Malignancy risk stratification for pulmonary nodules: comparing a deep learning approach to multiparametric...

Area of Science:

• Artificial Intelligence in Medical Imaging
• Deep Learning for Oncology
• Pulmonary Nodule Classification

Background:

• Pulmonary nodules are frequently detected incidentally, posing a challenge for accurate risk stratification.
• Existing methods like the Brock model and Lung-RADS® have limitations in classifying nodule malignancy risk.
• There is a need for reliable decision support systems to aid clinicians in managing pulmonary nodules.

Purpose of the Study:

• To evaluate the performance of a deep learning model, lung-cancer-prediction-convolutional-neural-network (LCP-CNN), for pulmonary nodule risk classification.
• To compare LCP-CNN against established multiparametric statistical methods (Brock model and Lung-RADS®).
• To assess LCP-CNN's efficacy across patient cohorts with varying risk profiles and underlying pulmonary diseases.

Main Methods:

• Retrospective analysis of CT scans from 297 patients with 422 pulmonary nodules (5-30 mm).
• Ground truth established by histology or follow-up stability; 105 nodules were malignant.
• Performance evaluation using ROC analysis, comparing LCP-CNN, Brock model, and Lung-RADS® across different subcohorts.

Main Results:

• LCP-CNN demonstrated superior performance (AUC 0.92-0.93) compared to the Brock model in total and screening cohorts.
• LCP-CNN showed significantly higher sensitivity than Brock model and Lung-RADS® at a 5% risk threshold in multiple cohorts.
• No significant performance differences were observed for LCP-CNN across various patient risk profiles or lung disease types.

Conclusions:

• Deep learning-based decision support systems, like LCP-CNN, show significant potential for integration into clinical workflows.
• LCP-CNN offers improved accuracy and efficiency in pulmonary nodule risk classification, addressing limitations of traditional models.
• These AI-driven approaches can complement or potentially replace current methods, enhancing clinical decision-making for pulmonary nodules.