# Reproducing imagesTr (the original images)

The labels in `labelsTr_26classes_in_orig_space/` are aligned to the images in
`imagesTr/`, produced from the official RSNA 2025 Intracranial Aneurysm
Detection Challenge DICOMs. This document describes the exact pipeline.

## Pipeline

1. **Download** the RSNA 2025 challenge dataset (DICOM series + `train.csv`) from Kaggle.
2. **Organize** DICOMs into one directory per `SeriesInstanceUID`.
3. **Convert** with [`process_RSNA2025_all_data.py`](https://github.com/PengchengShi1220/RSNA2025_Intracranial-Aneurysm-Detection):
   - `dicom2nifti.dicom_series_to_nifti(..., reorient_nifti=False)`
   - nibabel reorientation to **LPS**
   - output: `imagesTr/<SeriesInstanceUID>_0000.nii.gz`
4. **Match**: label `<uid>.nii.gz` ↔ image `imagesTr/<uid>_0000.nii.gz` (same UID).

## Environment

- `dicom2nifti` 2.6.2, `nibabel`, `pydicom`, `pandas`, `numpy`