Research On X-ray Dental Image Aided Diagnosis Method Based On Fusion Neural Network And Attention Mechanism

With the improvement of people’s requirements for health quality,dental health has become an important indicator of the physical and mental health of residents in a certain country or region.As an image with pathological features,medical images have extremely important reference value in the process of medical diagnosis for patients.X-ray photography of teeth plays an irreplaceable role in clinical diagnosis,treatment and surgery.The daily diagnosis workload of doctors is very large.In order to assist doctors in medical diagnosis and reduce the burden of diagnosis.Therefore,accurate localization of dental disease focus is a crucial technology for dental clinical diagnosis and treatment planning.In this paper,we propose a new end-to-end fusion network for the detection of tooth lesions,and propose a new network based on the fusion neural network and attention mechanism,which is mainly based on the X-ray dental image aided diagnosis network.It is mainly composed of a backbone network and attention mechanism module,with a pre-training structure and multi-scale feature aggregation module.The main work of this paper is as follows:(1)In order to help dentists diagnose diseases,and reduce the burden of doctors,this paper constructs a convolutional neural network for dental X-ray detection to find suitable medical data sets.Throughout the application of convolution neural network in medical images,the depth of neural network should not be too deep.In order to solve these problems,this paper selects the convolutional network with pre-training structure as the backbone network,adds the feature fusion module to the original network,and innovates the network to further improve the network performance.(2)In order to achieve more accurate heat map regression,we use a new loss function,the exponential weighted central loss function,which is improved on the basis of the weighted central loss function in this experiment.This function in this paper can suppress the distant loss,and pay attention to the loss near the marker.The best epoch value of the loss function is determined through the ablation experiment,and further optimize to reduce the computational loss of the tooth X-ray image detection network.(3)A new fusion network based on traditional convolution neural network and attention mechanism algorithm.In this paper,the fusion algorithm with attention mechanism algorithm is based on the original convolution neural network.By comparing the results of several groups of experiments,we decided to insert the attention mechanism in each upsampling stage for network fusion.We compare the detection effect of dental radiography detection network on the same data set after the backbone network has fused multi-scale feature module and attention mechanism module respectively.The experimental results mainly compared the detection accuracy,and mean relative error(MRE)when detecting lesions of different sizes.Through the comparison of the final experimental results,this paper concludes that the accuracy of the new network integrated with the attention mechanism and the traditional network has been greatly improved in image recognition tasks,and the improved accuracy varies from 0.95% to 1.84% in the detection of lesions with different radius sizes.The final result is that the network tested in this paper is better than several popular networks today,both in terms of average relative error(MRE)and success detection rate(SDR).