Paired cardiac CT for coronary and aortic valve calcium scoring, up to 681 exams
OpenOverview
Paired non-contrast cardiac CT for validation of calcium-scoring AI. Composition: - 256 paired patients, each with an ECG-gated calcium-scoring NCCT and a matching non-gated NCCT (512 exams) - 104 unpaired non-gated CAC scans - 65 unpaired non-gated AVC scans - Maximum around 681 exams, roughly 628 unique patients if about 53 scans serve both CAC and AVC Pairing: gated and non-gated NCCT on the same patient with acquisition dates less than 365 days apart. De-identification must preserve that interval through consistent date shifting. Acquisition: gated vs non-gated verified from DICOM metadata and protocol, non-contrast confirmed, adequate chest and heart coverage, original diagnostic-quality DICOM. Dose classified as low or regular prospectively from protocol name plus CTDIvol, DLP and tube current, not from free text. Scanner manufacturer, model, software version and reconstruction kernel taken from the selected non-gated series, with Toshiba and Canon normalised. Labels: CAC Agatston taken from the paired gated report or structured report; where absent, blinded expert rescoring is required rather than estimating from narrative text. AVC needs a numeric Agatston plus sex-specific severity threshold, and expert retrospective scoring should be assumed since reports usually describe valve calcification without a number. Predefined reader qualifications, blinded reads, documented disagreement resolution and a full audit trail from label to source. Provenance: unique patients, tokenised IDs, no overlap with any training or tuning set, site-level provenance documented. Observed stent prevalence should be reported, not filtered or enriched.
Progress
Data Specifications
| Category | Medical imaging |
|---|---|
| Required quantity | 681 |
| Data types | Medical imaging, CT, Cardiac, DICOM |
Use Cases
- Training and validating Medical imaging AI/ML models
- Benchmarking Medical imaging detection and segmentation algorithms
- Building de-identified Medical imaging research datasets for academic studies
- Augmenting existing Medical imaging datasets to reduce class imbalance