Knowledge-based Collaborative Deep Learning for Benign-Malignant Lung Nodule Classification on Chest CT.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Accurate identification of malignant lung nodules on chest CT scans is crucial for early lung cancer detection and patient survival.
• Deep learning models show promise in computer vision but face challenges in nodule detection due to limited training datasets.

Purpose of the Study:

• To develop a deep learning model for accurate classification of malignant versus benign lung nodules using limited chest CT data.
• To improve the early detection of lung cancer through enhanced nodule identification.

Main Methods:

• Proposed a multi-view knowledge-based collaborative (MV-KBC) deep model that learns 3-D nodule characteristics by decomposing them into nine fixed views.
• Each view utilizes a knowledge-based collaborative (KBC) submodel with pre-trained ResNet-50 networks to analyze nodule appearance, voxel, and shape heterogeneity.
• An adaptive weighting scheme and penalty loss function were employed for end-to-end training and reduction of false negatives.

Main Results:

• The MV-KBC model achieved 91.60% accuracy and 95.70% AUC for lung nodule classification on the LIDC-IDRI dataset.
• Demonstrated superior performance compared to five state-of-the-art classification approaches.
• The model effectively handles limited training data while minimizing the false negative rate.

Conclusions:

• The MV-KBC deep model offers a robust and accurate solution for classifying malignant lung nodules from CT images.
• This approach shows significant potential for improving early lung cancer diagnosis and patient outcomes.
• The method's effectiveness with limited data makes it valuable for clinical applications.