A Novel Approach Using AI-vision Transformer for CT Scan Analysis for Lung Cancer Detection.

Area of Science:

• Artificial Intelligence in Medical Imaging
• Oncology Diagnostics
• Machine Learning for Healthcare

Background:

• Lung cancer diagnosis is challenging due to subtle imaging differences between benign and malignant tumors.
• Traditional Convolutional Neural Networks (CNNs) have limitations in robustness with varying imaging methods.
• Machine learning advancements are transforming healthcare, enhancing accessibility and personalized patient care.

Purpose of the Study:

• To evaluate the efficacy of Vision Transformers (ViTs) for lung cancer classification using CT scans.
• To assess the robustness of ViTs in simulated real-time diagnostic scenarios with image variations.
• To compare ViT performance against conventional methods for lung cancer detection.

Main Methods:

• A Vision Transformer (ViT) model was trained on 1,190 CT images for lung cancer classification.
• Model performance was validated using simulated real-time scenarios with noisy and blurred image datasets.
• Comparative analysis was conducted against conventional validated methods.

Main Results:

• The ViT model achieved a 98.18% classification rate and a Matthews Correlation Coefficient of 0.9676.
• ViTs demonstrated superior performance in differentiating noisy and blurred images (80-88% accuracy) compared to conventional methods.
• The model exhibited robustness against image variations, maintaining strong diagnostic accuracy.

Conclusions:

• Vision Transformers show promising robustness and high accuracy for lung cancer classification from CT scans, even with image variations.
• ViTs may offer advantages for clinical oncology diagnostics and early cancer detection.
• Future research should utilize larger, clinically validated datasets to confirm ViT benefits in real-world settings.