Automated Stomach Cancer Detection System Using Dual-Layered CNN and SVM

Abstract

Bowel cancer is a difficult malignancy to diagnose early since it exhibits almost no signs at all. The most often used procedure for identifying stomach cancer is wireless capsule endoscopy. Due to the structure and color of the stomach, successfully identifying stomach cancer at an early stage with WCE images is a time-consuming medical procedure. The fundamental objective of this study is to develop a computer-aided method for more accurately detecting malignancy in WCE images at an early stage. This study employed a Convolutional Neural Network (CNN) and machine learning approach to early detection of stomach cancer. Dual-Layer CNN (DL-CNN) is used as a feature extractor to extract deep features from the input WCE images automatically. Additional texture features are measured for these WCE using a method called High Correlated Feature Extraction (HCFE). Finally, cancer and non-cancerous images are classified using the Multi SVM machine learning method. MAE and RMSE were used to prove the proposed stomach cancer detection method's training efficiency. The DL-CNN-HCFE approach extracts features that are separated into two groups and classified using SVM. The proposed DL-CNN-HCFE-SVM approach achieve a 95 per cent classification accuracy, which is higher than that of existing methods.