Optimization Research On Medical Imaging Processing And Recognition Based On Pulmonary CT Image

Human beings have entered a new era of intelligent life,with the development of deep learning in intelligence algorithms and parallel processing technology of mobile cloud computing.Under this tendency,opportunities coexist with threats in intelligent medical system.As for traditional medical science,images consist of graphs from imaging inspection and pathological section diagnosis,providing important reference for doctors.Image processing and image recognition are core methods of computer vision.Therefore,the medical image processing and recognition should take the specific features and the demand of the efficiency and accuracy into consideration.And the image adjuvant therapy meets the development requirement of smart medical system.The image processing includes preprocessing,segmentation,etc,aiming at improving the recognition effect by optimizing image quality and simplifying the content reasonably.It is the preliminary basis of image recognition,analysis and other types of image technology.The segmentation is the core step of the processing.The related algorithms used to divide the graph into connected domains with no intersection based on various features.The object region will then be extracted from background.There is no general method of image segmentation on account of the feature variety so far.Additionally,the diversity and pathological complexity of medical images also lay the barrier of segmentation.Therefore,it’s necessary to improve the preprocessing and segmentation according to medical features.The image recognition is composed of feature extraction and image classification,which simulates human vision perception.Convolution neural network(CNN)is one of the classical models of deep learning structure which combines two steps into an independent frame.On the one hand,convolution layers are able to extract image features without any subjective interference.On the other hand,some connotative features that haven’t be discovered by modern medical science may be further studied.But high algorithm complexity of tradition convolution is time consuming.What’s more,to clearly response the focus situation,medical images are generally of high resolution.In this case,the application of CNN in medical image recognition will lead to higher demands of the equipment performance.Hence,it is of great importance to accelerate convolution model with lower memory consumption.In view of the accuracy and efficiency of medical diagnosis,we analyze the features of medical images for the processing and recognition optimization.For medical image processing issues,the preprocessing will take the imaging process into account to analyze the quality issues first.Region growing algorithms are suitable for objects with connected area,and organs and other human tissues fit the property.Then enhance E-SRG region growing algorithms with space correlation to select the seed area automatically,and separate the complicated area.To handle with medical recognition,we propose MSM method and its enhanced method BLR,based on block-reshape method.Both optimization methods accelerate the calculation with no drop of accuracy and lower the memory overhead as well.The experimental results verify the superior performance of proposed methods.The preprocessing method can improve the quality of the images effectively.And E-SRG is demonstrated to be able to segment each organization area properly,with no restrictions of the dataset label and the image size.Both MSM and BLR accelerate the convolution and spare the memory space as well,which are independent of the acceleration hardware.MSM is more suitable for convolution with small kernel size.And there is no limitation of BLR.Finally,we segmented the lung area,and efficiently distinguished the benign nodules from the malignant ones.The innovation of the processing and recognition in this paper provide adjuvant therapy with new methods,which is of great significance.