Table 2 Quantitative evaluation of the COVID-19 lesion subtype segmentation algorithm.

From: Deep learning-based lesion subtyping and prediction of clinical outcomes in COVID-19 pneumonia using chest CT

 Parenchyma subtype

Dice coefficient

Average surface distance (mm)

Hausdorff distance (mm)

Healthy tissue

0.985 (0.021)

0.106 (0.113)

1.143 (1.679)

Ground glass opacities

0.912 (0.158)

0.301 (0.381)

3.019 (5.542)

Consolidation

0.841 (0.254)

6.265 (26.813)

16.997 (44.453)

  1. The table reports the average for each parenchymal pattern. Data in parenthesis are standard deviations. Both manual segmentations (radiologist and pulmonary imaging expert) were considered.
Back to article page