COVID-19 disease: CT Pneumonia Analysis prototype by using artificial intelligence, predicting the disease severity

Abstract

Background Since the beginning of 2020, coronavirus disease has spread widely all over the world and this required rapid adequate management; therefore, continuous searching for rapid and sensitive CT chest techniques was needed to give a hand for the clinician. We aimed to assess the validity of computed tomography (CT) quantitative and qualitative analysis in COVID-19 pneumonia and how it can predict the disease severity on admission. Results One hundred and twenty patients were enrolled in our study, 98 (81.7%) of them were males, and 22 (18.3%) of them were females with a mean age of 52.63 ± 12.79 years old, ranging from 28 to 83 years. Groups B and C showed significantly increased number of involved lung segments and lobes, frequencies of consolidation, crazy-paving pattern, and air bronchogram. The total lung severity score and the total score for crazy-paving and consolidation are used as severity indicators in the qualitative method and could differentiate between groups B and C and group A (90.9% sensitivity, 87.5% specificity, and 93.2% sensitivity, 87.5% specificity, respectively), while the quantitative indicators could differentiate these three groups. Using the quantitative CT indicators, the validity to differentiate different groups showed 84.1% sensitivity and 81.2% specificity for the opacity score, and 90.9% sensitivity and 81.2% specificity for the percentage of high opacity. Conclusion Advances in CT COVID-19 pneumonia assessment provide an accurate and rapid tool for severity assessment, helping for decision-making notably for the critical cases.

A Quantitative Lung Computed Tomography Image Feature for Multi-Center Severity Assessment of COVID-19

“…Assessment of 3-class severity (mild, moderate, severe) is crucial to determine the treatment route [14] and is well summarized in the WHO interim guidance on clinical management of COVID-19 2 . Some recent literature have shown chest CT to determine COVID-19 severity qualitatively and quantitatively with good correlation with clinical parameters [15][16][17]. Further, recent studies have also demonstrated that quantification of lung CT based severity can predict the short-term prognosis of COVID-19 [18,19].…”

“…Sun et al using commercial software quantified a number of CT features including the percent lesion cover and the type of deposition to classify 84 patients into severe and non-severe categories with about 91% specificity and sensitivity [15]. On the other hand, Gouda and Yasin quantified CT features of 120 patients including percentage of high and low opacity and total lesion cover using a CT pneumonia analysis algorithm by Siemens Healthineers [16]. With the aid of AI-rad —an artificial intelligence based lesion detection tool they could classify mild and non-mild cases with a sensitivity of up to 90%.…”

Quantitative evaluation of COVID-19 pneumonia severity by CT pneumonia analysis algorithm using deep learning technology and blood test results

“…The % opacity was also negatively correlated with SpO 2 . In a study using the same CT Pneumonia Analysis software, the opacity score, % opacity, volume of opacity, volume of high opacity, % high opacity and mean HU total were significantly higher in the moderate and severe groups compared to the mild group, while the total lung volume was significantly lower in the severe group compared to the mild group [14]. The results of our study also showed significant differences in % opacity and % of high opacity in univariate analysis, which is consistent with their results.…”

“…The positive rate for CT imaging for the diagnosis of suspected COVID-19 patients was 88% vs. 59% by RT-PCR, and the sensitivity of CT increased to 97% based on positive RT-PCR results, confirming the role of CT as a primary tool for diagnosis [6]. Several studies have shown the ability of visual quantitative evaluation of CT images to predict mortality and severity with high consistency [7][8][9][10], and these scoring techniques are now applied to the risk prediction and severity evaluation of COVID-19 pneumonia using artificial intelligence (AI) [11][12][13][14]. In these AI-based approaches, the use of routine blood test results is still limited to a few studies.…”

“…Reported cutoff values for the mean HU total to differentiate mild from moderate and severe cases were − 637.7 (AUC, 0.876; 81.8% sensitivity and 81.9% specificity) [14] and − 816 (AUC, 0.87; 91% sensitivity and 90% specificity) [23]. In these studies, mild pneumonia included patients with respiratory symptoms and evidence of pneumonia on CT.…”