Ensemble learning based lung and colon cancer classification with pre-trained deep neural networks

Lung and colon cancer are among the most common cancer types with the highest mortality rates worldwide. In recent years, studies have been carried out on the early diagnosis and detection of lung cancer and colon cancer. Also, new treatment methods and options have been investigated. With the development of early diagnostic methods, the survival rate of patients has increased. Therefore, in this study, a transfer learning-based deep ensemble learning model is proposed for the classification of lung and colon cancer from histopathological images.

Experiments were carried out on the LC25000 dataset with the proposed approach. DenseNet121, InceptionV3, MobileNet, ResNet50, ResNet101, VGG16, and Xception models were used for transfer learning. These models were implemented with different variations for ensemble learning.

It is seen that the most successful transfer learning model is the InceptionV3 model, and the ResNet101 model has the lowest performance. The DETL_V1 model, in which all models were used, was the most successful model with an accuracy rate of 99.78%.

By focusing on top models and using ensemble learning to combine their predictions, better performance was obtained compared to any single model on its own. Obtaining a more robust and generalizable model was achieved by combining multiple models that are trained on different subsets of the data with different architectures. The results proved that the proposed approach is supportive for decision in clinical diagnosis processes.