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

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Pulmonary nodules (PNs) require accurate preoperative risk assessment for malignancy.
• Distinguishing benign from malignant PNs is crucial for appropriate patient management.

Purpose of the Study:

• To develop and validate an AI-assisted prediction model for preoperative malignancy risk in solid and part-solid pulmonary nodules (PNs) ≤ 3 cm.
• To integrate automatically extracted quantitative imaging features with clinical data for individualized risk assessment.

Main Methods:

• Retrospective analysis of 951 patients with PNs ≤ 3 cm who underwent surgical resection.
• Utilized AI software (InferRead CT Lung AI) for automatic measurement of quantitative CT features.
• Developed a multivariable logistic regression model incorporating AI-derived features, clinical variables, and inflammatory markers, followed by internal bootstrap validation.

Main Results:

• The AI-assisted model achieved strong discrimination with an AUC of 0.836 and excellent calibration (MAE 0.015).
• Decision curve analysis showed meaningful clinical utility across a range of threshold probabilities.
• Risk stratification identified distinct malignancy rates (43.3% to 95.0%) across predicted risk strata in the surgically managed cohort.

Conclusions:

• An AI-assisted clinico-quantitative imaging nomogram provides a validated tool for preoperative malignancy risk assessment in indeterminate PNs.
• The model is best suited for preoperative surgical decision support in malignancy-enriched cohorts.
• External validation in unselected cohorts is necessary for broader implementation.