Multi-Classification and Segmentation of Esophageal Lesions Using an Improved Deep Learning Model from Endoscopic Images

It is important to have an autonomous and precise categorization of esophageal lesions to clinically evaluate the lesion of esophageal disorders and to come up with appropriate diagnostic strategies. This study presents a Hybrid Deep Learning Model (HDLM) for completely automated diagnostics, although endoscopists are required to correct the false data predicted by the testing process if much-supporting data is supplied. HDML comprises with Artificial Neural Network along XGBoost Classifier. This prototype doesn't simply take the place of endoscopists in decision-making. Instead, it complements the role that endoscopists play in decision-making. To help endoscopists improve their reliability when recognizing the kinds of lesions, a knowledge discovery component has indeed been introduced in the categorization job. The goal of this model is to provide a higher degree of assurance about the projected forms of esophageal tumors. The performance of the segmentation task in detecting the sites of esophageal lesions has also been improved by the inclusion of a mutual attention module. Because of its excellent recognition rate (97.81%) and segmented Dice factor (87.91 percent), our model seems to be well suited for real-world use. Endoscopists might benefit from a tool to help them grade esophageal lesions, such as the multi-task deep learning model that has been proposed.