Skip to main content
. 2021 Jul 29;14(10):1435–1445. doi: 10.1016/j.jiph.2021.07.015

Table 7.

Evaluation of the proposed severity detection method with state-of-the-art methods.

Author Class Subject Method Performance metric Value
Changati et al. Lung infection percentage 9749 chest CT volume Deep learning, deep reinforcement learning Ref. [2] Pearson correlation coefficient 0.92 for percentage of opacity (P < 0.001)
Shen et al. Severe/non severe CT images from 44 patients Thresholding and adaptive region growing Ref. [26] Pearson correlation coefficient r ranged from 0.7679 to 0.837, P < 0.05
Xiao et al. Severe/non severe 23,812 Covid-19 images ResNet-34 Ref. [24] Precision AUC 81.3%
98.7%
Shan et al. Segment and quantify infection regions 549 CT volumes VBNet Ref. [27] Dice similarity coefficient 91.6%10.0
Pu et al. COVID-19 severity and progression 72 COVID-19 and 120 others volumes UNetBER Algorithm Ref. [25] Sensitivity
Specificity
95%84%
Tang et al. Severe/non severe Chest CT images of 176 patients Random forest algorithm Ref. [28] Accuracy 87.5%
Proposed method Covid 19 severity (High, Moderate, Low) COVID-CT data set (public) (349 images with metadata) ResNet-50 and Densenet-201 algorithm Sensitivity
SpecificityPrecision
Accuracy
95.23%97.41%95.3%97.84%