Preoperative CT-based artificial intelligence-derived quantitative parameters and imaging features for predicting the invasiveness of histologically confirmed subcentimeter adenocarcinomatous nodules: a two-center study

Le résumé fourni par la source

Background: Assessing the invasiveness of subcentimeter pulmonary adenocarcinomas remains challenging. Computed tomography (CT) provides objective data, but radiologic interpretation is subjective. Artificial intelligence (AI)-derived quantitative parameters may offer a more reproducible assessment for small nodules. This study aimed to develop and validate models based on CT-derived AI quantitative parameters nodules and imaging features to predict the invasiveness of subcentimeter adenocarcinomatous nodules. Methods: Patients diagnosed with subcentimeter adenocarcinomatous nodules from two centers between January 2021 and December 2022 were retrospectively included, and their associated quantitative AI parameters and CT radiologic features were recorded and analyzed. Independent predictors associated with the invasiveness of subcentimeter adenocarcinomatous nodules were identified through univariate and multivariate logistic regression analyses. A qualitative model based on CT radiologic features (density, irregular morphology, and leaflet), a quantitative model based on AI-derived parameters (longest, entropy, and mass), and a combined model integrating both were subsequently constructed. The performance of the models was evaluated by calculating the area under the curve (AUC), sensitivity, specificity, positive predictive value, negative predictive value, accuracy and F-measure. Results: A total of 337 patients from Shanghai Changzheng Hospital and Kunshan Third People's Hospital were included in this study. Among the qualitative features, density, irregular morphology, and leaflet were identified as independent predictors of invasive lung cancer, and a qualitative model was constructed. Among the quantitative parameters, longest diameter, entropy, and mass were identified as independent predictors, and a quantitative model was developed accordingly. In the training set, the areas under the curve of the qualitative, quantitative, and combined models were 0.818, 0.849, and 0.901, respectively. Conclusions: In this cohort, models combining quantitative CT AI parameters with CT radiologic features demonstrated high performance in predicting the invasiveness of subcentimeter adenocarcinomatous nodules.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.