Discrimination between transient and persistent subsolid pulmonary nodules on baseline CT using deep transfer...

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Subsolid nodules (SSNs) on CT scans require differentiation between transient and persistent types.
• Accurate discrimination is crucial for appropriate patient management and avoiding unnecessary interventions.

Purpose of the Study:

• To develop and validate a deep learning (DL) model for discriminating transient from persistent SSNs on baseline CT scans.
• To assess the performance of the DL model against experienced radiologists.

Main Methods:

• A cohort of 1414 SSNs was used, with a transfer learning approach applied to a pre-trained pulmonary nodule classification model.
• The dataset was divided into development, tuning, and validation sets.
• Model performance was evaluated using AUC, accuracy, sensitivity, and specificity, and compared to radiologists' performance and Lung-RADS categorization.

Main Results:

• The DL model achieved an AUC of 0.926, with 0.859 accuracy, 0.863 sensitivity, and 0.858 specificity on the validation set.
• The model outperformed two experienced radiologists in discriminating SSNs.
• Effective feature extraction was confirmed through visualization techniques.

Conclusions:

• A transfer learning-based DL model demonstrates strong performance in differentiating transient and persistent SSNs.
• Reliable diagnosis of nodule persistence is achievable on baseline CT, enabling earlier diagnosis and enhanced patient care strategies.