Hierarchical classification of early microscopic lung nodule based on cascade network

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Early-stage lung cancer detection relies on identifying isolated lung nodules from numerous CT slices.
• Microscopic lung nodules pose challenges due to their small size and limited characterization.
• Accurate classification of lung nodules is crucial for effective treatment strategies.

Purpose of the Study:

• To develop a sensitive lung nodule classification model for microscopic nodules.
• To propose a novel cascade classification method for lung nodules.
• To enhance the accuracy of lung nodule categorization.

Main Methods:

• Utilized the Resnet34 network as the foundational classification model.
• Introduced a cascade lung nodule classification approach.
• Classified nodules into six distinct categories: ground-glass, solid, benign, malignant, and mixed components.

Main Results:

• The proposed cascade classification method achieved an accuracy of 80.04% on clinical data.
• This accuracy surpasses the performance of conventional multi-classification methods.
• Demonstrated effective classification of six different nodule types.

Conclusions:

• The novel cascade method offers more accurate classification of lung nodules into six categories compared to existing methods.
• This approach provides a rapid, precise, and dependable tool for classifying diverse lung nodule types.
• The study highlights improved accuracy in nodule categorization over traditional multivariate methods.