A Synthesizing Semantic Characteristics Lung Nodules Classification Method Based on 3D Convolutional Neural Network.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Early lung cancer detection significantly impacts patient survival and recovery.
• Computer-aided diagnosis (CAD) systems offer decision support for early lung cancer diagnosis.
• Current deep learning models for CAD lack interpretability, hindering clinical trust.

Purpose of the Study:

• To develop an interpretable deep learning model for classifying malignant pulmonary nodules.
• To enhance the diagnostic capabilities of computer-aided diagnosis systems in lung cancer detection.
• To provide explainable predictions that assist radiologists in clinical decision-making.

Main Methods:

• Proposed a Semantic Characteristic-combined Convolutional Neural Network (SCCNN) model.
• Utilized 3D multi-view lung nodule samples extracted via spatial sampling.
• Incorporated semantic characteristics from radiology reports as an auxiliary task and an attention module for feature fusion.

Main Results:

• Achieved 95.45% accuracy and 97.26% ROC curve area on the LIDC-IDRI dataset.
• The SCCNN model demonstrated superior performance compared to standard 3D CNN approaches.
• The model provided intuitive explanations for its predictions, enhancing interpretability.

Conclusions:

• The proposed SCCNN model effectively classifies benign and malignant lung nodules with high accuracy.
• The model's interpretability assists in understanding its prediction process, supporting clinical diagnosis.
• This approach advances the application of deep learning in medical imaging for lung cancer diagnosis.