By looking at the train_mimic.json created using the make-interpret-mimic-cxr.py script, I found that there are a lot of empty reports where the findings and impression sections are empty. However, when looking at the unprocessed reports from the original MIMIC dataset, you can see that there is either findings or impression.
Is this on purpose? Can we try to replace the missing data? I'm worried that using the current data could teach the models to produce empty reports
Hi!,
By looking at the train_mimic.json created using the make-interpret-mimic-cxr.py script, I found that there are a lot of empty reports where the findings and impression sections are empty. However, when looking at the unprocessed reports from the original MIMIC dataset, you can see that there is either findings or impression.
Is this on purpose? Can we try to replace the missing data? I'm worried that using the current data could teach the models to produce empty reports
Thank you so much!