RSNA 2022 Spine Fracture Detection - Metadata

This dataset contains metadata extracted from train image dicom files relevant to the RSNA 2022 Cervical Spine Fracture Detection competition.

1. meta-train - original metadata extracted (ignore this file)
2. meta-train-clean - cleaned version of meta-train (easier to use)
3. meta-segmentations - meta-data for images with segmentations (including correct labels C1-C7 extracted from unique values in segmentations)
4. meta-segmentation-clean - cleaned version of meta-segmentations.
5. meta-train-with-vertebrae - meta-data for all train images (with 88% accurate RF predictions of which vertebrae is in each image)
6. train-segmented - meta-data for all train images (with 95% accurate EffNetV2 predictions of which vertebrae is in each image from this notebook)
7. train-vert-fold4 - this is similar to 6. train-segmented, made by cleaning the segmentations and training an image+tabular model. Also includes additional columns created from feature engineering.
8. train-vert - ensembled predictions of train-segmented and train-vert-fold4.

The notebooks used to create these files are below:

• RSNA Fracture Detection - in-depth EDA
• Extracting Vertebrae C1, ..., C7