holajoa / Adaptor-VL-SSL

Freeze the Backbone: A Parameter-Efficient Contrastive Approach to Robust Medical Vision-Language Pre-training. Thesis of MSc AI degree at Imperial College London.
2 stars 1 forks source link

request for mimic_cxr_sectioned.csv #14

Open YJSYJSYJS opened 4 days ago

YJSYJSYJS commented 4 days ago

Thank you for your great work!

I'm reproducing your work with your repo. I downloaded all of the datasets including mimic-cxr-jpg. However i couldn't find mimic_cxr_sectioned.csv, which is required in preprocess/mimic_cxr.py. It seems like the csv file is a combined version of text files from mimic-cxr-jpg 2.0.0. Can you share your csv file or the way you made it?

holajoa commented 3 days ago

Thanks for your message! mimic_cxr_sectioned.csv is the officially recommended train/validation/test splits of the MIMIC-CXR-JPG dataset, and is included in the data files. Please find further information and download instructions here.

YJSYJSYJS commented 3 days ago

Thanks for your reply!

I read the instruction about mimic-cxr data at the website you linked. I think mimic-cxr-2.0.0-split.csv is just what you described.

mimic_cxr_sectioned.csv is read by pandas dataframe with the name of text_df in mimic cxr.py. So I thought that it might be the combination of text raw data composed of findings and impressions mainly