Using Multi-level Convolutional Neural Network for Classification of Lung Nodules on CT images

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Lung cancer is a leading global cancer, necessitating early and accurate diagnosis for improved survival rates.
• Computed Tomography (CT) is a crucial tool for detecting lung cancer, aiding clinicians in diagnosis.
• Classifying lung nodules accurately is vital for effective lung cancer management.

Purpose of the Study:

• To develop and evaluate a novel Multi-Level Convolutional Neural Network (ML-CNN) for lung nodule malignancy classification.
• To investigate the efficacy of multi-scale feature extraction and concatenation for improving classification performance.
• To perform ternary classification of lung nodules into benign, indeterminate, and malignant categories.

Main Methods:

• Development of a Multi-Level Convolutional Neural Network (ML-CNN) integrating three Convolutional Neural Networks (CNNs).
• Extraction of multi-scale features from lung nodule CT images using the CNNs.
• Concatenation of flattened feature vectors from each CNN level to enhance model performance.

Main Results:

• The ML-CNN model achieved an accuracy of 84.81% in ternary classification of lung nodules.
• The model demonstrated high performance without requiring any manual preprocessing algorithms.
• The proposed ML-CNN outperformed existing methods in the ternary classification task.

Conclusions:

• The developed ML-CNN is an effective deep learning model for accurate lung nodule malignancy classification.
• Multi-scale feature extraction and fusion significantly contribute to the model's classification capabilities.
• This approach shows promise for improving early lung cancer diagnosis and patient outcomes.