Abstract
BACKGROUND: The risk factors for postoperative recurrence in pathological stage IA invasive lung adenocarcinoma (LUAD) remain unclear. This study aimed to evaluate the incremental prognostic value of deep learning (DL)-based quantitative parameters for pathological stage IA invasive LUAD after surgery. METHODS: The maximum total size on axial images (MTSA) and the maximum solid component size on axial images (MSSA) of tumors were manually measured, and consolidation tumor ratio was calculated as MSSA/MTSA. Maximal total size on multiplanar reconstructed images, total volume, solid component volume, and solid component volume ratio (SV%) were evaluated by DL software, with different densities as thresholds to separate solid component. In addition, the radiomics parameters including variance, skewness, kurtosis, entropy, and sphericity were extracted from DL software. Incorporating clinical and pathological characteristics with DL-based quantitative parameters, competing risk model was employed to identify independent predictors of recurrence, and three nested predictive models were constructed. The predictive performance was assessed using Harrell's concordance index, receiver operating characteristic curve, net reclassification improvement index, integrated discrimination improvement index, calibration curve, and decision curve analysis. RESULTS: A total of 2,117 patients with pathological stage IA invasive LUAD were included, of which 139 experienced recurrence and 41 died. The predictive performance of manual measurements was inferior to that of DL-based quantitative parameters. Among the DL-based quantitative parameters, SV% with 0 HU as the solid component threshold (SV(0) (HU)%) was most strongly associated with recurrence. Univariate and multivariate analyses identified pathological stage, histologic subtype, vascular or perineural invasion, spread through air space, SV(0) (HU)%, and entropy as independent predictors of recurrence. Among the three predictive models, the model incorporating SV(0) (HU)% and entropy demonstrated the best predictive performance. CONCLUSIONS: DL-based quantitative parameters are superior to manual measurements in predicting the recurrence of pathological stage IA invasive LUAD. Pathological stage, histologic subtype, vascular or perineural invasion, spread through air space, SV(0) (HU)%, and entropy are significant risk factors for recurrence. SV(0) (HU)% and entropy can provide incremental prognostic value for this population.