Gastric cancer is the fourth most common cancer and the second major cause of cancer death worldwide. Early detection of gastric cancer by endoscopic surveillance is actively investigated to improve patient survival, particularly utilizing the newly developed magnifying narrow-band imaging endoscopy in the stomach. However, reviewing the endoscopic data is time consuming and obliges intense labor of profoundly experienced doctors.
In this work, we have proposed a method for detecting ulcer regions from endoscopic images. The basic goal is to classify images as cancerous (abnormal) or non-cancerous (normal) in the light of these extracted features. We perform feature extraction, selection and ranking technique to identify the pertinent set of features. We further used these features to build and train our binary classifier to classify images into its correct class label. Extensive classification experiments are carried out using the naive based classifier, support vector machines (SVM), bagging and random forest on our image data which validate that it is promising to employ the proposed texture features to recognize cancer in endoscopic images. We conclude that our proposed approach for Automated Diagnostic System for Gastric Cancer Detection can assist physicians in faster and efficient detection of cancer. We have achieved “an accuracy of 87% on training dataset and 89% on testing dataset using Random Forest Classifier, thus improving the overall system efficiency by 11% and reducing manpower by 40%.”