Medical Pathway Image Recognition Based On Deep Learning

Medical pathway images describe the relationship of molecular transformation in organisms in detail with the form of node-edge-node.A whole pathway image shows all metabolic pathways related to a certain physiological state.The analysis and mining of pathway images can help doctors discover new therapeutic pathways and determine the most effective treatment options.Because of the large number and complex structure of medical pathway images,it will take a lot of time to recognize them manually.This paper focuses on the research of target detection and recognition model based on deep learning and applies it to automatic detection and recognition of pathway images.The specific work of the paper is as follows:Firstly,by analyzing the advantages and disadvantages of various target detection methods based on in-depth learning and their applicable scenarios,in view of the characteristics of medical pathway images with many kinds of elements and complex overall structure,a multi-target detection model SSD(single shot multibox detector)with good detection accuracy and fast detection speed is adopted to realize the detection and location of multiple elements in images.Through the construction of data sets of each element,the detection model is trained.Through the analysis of the accuracy and recall of the detection results,the detection performance of the model is evaluated.The results show that,with a certain number of data sets,increasing the number of iterations in training can effectively improve the accuracy of detection results and reduce the missed detection rate of targets.Then,the method of segmentation and recognition is used to recognize the strings in the image.The commonly used projection segmentation method has good performance in the segmentation of non-cohesive characters.For the cohesive string,this paper uses the drip method to segment the characters.In view of the fact that the strings in pathway images contain many special characters(such as Greek letters,punctuation symbols,etc.),this paper constructs one hundred 32*32 pixel data sets for each character,and then trains them with convolutional neural network to get the character recognition model.Finally,the module of element detection and character recognition is integrated,the framework of medical pathway image recognition is designed and the experimental demonstration system is implemented.