Interstitial lung disease pattern segmentation on HRCT
Standardized segmentation of interstitial patterns (reticulation, honeycombing, ground-glass, consolidation, emphysema, cysts) on high-resolution chest CT.
A medical AI team building an algorithm for diffuse interstitial lung diseases (DILD) needed radiologists to segment interstitial abnormalities on HRCT following a standardized nomenclature, with strict acquisition criteria (thin slices, ≥512×512, inspiration phase, no respiratory artifacts).
Produce homogeneous segmentations of the interstitial patterns of interest — reticulation, honeycombing, bronchiectasis, ground-glass, consolidation, emphysema and pulmonary cysts — with the segmentation margin and quality criteria defined in the client's guideline.
Radiologists experienced in thoracic imaging were trained on the client's segmentation criteria per pattern, with reference figures for margin behavior and extreme cases. Cases not meeting the acquisition or quality criteria were flagged and excluded rather than force-segmented.
The client received a normalized dataset with harmonized terminology and margins across annotators, ready to feed a segmentation model for ILD without additional label reconciliation.