Deep Learning Techniques for Lung Cancer Recognition

Abstract

Globally, lung cancer is the primary cause of cancer-related mortality. Higher chance of survival depends on the early diagnosis of lung nodules. Manual lung cancer screenings depends on the human factor. The variability in size, texture, and shape of lung nodules may pose a challenge for developing accurate automatic detection systems. This article proposes an ensemble approach to tackle the challenge of lung nodule detection. The goal was to improve prediction accuracy by exploring the performance of multiple transfer learning models instead of relying solely on deep learning models. An extensive dataset of CT scans was gathered to train the built deep learning models. This research paper is focused on the Convolutional Neural Networks' (CNNs') ability to automatically learn and adapt to discernible features in the lung images which is particularly beneficial for accurate classification, aiding in identifying true and false labels, and ultimately enhancing lung cancer diagnostic accuracy. This paper provides a comparative analysis of the performance of CNN, VGG-16, and VGG-19. Notably, the built transfer learning model VGG-16 achieved a remarkable accuracy of 95%, surpassing the baseline method.