Algorithm Research On Organ Segmentation And Tuberculosis Detection Of Chest X-ray

For curable and controllable lung diseases such as tuberculosis,early screening and diagnosis play a vital role.Chest X-ray is an important imaging method for pulmonary tuberculosis screening,but due to the different characteristics of multi-category of tuberculosis lesions,it is difficult to quickly and accurately analyze.Computer-assisted detection can help radiologists screen chest X-rays more accurately.The main research contents of this article are:(1)Aiming at the problem that multiple organs in chest X-rays overlap and cause feature missing and difficult to segment,a multi-organ segmentation algorithm with global information is proposed.In the encoding-decoding network architecture,a module for multifeature fusion is proposed by the combination of multi-scale input block,hypercolumn block,and multi-receptive field block,and a module for multi-feature enhancement is proposed by the combination of attention gating block,and concurrent spatial and channel squeeze and excitation block.The FGDL loss for segmentation of ribs,clavicle,heart,and lung is proposed.In this study,the Dice value on the bone and JSRT segmentation datasets increased by 1.83 and 0.93 percentage points on average,which is better than other segmentation algorithms.(2)Aiming at the problem that a single chest X-ray contains multi-category tuberculosis lesions and is difficult to accurately classify,a multi-category tuberculosis lesion detection algorithm is proposed.The algorithm focuses on the analysis of specific lesion areas,introduces the learning scalable feature pyramid structure into the faster region-based convolutional neural network(Faster RCNN),and uses reinforcement learning to mine indistinguishable samples during the training process to reduce the number of false positive lesion detections,the detection accuracy of multi-category tuberculosis lesions and nodular small area lesions increased by 2.78 and 5.23 percentage points,respectively.Based on the results of the tuberculosis lesion detection model,this paper proposes a tuberculosis classification rule,which is superior to other tuberculosis classification algorithms on two public datasets.(3)The automated quantitative analysis of chest X-rays,based on the results of multi-organ segmentation,the automatic measurement of the width of the mediastinum,the area and density of the lung area,and the length of the segmentation line of the lung area was realized by accurately positioning the relevant measurement positions,and the lung area is automatically partitioned.Then the results are measured and visualized on the public dataset,the measured values are analyzed,and the role of the measured index values in the auxiliary diagnosis of lung diseases is explored.The experimental results show that,based on the results of multi-organ segmentation,this article performs multi-category tuberculosis lesion detection and automated chest X-ray measurement,which improves the accuracy of tuberculosis detection and realizes automated quantitative analysis of multiple clinical indicators,which can be used to assist public health service providers to screen for tuberculosis in areas where tuberculosis is endemic.It can also be used to help radiologists diagnose lung diseases and reduce their workload.