Research On Detection Of Intestinal Metaplasia In Gastroscope Images Based On Semantic Segmentation

Intestinal metaplasia refers to the replacement of gastric mucosal epithelial cells by intestinal epithelial cells and is a common lesion of gastric mucosa.A large number of studies and findings of early gastric cancer indicate that the regularity of gastric cancer evolution is: normal gastric mucosa-non-atrophic gastritis-atrophic gastritis-intestinal metaplasia-dysplasia-gastric cancer.Therefore,it is believed that the intestinal metaplasia of the gastric mucosa has a very close relationship with gastric cancer.Gastric cancer is one of the most common malignant tumors in the world.The clinical incidence of gastric cancer is very high,and there is no clear status of early gastric cancer under endoscopic observation.This undoubtedly brings great difficulties to microscopy,causing most patients Unable to get timely diagnosis and treatment until the late diagnosis.If prevention is done early and the detection rate is increased,the incidence of gastric cancer can be reduced.The general method for the detection of early gastric cancer is to use gastroscopy.However,as the imaging rate of the endoscope increases,the doctor ’s burden of reading is increasing.The doctor ’s observation of the lesion through the microscope can not avoid the constraints of subjective factors.The emergence of computer-aided diagnosis technology undoubtedly opened up a new direction for the current predicament.How to improve the accuracy of early gastric cancer examination while reducing the work intensity of endoscopists,especially to reduce the miss rate of intestinal metaplasia has become a problem faced by image processing experts.Under such demand,computer-aided diagnosis technology has been widely used and has become one of the hot spots in medical imaging research.This article focuses on the recognition of computer-aided diagnosis of intestinal metaplasia lesions during endoscopic examination of white light endoscopic images.To improve the accuracy and efficiency of computer-aided diagnosis.The main research work and contributions of this article are as follows:1.Summarize the defects of the current semantic segmentation algorithm applied in the detection of intestinal metaplasia.The advantages and disadvantages of traditional methods for intestinal metaplasia lesion tissue examination are summarized.2.The paper studies the application of semantic segmentation neural network based on deep learning technology in the detection of intestinal metaplasia lesions in endoscopic images under white light.The paper introduces the specific methods and design ideas of current mainstream semantic segmentation networks,analyzes the core modules and structures in these networks,and analyzes the detection of intestinal metaplasia lesions in white light endoscopic images by semantic segmentation networks through experimental tests.Pros and cons of the application scenario of semantic segmentation.3.This article uses an improved Deep Labv3 + network for deep learning methods to segment intestinal metaplasia lesions.Multi-scale feature information is captured through the optimal DPC architecture during the encoding process,replacing the original ASPP architecture.It effectively solves the phenomenon of incomplete segmentation within the lesion,and is more in line with the requirements for fine boundaries in semantic segmentation of intestinal metaplasia lesions.This method has been confirmed in the experimental results of the data set.In the binary classification task based on ordinary white light endoscopic images,the average cross-combination ratio reaches 85.17%,the global pixel accuracy rate reaches 92.7%,and only takes 0.05 seconds.Identify an endoscopic image.Therefore,the network model proposed in this paper has excellent performance in terms of speed and accuracy,and can be used as a clinician’s auxiliary diagnosis method,which can effectively improve the accuracy of diagnosis of intestinal metaplasia lesions.