Computer-aided Classification of Lung Nodules on CT Images with Expert Knowledge

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Accurate pulmonary nodule classification is crucial for early lung cancer detection.
• Assessing the impact of COVID-19 on lung health necessitates precise nodule analysis.
• Existing methods may lack the nuanced interpretation of nodule characteristics.

Purpose of the Study:

• To develop a computer-aided classification method for lung nodules.
• To leverage expert knowledge for improved nodule diagnosis.
• To enhance the accuracy of lung nodule classification in CT images.

Main Methods:

• A decoupling metric learning model was employed to capture deep nodule features.
• Similarity calculations were performed between current nodules and a database.
• Expert knowledge was extracted from similar nodules for decision-making.

Main Results:

• The method achieved 95.7% accuracy on the LIDC-IDRI dataset.
• An Area Under the Curve (AUC) of 0.9901 was obtained.
• The approach demonstrated effective utilization of expert knowledge.

Conclusions:

• The proposed computer-aided classification method shows high accuracy for lung nodules.
• This technique can significantly aid in lung cancer detection, diagnosis, and therapy.
• Integrating expert knowledge enhances the reliability of automated nodule classification.