AI-assisted clinico-quantitative imaging nomogram for preoperative malignancy risk in solid and part-solid pulmonary...

To develop and internally validate an artificial intelligence (AI)-assisted clinico-quantitative imaging prediction model that combines automatically extracted quantitative imaging features with clinical data to preoperatively assess individualized malignancy risk in solid and part-solid pulmonary nodules (PNs) measuring ≤ 3 cm.

This retrospective study analyzed data from 951 consecutive patients who underwent surgical resection for PNs ≤ 3 cm (210 benign, 741 malignant). Quantitative CT features, including nodule size, minimum and maximum computed tomography attenuation, consolidation-to-tumor ratio, and nodule type, were automatically measured using InferRead CT Lung AI software (Infervision, Beijing, China; version 4.0). These AI-assisted quantitative measurements were evaluated together with clinical variables and inflammatory markers as candidate predictors. The final prediction model was a multivariable logistic regression model, with predictor selection performed in the full cohort before internal bootstrap validation. Model performance was assessed using 1,000 bootstrap resamples, with discrimination quantified by the area under the receiver operating characteristic curve (AUC), calibration assessed by calibration plots, slope, intercept, and mean absolute error, and clinical utility evaluated using decision curve analysis.

The final AI-assisted model demonstrated strong discrimination, with an AUC of 0.836 (95% confidence interval [CI], 0.804-0.869), and excellent calibration, with a mean absolute error of 0.015. Decision curve analysis indicated a meaningful net clinical benefit across threshold probabilities of 0.10-0.45. Risk stratification based on quartiles of predicted probability categorized patients into lower-, intermediate-, and higher-risk strata, with observed malignancy rates of 43.3%, 86.7%, and 95.0%, respectively, in this surgically managed cohort. The interactive calculator is publicly accessible at https://ruanyingding.shinyapps.io/myshinyapp/. Because the study cohort was restricted to surgically resected nodules, these risk estimates should be interpreted within a malignancy-enriched preoperative surgical setting.

An AI-assisted clinico-quantitative imaging nomogram was developed and internally validated to support individualized preoperative malignancy risk assessment for indeterminate PNs ≤ 3 cm. Because the model was derived in a surgically selected, malignancy-enriched cohort, it is best interpreted as a tool for preoperative surgical decision support rather than for screening or incidental pulmonary nodule populations. External validation and, if necessary, recalibration in independent unselected cohorts are required before broader implementation.