A deep learning model to enhance lung cancer detection using 'Dual-Branch' model classification approach

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Oncology

Background:

• Lung cancer is a leading cause of cancer-related deaths globally.
• Computed tomography (CT) is crucial for lung cancer diagnosis, but challenges like limited data and input modalities persist.
• Accurate lung cancer classification is vital for improving patient survival rates.

Purpose of the Study:

• To introduce and evaluate the Dual-Branch Model Classification Approach (DbMCA) for enhanced lung cancer detection.
• To assess the impact of sample size and dual-input modalities (image and mask) on diagnostic model performance.
• To address limitations in existing lung cancer detection methods, such as data scarcity and single-modality inputs.

Main Methods:

• Developed a two-stage Dual-Branch Model Classification Approach (DbMCA).
• Integrated both image and mask data for a multi-modal classification strategy.
• Conducted comparative experiments on the LIDC-IDRI dataset with varying sample sizes.

Main Results:

• DbMCA achieved 91.21% accuracy and 91.18% F1-score on smaller datasets, and 98.04% accuracy and 98.01% F1-score on larger datasets.
• Demonstrated superior performance compared to baseline models, especially with integrated multi-modal information.
• Highlighted the sensitivity of Convolutional Neural Network (CNN) performance to sparse mask data, with Deep Neural Networks (DNN) and Support Vector Machines (SVM) showing better scalability.

Conclusions:

• The DbMCA significantly improves lung cancer detection accuracy and scalability by integrating multi-modal data.
• The model shows potential for detecting subtle lung cancer patterns, outperforming weaker baselines.
• Future work should focus on enhancing image quality, expanding datasets, and addressing segmentation constraints for better generalization.