A multistage CT-based artificial intelligence system for microvessel density quantification in lung adenocarcinoma: correlation with histopathology

Microvessel density (MVD), assessed via histopathology, is a key indicator of angiogenesis in lung adenocarcinoma (ADC). This study aimed to develop and validate an artificial intelligence (AI) system based on computed tomography (CT) images for the noninvasive preoperative assessment of MVD in ADC. This retrospective study included 152 surgically resected ADC patients from center 1(internal cohort) and center 2 (external cohort). A multistage AI system was developed to automatically segment ADC lesions and intratumoral vessels on unenhanced CT (UECT) and contrast-enhanced CT (CECT), and calculate CT-derived MVD (CT-MVD). CD34 immunostaining-derived MVD (CD34-MVD), including microvessel area (MVA, the microvessel area proportion) and microvessel count (MVC, the number of microvessels per unit area), was used as the pathological reference standard. Correlation, Bland–Altman, and intraclass correlation coefficient (ICC) analyses were performed to evaluate the relationship between CT-MVD and CD34-MVD. The AI models showed good segmentation performance, with Dice coefficients of 0.869 ± 0.053 for ADC lesions and 0.817–0.884 for vessels. In the internal and external cohorts, CT-MVD showed a highly significant positive correlation with MVA ( r = 0.665–0.859 and r = 0.652–0.846, respectively; all P < 0.05) and a moderate positive correlation with MVC ( r = 0.339–0.591 and r = 0.352–0.577, respectively; all P < 0.05), with the strongest correlations observed in the venous phase. Bland–Altman and ICC analyses further showed good agreement and consistency between venous-phase CT-MVD and MVA, with 95% limits of agreement of − 1.079 to 1.079 and − 1.098 to 1.098, and ICC values of 0.850 and 0.846 in the internal and external cohorts, respectively. AI-derived CT-MVD was correlated with CD34-MVD in ADC, particularly in the venous phase, suggesting its potential as a noninvasive imaging parameter for assessing tumor angiogenesis.