Abstract:
In this comprehensive study, highly effective advanced deep learning architectures are tested at detecting chest cancer from computed tomography (CT) scan images. The research is centered around six state of the art convolutional neural network (CNN) base models VGG19, ResNet152V2, DenseNet201, SE-ResNet152, Xception and InceptionV3 and evaluates the performance of these models in their original settings and after applying transfer learning techniques. The models' diagnostic accuracy and reliability for chest cancer diagnosis, a major global health problem of concern, is to be determined through automated image based analysis. In this study, a very rigorous methodological framework that includes dataset preparation, preprocessing and model training is applied, and transfer learning is used to improve the efficiency and generalization of learning. Also, with strong focus on test accuracy, each model’s performance was assessed using key evaluation metrics. The experimental results show that DenseNet201, ResNet152V2, Xception and InceptionV3 are most accurate with an original accuracy of 97%. VGG19 dropped from 97% to 95% of accuracy under transfer learning. The results suggest that transfer learning tends to continue to support or enhance model effectiveness generally, although VGG19 is an example of an exception. This study in overall indicates the possibility of CNN based models like DenseNet201 and ResNet152V2 in development of accurate and scalable AI based solutions for early chest cancer detection from CT scans.
