Closed AlexTS1980 closed 3 years ago
That's correct - in the original CNCB dataset, they provide a file called lesions_slices.csv
which lists CT slices for which radiologists have observed lung lesions due to CP/NCP. We use only the slices listed in this file when constructing the COVIDx-CT dataset, as we cannot guarantee that other slices from CP/NCP volumes contain abnormalities. For normal control cases, we use all slices.
I don't fully understand the labelling of the data. The paper says 'For NCP and CP CT volumes, slices marked as containing lung abnormalities were leveraged' - does it mean you only added scans that clearly display symptoms of (N)CP, leaving out slices without these symptoms?