Assessing the Accuracy of a Deep Learning Method to Risk Stratify Indeterminate Pulmonary Nodules.

Area of Science:

• Pulmonary Medicine
• Artificial Intelligence
• Medical Imaging

Background:

• Management of indeterminate pulmonary nodules (IPNs) is challenging, often leading to invasive procedures and delayed diagnoses.
• There is a need for improved strategies to optimize surveillance and reduce unnecessary interventions for IPNs.

Purpose of the Study:

• To develop and validate a deep learning (DL) method for enhanced management of IPNs.
• To improve the accuracy of risk stratification for IPNs.

Main Methods:

• A Lung Cancer Prediction Convolutional Neural Network (LCP-CNN) model was developed using CT images of IPNs.
• The LCP-CNN model was trained on data from the National Lung Screening Trial and validated on external cohorts from two academic institutions.

Main Results:

• The DL model demonstrated superior performance in external validation cohorts compared to a common clinical risk model (AUCs of 83.5% and 91.9% vs. 78.1% and 81.9%).
• The LCP-CNN model achieved significant net reclassification improvements for both ruling in and ruling out malignancy in validation cohorts.
• The DL algorithm improved accuracy in predicting disease likelihood across different management thresholds.

Conclusions:

• The developed deep learning algorithm effectively reclassifies IPNs into low- or high-risk categories.
• This AI-driven approach has the potential to significantly reduce unnecessary invasive procedures and diagnostic delays for lung nodules.
• The study highlights the clinical utility of DL in improving the management of indeterminate pulmonary nodules.