Comparison of Four Transfer Learning and Hybrid CNN Models on Three Types of Lung Cancer
Abida Sultana, Tahsin Tasnia Khan, Tahmim Hossain
Abstract
The abnormal growth of cells in the lungs led to lung cancer is the second most fatal cancer in the world according to the World Health Organization(WHO) reported in 2020. Lung cancer starts affecting one single lymph node of the lungs and can spread to other body organs, even the brain. Early classification of the types of cancer might ease the way of treatment, which can save millions of lives every year. The faster and precise classification of lung cancer types is the motivation of this study.Our dataset consists of 15000 CT scan images of three types: Lung benign tissue, Lung adenocarcinoma, and Lung squamous cell carcinoma. Five artificial deep neural networks (2-D CNN with SVM, ResNet-50, InceptionResNetV2, Inception-V3, and VGG-19) have been applied in the classification of three types of lung cancer in this research. Various measures including accuracy, precision, recall, and F1 score have been applied to assess the performance of these models. Inception-V3 has been shown the best validation accuracy of 99.13% among the implemented transfer learning and CNN-SVM approaches. Moreover, the comparative analysis of this paper concludes that Inception-V3 is the best-suited model for accurately classifying the types of lung cancer.