of such radiological parameters as absolute inflammation volume, absolute ground glass volume, absolute consolidation volume (ACV), percentage inflammation volume, percentage ground glass volume, percentage consolidation volume (PCV), and severity of pneumonia classified as none, mild, moderate, or critical. The automatically assessed radiological parameters correlated with the clinical parameters that reflected the severity of pneumonia (P <0.05). The patients with critical pneumonia, as compared with mild or moderate one, were more frequently men, had significantly lower oxygen saturation, higher respiratory rate, higher levels of inflammatory markers, as well as more common need for mechanical ventilation and admission to the intensive care unit. They were also more likely to die during hospitalization. Notably, as determined by the receiver operating characteristic curve analysis, radiological parameters above or equal to the cutoff points were independently associated with in‑hospital mortality (ACV odds ratio [OR], 4.08; 95% CI, 2.62-6.35; PCV OR, 4.05; 95% CI, 2.60-6.30). Using AI to analyze HRCT images is a simple and valuable approach to predict the severity of COVID‑19 pneumonia.