Skip to content

data.md

Latest commit

 

History

History
79 lines (59 loc) · 3.11 KB

data.md

File metadata and controls

79 lines (59 loc) · 3.11 KB
Raw

Datasets

Knee MRI datasets for self-supervised pretraining and downstream evaluation.

Directory Structure

After preprocessing, volumes are stored as compressed NIfTI (.nii.gz) with shape (W, H, D) following the TorchIO convention. All datasets share the same layout:

<DatasetName>/
├── <split>/[<sequence>/]<plane>/<uid>.nii.gz
└── <split>.csv

The optional <sequence>/ subfolder is present when a dataset contains multiple MRI sequences (e.g., pd/, t2_fs/ for fastMRI; DESS_E1/, DESS_E2/ for SKM-TEA). Single-sequence datasets (MRNet, kneeMRI, KMAR-50K) omit this level.

Datasets Overview

Dataset Modality Sequence Plane Size Labels Download Source
MRNet Knee MRI PD / T2 (varies) Sagittal, Coronal, Axial 1,130 / 120 (train/test) exams Abnormal, ACL tear, Meniscal tear Stanford ML Group
fastMRI Knee Knee MRI PD, PDFS, T1, T2, T1PRE, T1POST, ... Sagittal, Coronal, Axial ~10,000 exams (~164 GB) None (unlabeled) NYU fastMRI
KMAR-50K Knee MRI Various (clinical routine) Sagittal, Coronal, Axial ~50,000 volumes Available but unused Zenodo
SKM-TEA Knee qDESS MRI DESS Echo 1 (T1w, TE≈6ms), DESS Echo 2 (T2w, TE≈34ms) Sagittal 86 / 43 / 26 (train/val/test) exams Meniscal tear, Ligament tear, Cartilage lesion, Effusion Stanford AIMI
kneeMRI Knee MRI Sagittal PD Sagittal 817 / 100 (train/test) volumes ACL tear (binary & multiclass) Zenodo

MRNet, fastMRI Knee, and KMAR-50K are used for self-supervised pretraining. SKM-TEA and kneeMRI are used for downstream evaluation.

Preprocessing

1. MRNet.

python scripts/preprocess_dataset/preprocess_MRNet.py \
    --data-dir /path/to/MRNet/MRNet-v1.0 \
    --save-dir /path/to/datasets/preprocessed/MRNet \
    --workers 8

2. fastMRI Knee

Preprocessing converts DICOM series to NIfTI, groups by patient/exam/series, and determines the acquisition plane and MRI sequence from DICOM metadata:

python scripts/preprocess_dataset/preprocess_fastMRI.py \
    --data-dir /path/to/fastMRI/knee \
    --save-dir /path/to/datasets/preprocessed/fastMRI \
    --split train \
    --workers 32

3. KMAR-50K

python scripts/preprocess_dataset/preprocess_KMAR-50K.py \
    --data-dir /path/to/KMAR-50K \
    --save-dir /path/to/datasets/preprocessed/KMAR-50K \
    --workers 8

4. SKM-TEA

python scripts/preprocess_dataset/preprocess_SKM-TEA.py \
    --data-dir /path/to/SKM-TEA/qdess/v1-release \
    --save-dir /path/to/datasets/preprocessed/SKM-TEA \
    --workers 8

5. kneeMRI

python scripts/preprocess_dataset/preprocess_kneeMRI.py \
    --data-dir /path/to/kneeMRI \
    --save-dir /path/to/datasets/preprocessed/kneeMRI \
    --split test \
    --plane sagittal \
    --workers 8