Research Of Benign And Malignant Diagnosis Methods For Pulmonary Nodules Based On Three Dimensional

In lung cancer computer-aided diagnosis system,pulmonary nodule detection and different types of nodule classification diagnosis are the core of the whole computer-aided diagnosis system,and are also the hotspots and difficulties of the current auxiliary diagnosis system.For the pulmonary parenchyma,there are more vascular trachea and other tissues in the lung parenchyma region.The overall performance is more complicated.It is difficult to accurately identify and diagnose lung lesions in two-dimensional CT images.In this paper,three-dimensional lung parenchymal data is constructed based on sequential two-dimensional CT images.Pulmonary lesions are detected in three-dimensional space and the characteristics of three-dimensional lesions are analyzed.Finally,the suspected lesions are classified and diagnosed.The specific research content of this paper is as follows:(1)Pulmonary nodule detection method based on 3D Shape Index-like spherical filter.Faced with the cross-sectional vascularity in the lung parenchyma region and the complex effects of other tissues such as trachea,the accuracy of nodule detection in two-dimensional CT images was low,and there were many other tissue regions with high false positives in the test results.To solve this problem,this paper proposes a nodule detection algorithm for three-dimensional spherical filter.First,the sequence CT image raw data is preprocessed to obtain a three-dimensional lung parenchymal data region.According to the three-dimensional characteristics of blood vessels and nodules,a three-dimensional mathematical model was constructed to use the three-dimensional model features of nodules and blood vessels to improve the three-dimensional shape index,and to construct a spherical filter nodule detection function for three-dimensional lung parenchymal data nodules.The suspected nodule coordinate set was detected and the area of confidence connection was increased to realize the segmentation of the three-dimensional pulmonary nodule,and nodule features were extracted for further diagnostic classification.The spherical-like filter constructed in this paper has the following two innovations:1.Constructing nodule detection function using shape index;2.Improve shape index so that it can be applied to three-dimensional lung parenchyma to perform three-dimensional nodule detection.Experimental analysis shows that the lung nodule detection algorithm proposed in this paper can detect different types of nodules more comprehensively and to a certain extent can reduce false positives.(2)Diagnosis of benign and malignant pulmonary nodules based on multi-dimensional feature fusion.In the nodule feature extraction,single features can not fully reflect the characteristics of nodules,and simply the feature vector composed of multi-dimensional features in series can not fully use the effective information of multi-dimensional features.To solve this problem,this paper proposes a multi-dimension feature fusion classification diagnosis algorithm to identify and diagnose different types of nodules.First,a three-dimensional texture model is constructed,and the histogram statistics of the three-dimensional global angular direction and the three-dimensional shape index are calculated,and the extreme value calculated by the three-dimensional shape index is used to improve the local SIFT feature.The multi-view features constructed in this paper are extracted from the training data,and feature vectors are classified using different kernel functions.Finally,decision-level fusion is used to perform fusion optimization on different feature diagnosis results to obtain classification diagnosis results for different types of nodules.The multi-feature fusion lung nodule diagnosis method constructed in this paper has the following two innovations:1.Three-dimensional extension of texture features,and construction of a global angle direction and shape index histogram;2.Using the three-dimensional shape index extreme points to improve the local SIFT features,densely extract key points on the three-dimensional nodal body data,and perform comprehensive analysis on three-dimensional volumetric data samples.The experimental analysis shows that the multi-feature fusion strategy proposed in this paper has higher classification accuracy,can fully utilize the extracted features,and has higher accuracy and sensitivity than the individual features.