Comparative Analysis of Deep Learning Methods on CT Images for Lung Cancer Specification

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Lung cancer is a leading cause of cancer deaths globally.
• Early detection and precise tumor localization are critical for patient survival.
• Computed tomography (CT) scans are vital for lung cancer diagnosis.

Purpose of the Study:

• To apply deep learning techniques for early lung cancer detection using CT images.
• To accurately segment identified lung tumors for better treatment planning.

Main Methods:

• Utilized pre-trained convolutional neural networks (CNNs) for lung cancer detection, including MobileNetV2, ResNet152V2, InceptionResNetV2, Xception, VGG-19, and InceptionV3.
• Employed segmentation models like UNet, SegNet, and InceptionUNet to delineate tumor regions.
• Evaluated model performance using detection accuracy and Jaccard index for segmentation.

Main Results:

• InceptionResNetV2 achieved a high detection accuracy of 98.5%.
• UNet demonstrated superior performance in tumor segmentation, yielding a Jaccard index of 95.3%.
• The study highlights the efficacy of specific deep learning architectures in lung cancer analysis.

Conclusions:

• Deep learning models, specifically InceptionResNetV2 and UNet, show strong potential for enhancing early lung cancer diagnosis.
• These models can effectively detect and segment lung tumors, aiding clinicians in treatment decisions.
• Further research can refine these AI tools and explore their broader applications in medical diagnostics.