ACDC Cardiac Structure Segmentation Results with 2.5% Labeled Data

Segmentation performance metrics for the right ventricle, myocardium, and left ventricle from the Automated Cardiac Diagnosis Challenge (ACDC) dataset. Results are based on a limited training set of 2.5% labeled data and reported as mean and standard deviation across three experimental runs. The data was contributed by Yafei Mi and last updated in April 2026.

Use Cases

• Benchmark semi-supervised segmentation models by comparing performance metrics (mean ± standard deviation) for the RV, Myo, and LV structures.
• Analyze the variance in model performance across three runs to assess the stability of training with minimal labeled data.
• Use the reported metrics for the right ventricle (RV), myocardium (Myo), and left ventricle (LV) as a baseline for new methods on the ACDC dataset.

Strengths

• Performance metrics are aggregated over 3 runs, providing a measure of variance.
• Clear focus on a specific, challenging machine learning scenario using only 2.5% of labeled ACDC data.
• Data is shared under a permissive CC-BY-4.0 license for open reuse.

Limitations

• The dataset is extremely small at 5.5 KB, containing only summary statistics, not raw images or segmentations.
• No individual patient data, image slices, or model predictions are included, only aggregated results.
• The underlying ACDC data's temporal and geographic coverage for the original cardiac MRI scans is not specified.

Provenance

Source

Results derived from the Automated Cardiac Diagnosis Challenge (ACDC) dataset.

Collection Method

Machine learning experiment results (segmentation performance metrics) reported after model training.

Freshness

Last updated in April 2026.