Deep Learning-Based Lung Cancer Detection Using CT Scan Images: A Comparative Study of VGG16, ResNet50, and DenseNet121

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Lung cancer is a major problem because it kills more people every year than any other type of cancer. If we can find the abnormal areas early, people are much more likely to recover. Doctors usually use a type of scan called a CT scan to look for these abnormal areas. The problem is that reviewing all the scan images is very tedious, and even experienced doctors can miss something. We wanted to see whether computers could help with this task. We tried using three kinds of computer programs, VGG16, ResNet50 and DenseNet121, to look at the pictures from the scan. We used a collection of pictures called the LIDC-IDRI collection to teach the computers what to look for. DenseNet121 performed best at identifying the regions. We think this is because it is good at using the information it has to make decisions. We also used a tool called Grad-CAM to see what the computers were actually looking at. It showed that the better programs focused on the abnormal areas, not just the normal tissue. Our results show that using computers to help with this task is promising, and it could help doctors make better decisions. We also know there is still a lot of work to do to make it perfect.