Research On Attention Mechanism Based Deep Learning Method For Medical Image Analysis

Medical image is an important basis for assisting doctors in clinical diagnosis.Manual analysis of medical images is subjective,and it is difficult to ensure the consistency of results.Therefore,it is of great significance to use computer technology to analyze medical images automatically.Segmentation and classification are two important steps in medical image analysis.The attention mechanism enables the model to focus on important information in the region of interest.Based on this,this project introduces the attention mechanism into the medical image segmentation and classification tasks and proposes the two-branch network segmentation method of spatial attention constraint graph based on difficulty perception and the two-level convolutional neural network classification method respectively.Due to the influence of patients’ own physiological conditions and shooting environment,medical images have certain differences.The existing deep learning methods mostly use the same network to segment images,which is difficult to improve the segmentation accuracy and efficiency at the same time.In addition,the common deep learning segmentation methods ignore the relationship between pixels when acquiring the probability map of pixel categories,which limits the improvement of segmentation accuracy.To solve these two problems,a two-branch network of spatial attention constraint graph based on difficulty perception is proposed in this paper,which mainly consists of difficulty grading module and two-branch segmentation network.The difficulty classification module can automatically classify images.Based on the classification results,the model can segment images adaptively,thus improving the efficiency of the segmentation model.In order to process the difficult image,the difficult image segmentation branch is introduced into the two-branch network structure.In this branch,a new framework of spatial attention constraint graph is introduced.Firstly,spatial attention is introduced to obtain the robust discriminative characteristics of gray scale changes.Finally,spatial attention is embedded into the image frame to further improve the segmentation performance of difficult images.Experiments on ultrasonic images of breast cancer axillary lymph nodes have proved that the proposed method is superior to the classical deep learning segmentation method,especially in difficult images,achieving more significant performance improvement.In addition,the proposed method achieves both improved segmentation accuracy and efficiency.In some complex medical image classification tasks,most of the methods often extract features based on the region of interest(ROI),but this kind of methods often cause the loss of some important information.In addition,due to the complexity of images,it is difficult to extract sufficient discriminative information using only a single network.To solve the above problems,this paper proposes a two-level convolutional neural network for attention.The network structure mainly includes a multi-branch local attention feature extraction module and a global attention module.The local attention module includes parallel regional convolution and channel feature recalibration.Regional convolution is used to learn the features of several important regions.Channel feature recalibration is used to learn the importance of each important region feature.This module can simultaneously learn the features of the region of interest and the background region,and give high attention to the important features.The global attention module can use multiple branches to learn diverse effective features from a global perspective,and use channel features to re-calibrate and assign weights to the learned features,which further improves the feature differentiation and thus the classification accuracy.Experimental results on OCT images of choroidal neovascularization in macular area show that the proposed method is superior to the classical deep learning classification method.In this paper,attention mechanism is integrated into the process of medical image segmentation and classification,which enriches the technical means of using attention mechanism.The research of this paper runs through the key process of clinical diagnosis and treatment,and helps to improve the effect of clinical diagnosis and treatment.