Positive and negative cohort definitions
Define acute positive pulmonary embolism, chronic findings, equivocal studies and technically limited examinations. Negative controls should come from the intended clinical pathway rather than an unrelated healthy population, with duplicate and repeated-patient rules stated explicitly.
CTPA protocol and paired imaging
Contrast timing, pulmonary-artery opacification, motion, slice thickness and reconstruction affect task fitness. If paired non-contrast CT is required, define the acceptable time window and ensure the pair belongs to the correct clinical episode.
Ground-truth hierarchy
A defensible hierarchy can combine final radiology interpretation, expert re-review, adjudication and relevant clinical follow-up. The specification should distinguish study-level PE labels from embolus localization or segmentation requirements.
Diversity, metadata and QC
Report scanner vendor, model, reconstruction, protocol, site and geography distributions. The release manifest should record reports, acquisition metadata, completeness, quality exclusions and label provenance rather than reducing the cohort to a scan count.
External validation and permitted use
External-validation cohorts should remain independent from development patients and, where required, sites or scanners. Commercial use, programme-specific authorization, privacy and de-identification controls must be confirmed before an approved release.
Questions, answered directly.
Does MedCorpora claim a ready CTPA cohort?+
No. This guide defines a sourcing specification; eligible inventory, sites and timing require feasibility.
Should technically limited studies be excluded?+
Not automatically. They may be exclusions, a separate class or important failure cases depending on the intended use.
Can the cohort support external validation?+
Potentially, if patient and source independence, ground truth and acquisition coverage meet the defined protocol.