Lung And Colon Cancer Classification Using Medical Imaging: With Best Image Pre-Processing Techniques Employing Transfer Learning Approach

Abstract

Early detection and treatment of lung and colon cancer can reduce death rates globally. The most popular, effective, and successful imaging method for detecting lung and colon cancer at the moment is transfer learning. The goal of this research is to determine which of the following 10 CNN architectures—VGG19, MobileNet, InceptionV3, ResNet50, ResNet50V2, ResNet101, MobileNetV2, DenseNet201, VGG16 and Xception—offers the most efficiency in terms of identifying lung and colon cancer while minimizing data loss and finishing time. A matrix for evaluation is used to compare completion time, performance, and data loss. When it comes to accuracy, MobileNet tops the list, followed by VGG19 in second place and VGG16 in third. There are 25,000 photos in the dataset. MobileNet's training accuracy, validation accuracy, and testing accuracy were the highest at 99.90%, 99.88%, and 99.58%, respectively. With 45 seconds each epoch, it has the shortest completion time, 0.332 percent data loss, and the best outcome in the shortest epoch. The suggested approach, which is based on image processing and transfer learning, produces best accuracy, quickest completion time, and the least amount of data loss.