Lung and colon cancer detection with convolutional neural networks on histopathological images

Abstract

Lung cancer is deadly cancer same as colon cancer, both of them can grow simultaneously. Most researchers conduct research to detect one disease on one single body organ. So, in this study, we proposed a computer-aided diagnosing system using the Convolutional Neural Network (CNN) to detect lung and colon cancer tissues on the LC25000 dataset. The LC25000 dataset contains 25000 histopathological color image samples of colon and lung tissues which indicated cancer (adenocarcinoma) or not. In this study, we used three pre-trained CNN models, which are ShuffleNet V2, GoogLeNet, and ResNet18 also one simple customized CNN model. From the evaluation metrics tables given in this study, the highest accuracy to classify lung cancer was gained by ResNet18 with 98.82% accuracy but the shortest training time gained by ShuffleNet V2 with 1749.5 seconds. ShuffleNet V2 was the best model used to classify colon data, it gives 99.87% accuracy with the fastest training times 1202.3 seconds. The customized CNN model proposed by us get 93.02% accuracy to classify lung cancer and 88.26% accuracy for colon cancer. The proposed CNN model also gained the shortest training time which was better than GoogLeNet and ResNet18.