Meta-Analysis Of Deep Learning Diagnosis Of Focal Liver Lesions And Research On Target Detection Algorithm Of Hepatic Cystic Echinococcosis

Objective: 1.To explore the diagnostic value of deep learning technology and traditional machine learning method in focal liver lesions by diagnostic Meta-analysis.2.Based on the results of Meta-analysis,the target detection algorithm of cystic hepatic echinococcosis is developed and researched,which can locate and identify cystic hepatic echinococcosis on ultrasound images,make patients receive relevant treatment in time and prevent the further development of the disease.Methods: The data of this study came from 972 patients with cystic hepatic hydatid disease in the Department of Abdominal Ultrasound of the First Affiliated Hospital of Xinjiang Medical University from 2008 to 2020,with a total of 3083 liver ultrasound images.1.Research papers on computer-aided diagnosis of focal liver lesions by searching Chinese and English databases and building their own databases until March 2022.Through literature screening and quality evaluation,the sensitivity and specificity of forest maps were generated by using bivariate method and hierarchical total subject operation characteristic curve after extracting research data.Meta-regression was used to explore the possible sources of heterogeneity.2.Based on Poly-Yolo segmentation algorithm,the nonimaging regions in ultrasound images are removed,and the computation of subsequent models is reduced.Build Yolov5 target detection model for focus location and classification.On this basis,build an accurate classification model of cystic hepatic hydatid disease through integrated model,and evaluate the model through indicators such as accuracy rate and recall rate.Results: Diagnostic Meta-analysis included 24 related literatures.Compared with traditional machine learning model,deep learning method achieved better performance,including sensitivity(91% vs 87%)and specificity(93% vs 87%).2.2.The segmentation algorithm based on Poly-YOLO network can effectively realize the accurate segmentation of the imaging region.Among them,DCS coefficient(U-Net: 0.97 vs OSTU: 0.83 vs Markov: 0.85)and IOU intersection ratio(Poly-Yolo: 0.95 vs OSTU: 0.79 vs Markov: 0.81).The Yolov5 l model was used as the target detection model for cystic hepatic echinococcosis,and the mean average precise(m AP)was 88.1%.Conv Ne Xt-T achieves the best results with an accuracy of 86.0%,a recall of 85.95% and an F1-score of 86.0% by using the snapshot integration algorithm to obtain sub-models and build integration models.Conclusion: In the task of diagnosing focal liver lesions based on ultrasound images,Deep learning method is better than traditional machine learning method.Deep learning approach to analyze ultrasound images of cystic liver worm disease has theoretical feasibility.polyYolo segmentation algorithm can effectively remove the non-imaging areas of ultrasound images,so that Yolov5 l can effectively analyze the liver areas under ultrasound.The model based on ensemble algorithm can further improve the classification accuracy of lesions.The method proposed in this study is expected to be a potential auxiliary diagnostic tool for cystic hepatic hydatid disease.