Early Cancer Diagnosis and Ulcer Classifications In Endoscopy Videos Using Deep Learning Approach

The latest development of AI with endoscopy video surveillance and analysis techniques enables viewing both ends of a patient’s digestive tract, which helps to detect abnormalities like cancer and ulcer in the stomach. Using traditional endoscopy techniques, an experienced clinician must spend at least an hour for classification by viewing and analyzing approximately 5,000 images produced during one examination. Therefore, it is essential to identify the optimal features using Deep Learning in endoscopy video surveillance techniques. The video observation for object detection strategy to find the connected region to characterize and relate with the diagnostic program constructed based on optimal features with deep learning architecture and is trained with 10,000 (Esophago Gastro Duodenoscopy) EGD data set. The paper proposes a Medical Diagnostic Decision Support System (MDDSS) framework, which comprises two phases. In the initial phase, the object recognition and characterization of gastric ulcer and gastric cancer in the endoscopy video, we use Flower Pollination (FP) algorithm. This procedure deals with using color features selection, as well it realizes the efficacy of 97.56% for detecting and classifying bleeding cancer and ulcer growth regions.