Research On Screening Methods For Chronic Atrophic Gastritis Based On Deep Learning

Chronic atrophic gastritis is an important stomach disease,and its occurrence and development seriously threaten human health.When the human body suffers from such diseases,it is not only necessary to endure dyspepsia,stomach upset and other discomforts,but also causes frequent bleeding,induces other stomach diseases,and even transforms into gastric cancer.Therefore,it will be chronically atrophic in medicine.Gastritis is characterized as a serious precancerous lesion.If chronic atrophic gastritis can be detected and diagnosed at an early stage,the patient's condition can be intervened and treated in advance to reduce the risk of worsening the disease and cancer.However,the clinical manifestations of chronic atrophic gastritis are very complicated.The diagnostic effect is directly related to the doctor's experience and medical level,so that the gastroscopic image cannot be the main basis for the diagnosis of chronic atrophic gastritis.It must be combined with gastric mucosal biopsy.In order to get the final diagnosis.The manual diagnosis process has the drawbacks of time-consuming,laborious and subjective differences.With the rapid development of computer hardware and algorithms,artificial intelligence technology with deep learning as the core has been widely used in various industries.In the field of health care,due to the excellent performance of deep learning in tasks such as image recognition,it has been favored by medical workers as an auxiliary diagnostic tool.It provides a new research direction for the automatic diagnosis of chronic atrophic gastritis.Gastroscopic images are an indispensable imaging medium for the diagnosis of chronic atrophic gastritis.Highly clear gastric morphology and mucosal epidermal features can be observed.It also provides a rich data resource for deep learning and assisted screening for chronic atrophic gastritis.In this study,the images of gastric antrum and stomach angle in the gastroscope image were first classified,and the chronic atrophic gastritis on the gastric antrum and stomach angle images were screened on the basis of classification.Aiming at the problem of image classification of gastric antrum and stomach angle,this paper proposes GAH-CNN deep learning model and combined model method.In the experiment process,the image is preprocessed by INPAINT_TELEA algorithm to remove the watermark noise of the image.Then the SIFT feature extraction algorithm,HOG algorithm and LBP algorithm are used to extract the traditional features of the image,and the support vector machine algorithm is applied to the gastric antrum angle.The images are classified;finally,these models are combined to generate a combined model to improve classification performance.Costsensitive algorithm design is used to process unbalanced data sets.The classification accuracy of SIFT,HOG and LBP combined with SVM algorithm is 92%,93% and 88% respectively;the accuracy of GAH-CNN model is 95%,which exceeds the traditional image feature combined with SVM classification algorithm;The combined accuracy of the model's combined model reached 99% of the best results.Sy BN-Dense Net and Sy BN-Res Net models were proposed for the screening of chronic atrophic gastritis on gastric antrum and gastric angle images.The two models achieved good results in the screening of chronic atrophic gastritis.The corresponding recognition accuracy rates were 98.6% and 94.3%,respectively.The CAM generated thermograms were used to explore the discriminant basis for the model to identify the lesion images.At the same time,the effects of different CNN network structures and parameters on the performance of automatic screening for chronic atrophic gastritis were discussed.The experimental results show that the chronic atrophic gastritis at the gastric antrum is easier to be screened and diagnosed than the gastric horn.The accuracy of the deep learning algorithm for screening chronic atrophic gastritis has reached the high-level doctor's manual screening.s level.Therefore,the deep learning model studied in this paper provides a feasible solution for the automatic screening of chronic atrophic gastritis.