Fig 1.
Overview of the proposed method for our COVID-19 study.
Fig 2.
Flowchart of radiomic feature extraction.
Fig 3.
t-SNE dimensions of the class distributions using whole CXR images (left) and lung segments (right).
Table 1.
CXR data information.
Table 2.
Best classification metrics for each feature-level approach.
Table 3.
Classification accuracy obtained using handcrafted and radiomic features.
Table 4.
Classification accuracy obtained using all handcrafted and radiomic feature combinations.
Table 5.
Classification accuracy obtained using selected handcrafted (LBP + HOG + GLCM) and radiomic feature combinations.
Table 6.
Classification accuracy obtained using combined deep features (Pool5 of Resnet18 + Conv5 of Densenet121).
Table 7.
Classification accuracy obtained using all features.
Table 8.
Classification accuracy obtained using selected handcrafted, radiomic and deep features.
Table 9.
Comparison of the three-class classification studies from previous CXR-based COVID-19 studies.