A novel benign and malignant classification model for lung nodules based on multi-scale interleaved fusion integrated...

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Oncology

Background:

• Lung nodules are precursors to lung cancer, necessitating accurate classification for patient survival.
• Deep learning models for lung nodule classification require large datasets and parameters, often unavailable in medical imaging.
• Existing models struggle with multi-scale feature extraction, limiting performance on new data.

Purpose of the Study:

• To develop a lightweight deep learning network for accurate lung nodule classification.
• To address limitations in multi-scale feature extraction and data requirements in medical image analysis.

Main Methods:

• Proposed MIFNet, a lightweight interleaved fusion integration network with multi-scale feature learning modules.
• MIFNet utilizes MIF blocks with convolutional layers (1x1, 3x3) and shortcut links for multi-scale feature extraction.
• The model features a low parameter count (0.7M) and reduced computational cost.

Main Results:

• MIFNet achieved 94.82% accuracy, 97.34% F1 score, 96.74% precision, 97.10% sensitivity, and 84.75% specificity on the LUNA16 dataset.
• The model demonstrated superior performance compared to pre-trained deep networks and state-of-the-art methods.
• MIFNet offers an objective and efficient auxiliary tool for lung nodule classification.

Conclusions:

• MIFNet provides an effective solution for lung nodule classification with high accuracy and efficiency.
• The lightweight architecture and multi-scale feature learning make MIFNet suitable for medical imaging applications with limited data.
• This approach enhances diagnostic capabilities for lung cancer precursors.