Classification of Lung and Colon Cancer Using Deep Learning Method

Abstract

Cancer seems to have a significantly high mortality rate as a result of its aggressiveness, significant propensity for metastasis, and heterogeneity. One of the most common types of cancer that can affect both sexes and occur worldwide is lung and colon cancer. It is early and precise detection of these cancers which can not only improves the rate of survival but also increase the appropriate treatment characteristics. As an alternative to the current cancer detection techniques, a highly accurate and computationally efficient model for the rapid and precise identification of cancers in the lung and colon region is provided. For the training, validation and testing phases of this work, the LC25000 dataset is used. Cyclic learning rate is employed to increase the accuracy and maintain the computational efficiency of the proposed methods. This is both straightforward and effective which facilitates the model to converge faster. Several transfer learning models that have already been trained are also used, and they are compared to the proposed CNN from scratch. It is found that the proposed model provides better accuracy, reducing the impact of inter-class variations between Lung Adenocarcinoma and another class Lung Squamous Cell Carcinoma. Implementing the proposed method increased total accuracy to 97% and demonstrate computing efficiency in compare to other method.