Skip to main content
. 2021 May 21;118:102114. doi: 10.1016/j.artmed.2021.102114

Table 8.

Per-class accuracy for lesion categorization between manual readings of expert radiologists and the AI model without lung segmentation and AI model with segmentation. Values in parentheses indicate 95% confidence intervals (CI).

Model no_segm Model w_segm
Consolidation 77.8% (69.9–84.1) 97.9% (93.6–99.8)
Ground glass 18.6% (14.1–24.1) 41.3% (35.1–47.7)
Crazy paving 57.1% (49.4–64.4) 98.3% (94.8–99.8)
Negative 99.3% (98.6–99.7) 99.9% (99.5–100)
Average 63.2% 84.4%