A Comprehensive Analysis of Deep Learning-Based Approaches for the Prediction of Gastrointestinal Diseases Using Multi-class Endoscopy Images

The human gastrointestinal (GI) system can be affected by various illnesses which results in the death of about two million patients globally. Endoscopy helps to detect such diseases as identifying these abnormalities in GI tract endoscopic images is crucial for therapy and follow-up decisions. However, clinicians require adequate time to examine such follow-ups that hinders manual diagnosis. As a result, the aim of the study is to detect and classify various gastric based diseases using deep transfer learning models such as DenseNet201, EfficientNetB4, Xception, InceptionResNetV2, and ResNet152V2, which have been assessed on the basis of precision, loss, accuracy, F1 score, root mean square error, and recall. In this study, Kvasir's dataset has been used, which is divided into five categories: dyed-lifted polyps, esophagitis, normal cecum, dyed resection margins, and normal colon of endoscopic images. All the images are enhanced by removing the noise before being sent into the deep transfer learning algorithms. During experimentation, it has been analyzed that to detect dyed-lifted polyps, Inception ResNetV2 obtained the highest testing accuracy by 97.32%. On the other hand, Xception model efficiently detects dyed resection margins, esophagitis, normal cecum, and normal colon by computing the best testing accuracy of 95.88%, 96.88%, 97.16%, and 98.88%, respectively.