Lung cancer diagnosis with GAN supported deep learning models

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Oncology

Background:

• Lung cancer is a major cause of cancer mortality globally, necessitating early detection for improved outcomes.
• Conventional diagnostic methods like biopsies and manual CT interpretation are slow and variable.
• Deep learning (DL) presents an opportunity for faster, more objective medical image analysis.

Purpose of the Study:

• To classify lung CT images into benign, malignant, and normal categories using advanced DL.
• To enhance diagnostic accuracy in lung cancer detection through specialized AI models.

Main Methods:

• A dataset of 1097 lung CT images was augmented using Generative Adversarial Networks (GANs).
• Data preprocessing included histogram equalization and noise reduction, followed by a 70/30 train-test split.
• Multiple DL architectures (VGG19, AlexNet, InceptionV3, ResNet50, custom CNN) were trained and evaluated, with Faster R-CNN integration for detection.

Main Results:

• The custom CNN model achieved a peak accuracy of 99%, outperforming established models like VGG19 (97%).
• Integration of Faster R-CNN enhanced the model's sensitivity and classification precision.
• The study demonstrated superior performance of DL models in lung nodule classification.

Conclusions:

• GAN-enhanced deep learning models show significant potential for accurate lung cancer classification.
• These AI-driven tools can support clinicians in early lung cancer detection and diagnosis.
• The developed models offer a promising avenue for improving patient survival rates through timely intervention.