About MIMIC-CXR classification

[Hunan-Tiger]

Hello, thank you for your work.

I am currently using 'densenet121-res224-mimic_ch' to evaluate the MIMIC-CXR test set, but I am encountering a category inconsistency issue.

The official MIMIC categories are [‘Atelectasis’, ‘Cardiomegaly’, ‘Consolidation’, ‘Edema’, ‘Enlarged Cardiomediastinum’, ‘Fracture’, ‘Lung Lesion’, ‘Lung Opacity’, ‘No Finding’, ‘Pleural Effusion’, ‘Pleural Other’, ‘Pneumonia’, 'Pneumothorax', ‘Support Devices’]. However, the model's results are missing the ‘No Finding’, ‘Pleural Effusion’, and ‘Pleural Other’ categories.

How can this be resolved? Or have the category names changed?

Thank you again.

[ieee8023]

The models were trained to test dataset shifts so the labels were aligned to a common list:

torchxrayvision/torchxrayvision/datasets.py

Line 22 in 6fa28a2

default_pathologies = [

Plural Effusion is renamed to just Effusion

torchxrayvision/torchxrayvision/datasets.py

Line 1193 in 6fa28a2

self.pathologies = list(np.char.replace(self.pathologies, "Pleural Effusion", "Effusion"))

No finding as and plural other were dropped. No finding was used to set the labels for the pathologies to false.

[Hunan-Tiger]

Thank you for your reply!