Decision Level Fusion of Transfer Learning based Models for Diagnosing Lung and Colon Cancer

Abstract

Lung and colon cancers are one of the most known cancer types that cause death. However, in most cases, early detection of the disease has critical impact in limiting the spread of tumor cells. Therefore, the diagnosis of the disease is considered the first step for treatment to improve the chance of survival. Artificial intelligence provides great assistance for specialists in the field of diagnosing pathological tissue by saving effort and time with accurate decisions. In this study, the high potential of ensemble learning was adapted to diagnosing colon and lung cancers by establishing multiple deep networks based computer-aided diagnosis system for infected cancer cells. In the model ensemble, the individual classification results of DenseNet201, ResNet101 and EfficientNet-b0 models were combined in terms of hard voting idea. The individual classification accuracies of DenseNet201, ResNet101, and EfficientNet-b0 are resulted in 99.94%, 99.93%, and 98.64% for colon cancer and 99.05%, 98.93%, and 95.79% for Lung cancer, respectively. However, the ensemble approach increased the performances to 99.98% and 99.78% for colon and lung cancer datasets respectively.