Leveraging Transfer Learning and Attention Mechanisms for a Computed Tomography Lung Cancer Classification Model

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Lung cancer is a leading cause of cancer mortality globally.
• Late diagnosis significantly impacts patient survival rates.
• AI, particularly deep learning, offers potential to improve diagnostic accuracy and efficiency in radiology.

Purpose of the Study:

• To develop and evaluate a deep learning model for automated lung cancer subtype classification from CT images.
• The model integrates Residual Network 50 Version 2 (ResNet50V2) with Squeeze-and-Excitation (SE) blocks.
• The goal is to enhance diagnostic capabilities in radiology.

Main Methods:

• A dataset of 1,000 lung CT images was used, categorized into adenocarcinoma, large cell carcinoma, squamous cell carcinoma, and normal tissue.
• A fine-tuned ResNet50V2 architecture with SE blocks was employed for feature recalibration.
• The model was trained and evaluated using standard machine learning metrics including accuracy, AUC, precision, recall, and F1-score.

Main Results:

• The model achieved a test accuracy of 90.16% and an overall AUC of 0.9815.
• High class-wise AUCs were observed, with values ranging from 0.9523 to 0.9977.
• Strong precision, recall, and F1-scores across all lung cancer subtypes and normal tissue indicate robust performance.

Conclusions:

• The ResNet50V2 model with SE blocks demonstrates high performance in classifying lung cancer subtypes from CT images.
• This AI approach shows potential to assist radiologists, particularly in resource-limited settings.
• Future research should include external validation and exploration of advanced architectures like Vision Transformers.