TrackRAD2025: Sagittal Cine-MRI Frames for Cardiac Motion Tracking

TrackRAD2025 provides over 2.8 million unlabeled and over 10,000 labeled sagittal cine-MRI frames for a cardiac motion tracking challenge. The dataset was created by LMUK-RADONC-PHYS-RES and includes data from 477 unlabeled and 108 manually labeled patients. It was last updated on the Hugging Face platform in January 2026.

Use Cases

• Train semi-supervised learning models based on the large volume of unlabeled cine-MRI frames.
• Develop and benchmark cardiac structure segmentation algorithms based on the manually labeled frames.
• Study inter-observer variability in medical image annotation based on the subset of frames labeled by multiple observers.
• Create models for tracking anatomical motion over time based on the time-resolved 2D image sequences.

Strengths

• Large scale of unlabeled data with over 2.8 million frames from 477 patients.
• Includes a substantial labeled subset with over 10,000 frames, including a portion with multiple observer annotations.
• Focuses on a specific, clinically relevant modality: 2D sagittal cine MRI.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the labeled subset is not explicitly broken down beyond the total frame count.
• The temporal and geographic coverage of the patient cohort is not specified.

Provenance

Source

LMUK-RADONC-PHYS-RES

Collection Method

Acquired 2D sagittal cine MRI data from patients, with a subset manually labeled, potentially by multiple observers.

Freshness

Last updated 2026-01-10 09:46:59; freshness should be verified.