Localizing and Identifying Intestinal Metaplasia Based on Deep Learning in Oesophagoscope

Abstract

Intestinal metaplasia is a precancerous lesion, gastric cancer is a very common malignant tumor, and many people die every year from stomach cancer. Early diagnosis of gastric cancer is critical to reducing patient mortality and overall medical burden. However, traditional endoscopic intestinal metaplasia on the gastric mucosa lacks specific performance. Consequently, subtle changes in precancerous intestinal metaplasia are not obvious limiting diagnostic accuracy. As a clinical computer aid in the diagnosis of early gastric cancer, a deep learning framework model called W-Deeplab was proposed for the identification and localization of intestinal metaplasia lesions. It achieves high-precision semantic segmentation of endoscopic images. As a computer aid to clinicians, it can improve the accuracy and efficiency of intestinal metaplasia diagnosis and reduce misdiagnosis.