Exploration of the Possibility of Early Diagnosis for Digestive Diseases Using Deep Learning Techniques

Digestive Diseases are a commonality among Americans (almost one out of every five Americans are affected by digestive diseases). The most effective way of identifying such diseases is through Endoscopic Examinations. While patients normally have to wait for a long time to take high quality images from Endoscopic Examinations for diagnosis. This waiting period is problematic because some of these diseases can turn cancerous. This brings in the need for quick and accurate diagnosis of digestive diseases because the ability to identify diseases and the severity of the diseases vary from one doctor to another. This study analyzes the ability for our deep learning models to distinguish between the Z-Line, the Pylorus, the Cecum, Esophagitis, Ulcerative Colitis, Polyps, Dyed-Lifted Polyps and dyed resurrection margins. We developed a Convolutional Neural Network algorithm to distinguish between the anatomical landmarks to reach pathological findings. We also experimented with three transfer learning models to compare the final results, and ended up with 85 percent accuracy primarily with research optimization of the AlexNet type model. This indicates a significant impact of our artificial intelligence technology on making tremendous strides to be a tool that doctors may use in the near future, so as to improve the overall health of humanity.