Regarding MIMIC-ABN

[njan-creative]

I am looking at MIMIC-ABN shared at
https://huggingface.co/MK-runner/MLRG/tree/main/radiology%20report/mlrg_mimic_abn_annotation_v1.1.json

The file at link below also shows 844 images and it matches the MIMIC-ABN shared in the dataset.
https://github.com/mk-runner/MLRG/blob/main/generated-radiology-reports/MIMIC-ABN/test_reports_epoch-1_23-10-2024_10-25-20.csv

I observed that the distribution of view position of anchor image is below for 844 images
Anchor View
[PA] 430
[LATERAL] 412
[AP] 2

For anchor images that has lateral image as anchor image.
Did you use the lateral image only for generating the reports or other images were also used.

The original MIMIC-ABN dataset is not accessible now.

[mk-runner]

Thank you for your question regarding the use of anchor images and auxiliary references in the MIMIC-ABN dataset.

For cases where the anchor scan is a lateral image, if the auxiliary references are not empty, we do include them during report generation, regardless of their view positions (e.g., PA, AP, etc.).

Best regards,
Kang Liu

[njan-creative]

Thank you for the response.

I am looking for MIMIC-ABN dataset and x-rays belonging to MIMIC-ABN dataset.
The earlier links for MIMIC-ABN dataset are not working.

Can you post the list of images in MIMIC-ABN dataset.
I am looking for evaluation on MIMIC-ABN dataset.

[mk-runner]

In terms of the MIMIC-ABN dataset, all images in MIMIC-ABN originate from the MIMIC-CXR dataset. Compared to MIMIC-CXR, there are two key differences:

1. Subset of Data: MIMIC-ABN is a subset of MIMIC-CXR. As a result, the number of samples in MIMIC-ABN is smaller than in the full MIMIC-CXR dataset.
2. Only Abnormal Findings: The radiology reports in MIMIC-ABN contain only descriptions of abnormalities, whereas MIMIC-CXR includes full-length reports.

If you would like to reconstruct the MIMIC-ABN dataset, we recommend the following steps:

1. Download the MIMIC-ABN annotation file from Hugging Face: mlrg_mimic_abn_annotation_v1.1.json
2. For each sample, retrieve the corresponding X-ray image from the MIMIC-CXR dataset using the path specified in sample['anchor_scan']['image_path'].

Please note that due to copyright restrictions and the large size of the MIMIC-CXR dataset, we are unfortunately unable to provide the corresponding X-ray images directly.

Feel free to reach out if you have any further questions.

Best regards,
Kang Liu

[kailasdayanandan]

Thanks a lot for the response.