Detection of Lung and Colon Cancer Using XRx-Net With Histopathological Images

Abstract

Lung and Colon Cancer (LCC) is a prevalent type of cancer influenced by factors such as genetics, smoking, second-hand smoke exposure, and alcohol consumption. Traditionally, cancer diagnosis, including grading and staging, relies on histopathology, where pathologists examine samples under a microscope. This process is time-consuming and complex. To address these challenges, the Xception ResNeXT Network (XRx-Net) framework is introduced for more efficient LCC detection. In this approach, histopathological images of LCC are processed through an Adaptive Bilateral Filter (ABF) for initial filtering. Next, hookNet is used to segment the cancerous cells. Features are then extracted using Wavelet texture analysis and Convolutional Neural Networks (CNNs). Later, these features are input into the LCC detection phase, where the XRx-Net framework is a combination of the Xception model and ResNeXt Networks employed for classification. The framework successfully identifies various outputs, including lung benign tissue, lung adenocarcinoma, lung squamous cell carcinoma, colon benign tissue, and colon adenocarcinoma. The XRx-Net framework achieves a testing accuracy of up to 91.34%, with a True Positive Rate (TPR) of 92.36% and a True Negative Rate (TNR) of 90.14%.