Classification of lung nodules in CT images based upon a multiplane dense inception network

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Lung cancer remains a leading global cause of mortality.
• Early detection of cancerous lung nodules is crucial for effective treatment.
• Computer-aided diagnosis (CAD) systems are vital for analyzing lung nodules in CT images.

Purpose of the Study:

• To develop a deep learning-based CAD system for predicting lung nodule malignancy.
• To enhance the accuracy of lung nodule classification in computed tomography (CT) images.

Main Methods:

• A multiplane dense inception network (MPDINet) was developed, integrating DenseNet and GoogLeNet architectures.
• Texture features from Local Binary Patterns (LBP), Gray-Level Co-occurrence Matrix (GLCM), Gray-Level Run Length Matrix (GLRLM), and Gray-Level Size Zone Matrix (GLSZM) were utilized.
• A three-plane (axial, coronal, sagittal) network design with perinodular zones was implemented for robust nodule characterization.

Main Results:

• The MPDINet model was evaluated on the LIDC-IDRI dataset, comprising 1235 lung nodules.
• With inverse difference moment (IDM) feature concatenation, the model achieved high performance: AUC (0.9821), sensitivity (0.9426), specificity (0.9732), and precision (0.9499).
• These results demonstrate the model's capability for accurate lung nodule classification.

Conclusions:

• The developed MPDINet architecture, enhanced with handcrafted feature concatenation, shows significant promise.
• This approach is suitable for various lung nodule classification applications using CT imaging.
• The study highlights the potential of deep learning in improving lung cancer diagnosis.