Pixel-based acquisition condition classification · No DICOM metadata required
Sensitivity values: 154-case LIDC-IDRI perturbation study (arXiv:2603.26785)
"Detection sensitivity for pulmonary nodules ≥3mm is within established benchmarks for this study."
"Detection sensitivity for pulmonary nodules ≥3mm is within established benchmarks for this study."
"Detection of pulmonary nodules <6mm is limited under current acquisition parameters. Small nodules cannot be reliably excluded."
"Given acquisition parameters, detection sensitivity for pulmonary nodules in the 3–6mm range is reduced. Small nodules may not be reliably excluded."
"Given acquisition parameters, detection sensitivity for pulmonary nodules in the 3–6mm range is reduced. Small nodules may not be reliably excluded."
Sensitivity depends not just on reconstruction interval size — but on where the nodule falls within that interval. A nodule straddling two slices is harder to detect than one centered on a slice, even when the AI reports identical confidence.
Left: overall sensitivity by reconstruction interval. Right: sensitivity as a function of nodule z-phase — where the nodule falls within the reconstruction cycle. Phase 0 = nodule centered on a slice; phase 0.5 = nodule straddling two slices (worst case). AI confidence score does not change. Detectability does.