Research On Lung Medical Image Detection And Segmentation Methods For Disease Diagnosis And Treatment

The primary step in the diagnosis and treatment process of lung diseases is to conduct imaging examinations,and doctors use the information provided by lung medical images as assistance to make disease diagnosis results.During this process,different clinical experiences can lead to biased diagnostic results,which can lead to misdiagnosis,misdiagnosis,or missed diagnosis.Computer assisted lung medical image detection can serve as an auxiliary tool for doctors,and the detection results can assist doctors in efficient diagnosis of lung diseases;If there are pulmonary nodules or tumors in the detection results,computer-assisted segmentation methods can be used for segmentation.The segmentation results combined with the imaging features of the tumor can assist doctors in making preliminary diagnoses of benign and malignant tumors,as well as measuring the growth rate of the tumor;If the doctor determines that surgical treatment is necessary,computer-assisted 3D segmentation and reconstruction methods can be used to obtain 3D models of lung tumors and lung parenchyma,thereby assisting the doctor in preoperative planning.In summary,although lung medical image detection and segmentation methods for disease diagnosis and treatment can serve as auxiliary tools for doctors,improving the performance of the methods to achieve higher accuracy and accuracy remains a challenging task.This article conducted research on disease classification and detection based on lung medical images,achieved high accuracy detection of five common lung diseases;We had conducted research on automatic tumor segmentation based on lung medical images,achieved automatic and high-precision segmentation of lung tumors;We had conducted research on the application of detection and segmentation in auxiliary diagnosis and treatment,achieved the functions of auxiliary diagnosis of lung diseases,tumor segmentation,and preoperative planning.Specifically:A disease classification and detection model based on lung medical images had been proposed.In response to the problem of overfitting when facing relatively complex VGG16(Visual Geometry Group,VGG)neural networks with fewer lung medical datasets,transfer learning was used to improve and adjust the fully connected layer of VGG16.This solves the conflict between fewer lung images and complex VGG16,while preserved the ability of VGG16 to extract image features and increased the ability to classify lung images.Improved the cross entropy loss function in VGG16 and increased clustering performance.Combine the improved VGG16 and Gradient Boosting Decision Tree to construct a single branch structural model,and then combine five single branch structural models with a weighted voting algorithm to construct a classifier model.In the experiment,the proposed classification detection model was trained and tested using the Ches X-ray14 dataset.The training accuracy of the model remained between 87% and 90%,while the overall testing accuracy remained between 89% and 90%.The classification accuracy and robustness of the model were better than the best published results.A 2D automatic segmentation method for lung tumors based on improved region growth algorithm was proposed for the diagnosis of lung tumors.Combined the prior information of lung tumor with the maximum between-class variance(OTSU)algorithm,a lung tumor location method was constructed,and the automatic selection of initial seed points of the region growth algorithm was realized.After the seed point expansion,the automatic update mechanism of growth restriction and threshold was established.Finally,the combination result of multi-point growth was taken as the final segmentation result.The proposed method realizes automatic segmentation of lung tumors while eliminated a large number of human interactions.In the experiment,the average Dice coefficient obtained by this method was 0.936 ±0.027,and the average Jaccard distance was 0.114 ± 0.049.The segmentation performance was bettered than the current popular segmentation algorithm.A 3D automatic segmentation method for lung tumors based on CT images was proposed for 3D reconstruction and preoperative planning in 3D diagnosis and treatment.The automatic selection of seed points for adjacent slice images utilizes the similarity between adjacent slice images and the established dynamic constraints.The automatic selection method for seed points under normal and abnormal conditions had been set,and the automatic selection of seed points under abnormal conditions was achieved through dynamic constraints.By combined the automatic updating mechanism of threshold,automatic 3D segmentation of lung tumors was achieved.At the same time,the K-means clustering algorithm was used to achieve3 D segmentation of lung parenchyma by combined image preprocessing with morphological image post-processing.In the experiment,these two methods could smoothly and reliably achieved 3D segmentation of lung tumors and lung parenchyma.Apply the proposed detection and segmentation method to the auxiliary diagnosis and treatment of lung diseases.The proposed classification detection model had achieved auxiliary diagnostic function;The proposed automatic segmentation method for lung tumors had achieved tumor segmentation function;The proposed 3D segmentation method combined with reconstruction methods could obtained 3D models of lung tumors and lung parenchyma,and achieved preoperative planning function of thoracoscopic surgery through added interactive response and adjustable simulation line segments.Doctors can continuously adjust the simulated surgical approach through adjustable simulation line segments,thereby avoiding areas where danger may occur during the surgical process and achieving more efficient preoperative planning.