🦵 Knee MRI Dataset for ACL Tear Detection

📝 Description

The KneeMRI dataset was gathered retrospectively from exam records performed on a Siemens Avanto 1.5T MRI scanner using a proton density-weighted fat suppression technique. Scans were conducted at the Clinical Hospital Centre Rijeka, Croatia, between 2006 and 2014.

The dataset includes:

• 917 volumes of either left or right knees (12-bit grayscale)
• Each volume is labeled with the condition of the anterior cruciate ligament (ACL):
• Healthy
• Partially injured
• Completely ruptured

Each sample includes a manually extracted Region of Interest (ROI) and diagnosis assigned in a double-blind fashion.

📚 For more information, see the paper:
I. Štajduhar, M. Mamula, D. Miletić, G. Unal, "Semi-automated detection of anterior cruciate ligament injury from MRI", Computer Methods and Programs in Biomedicine, Volume 140, 2017, Pages 151–164.

🦴 ACL Tears

ACL tears are among the most common knee injuries in high-performance sports such as basketball and soccer. They may involve:

• Stretching
• Partial tearing
• Complete rupture (most common)

Symptoms:
Pain, popping sound, instability, joint swelling.

🧠 ~200,000 ACL tears and over 100,000 reconstruction surgeries occur annually in the United States.

🧠 Magnetic Resonance Imaging (MRI)

MRI is a non-invasive radiological imaging method used for diagnosis, treatment monitoring, and structural analysis of:
- Joint injuries (like ACL tears)
- Brain trauma
- Heart disease
- Vascular issues

🧐 Key Notes on the Dataset

• Slices vary significantly between and within planes.
• Some slices are better suited to detect ACL tears than others.
• MRIs are volume-based (stack of slices), and not all volumes have the same number of slices → batching isn't straightforward.
• Future work includes plane-wise classification models and ensemble methods.

📂 File Structure

• README: Overview of the dataset
• example.py: Script showing how to read the dataset
• metadata.csv: CSV file containing metadata and ACL diagnosis
• volumetric_data/: Contains compressed volumes (.7z) split across 10 files
• vol01.7z – vol10.7z: Each contains a subset of MRI volumes
• example.pck: Small pickle file to preview the structure

🧾 Metadata Columns

📌 Citation & Source

If you use this dataset in your work, please cite:

Clinical Hospital Centre Rijeka, Croatia
I. Štajduhar et al. (2017), Semi-automated detection of anterior cruciate ligament injury from MRI, Computer Methods and Programs in Biomedicine, 140, pp. 151–164.