AI reads CT scans to predict which lung nodules turn cancerous

Copied word for word from the study's registry entry, so the wording is the study team's rather than ours.

Inclusion Criteria: * Age ≥ 18 years. * Persistent pulmonary ground-glass nodule (pGGN or mGGN, 5-30 mm) on thin-slice chest CT (slice thickness ≤ 1.5 mm). * Baseline and follow-up thin-slice chest CTs of sufficient quality for 3D segmentation and registration. * Minimum interval between any two consecutive CTs \> 1 month. * Complete baseline clinical data available (age, sex, smoking history, family history of malignancy, relevant comorbidities). Cohort-specific inclusion * Group 1 (Development): surgical resection of the target GGN at PKUPH between Jan 2007 - Jun 2025, with ≥ 2 pre-operative thin-slice CTs available. * Group 2 (Surgical internal test): surgical resection at PKUPH between Jul 2025 - Jan 2026, with ≥ 2 pre-operative thin-slice CTs available. * Group 3 (Non-surgical internal test): non-operative management at PKUPH between Jan 2020 - Dec 2025, with ≥ 3 thin-slice CTs of the target GGN available. * Group 4 (Prospective external validation): prospective enrollment after model lock at participating centers, baseline CT plus ≥ 2 planned routine follow-up thin-slice CTs. Exclusion Criteria: * Coexisting severe pulmonary disease that obscures evaluation of the target GGN (e.g., active pulmonary tuberculosis, severe interstitial lung disease). * Prior history of any other thoracic malignancy, or active extrathoracic malignancy under treatment within 5 years, that would confound interpretation of the target GGN. * CT image quality insufficient for registration and feature extraction (severe motion artifact, slice thickness \> 1.5 mm at any required timepoint, or extensive metallic artifact projecting over the target GGN). * Pure solid nodule with no ground-glass component. * Target GGN already received treatment (resection, ablation, or radiotherapy) prior to the baseline CT used in this study.