Research On Attention-Guided Deep Learning Algorithm In Chest X-ray Tuberculosis Detection

Tuberculosis is the most deadly infectious disease in the world.Early detection and diagnosis is a key step in the treatment of tuberculosis.Existing computer-aided detection systems for chest X-rays for the diagnosis of tuberculosis have been preliminary studied.With the rise of deep neural networks in recent years,the use of deep convolutional neural networks to solve the above problems has a very large Strengths and potential.The attention mechanism in deep learning is modeled on the principle of human information acquisition,and filtering irrelevant information can greatly improve the performance of the network.This article proposes an improved deep convolutional neural network for chest X-rays,introduces the attention mechanism,and studies the focus detection and classification algorithm for tuberculosis.main tasks as follows:(1)Collected 4,990 chest orthotopic X-ray tuberculosis datasets from outpatient clinics of three different hospitals in Jilin,Guangzhou,and Shanghai.Among them,there were 1,500 healthy control images that did not contain the disease,and 2506 images of tuberculosis.There are 984 images of non-tuberculosis but possibly other lung diseases.When preprocessing the image,the DICOM format image is converted into a JPG format image,and the single-channel image is converted into three channels according to the input requirements of the deep learning network.Finally,the bilinear interpolation algorithm is used to unify the image size to avoid the experimental results.Make an impact.(2)The structure of CNN,the principle of attention mechanism,and the benchmark network ResNet are studied.Based on the analysis of several representative convolutional neural network algorithms,the convolutional neural network algorithm for tuberculosis detection is studied.Supervision method and attention guidance mechanism,a deep learning convolutional neural network model based on attention mechanism is constructed,and the skip connection operation in the residual structure is introduced into the attention mechanism(3)The deep residual convolutional neural network based on attention mechanism is appliedto the classification of tuberculosis,which solves the problem of large differences in the shape of tuberculosis lesions and results in poor detection of the lesions.Cost and improve detection accuracy.The experimental results show that compared with the four classic classification algorithms,the proposed method is better than the existing algorithms in classification accuracy and robustness.(4)Use the attention mechanism to locate the area of ??tuberculosis lesions,generate visualized significant activation maps and heat maps,and verify the effectiveness of the attention mechanism on tuberculosis detection tasks.The experimental results show that the network with attention mechanism can use the correlation between space and channels to assign different weight values ??to the feature map,effectively suppress the interference of useless information on the classification results,and improve the automatic classification effect.This paper improves the deep residual convolutional neural network,introduces an effective attention mechanism,and realizes the detection and classification of chest X-ray tuberculosis.It can help reduce the workload of doctors and provide a feasible intelligent solution for large-scale automated screening for tuberculosis.