The Research Of Detection Algorithm For Ground Glass Opacity

For years, automatic detection of lung nodules has puzzled medical experts, manyexperts and scholars have researched depth and come out a lot of detection methods,especially lung parenchyma division has also been made some significant results. But atpresent pulmonary nodules detection system in the domestic and overseas is aimed at solidlung nodules and the research of Ground Glass Opacity is very little. Although someoverseas scholars have proposed several algorithms for Ground Glass Opacity in recentyears, but common problems that the sensitivity and specificity are not high. Domesticresearch is nearly blank for Ground Glass Opacity, so this paper puts forward a kind ofdetective method for Ground Glass Opacity. The main work of this paper is following.Firstly, I segment lung area which contains Ground Glass Opacity. Because lungnodules must be located in the pulmonary parenchyma, the first step is the segmentation ofpulmonary parenchyma, so as to improve the efficiency of detection of lung nodules. Thispaper considers synthetically several steps of automatic segmentation of lung parenchyma,namely, separating Left and right lung, eliminating trachea/bronchial, repairing the edge ofthe lung.Secondly, regarding the problem that Ground Glass Opacity present light fuzzyshadow and it is not easy to be detected, putting forward a method based on AdaptiveNonlinear Filter (Adaptive Nonlinear Filter, AN) and band pass Filter (Band passFilter) to detect the candidate area of Ground Glass Opacity. The first affiliated hospitalof China medical university has35sets CT image including17Ground Glass Opacity,residual1; LIDC database has50sets CT image including21Ground Glass Opacity,residual2;the average residual rate of the two parts is7.89%.Thirdly, in view of the special gray and morphological characteristics of GroundGlass Opacity, I extract the22d characteristic data, and then the22d characteristic data willbe dropped to8d by PCA. The contribution rate of this8d is97%, exceeding therequirements of the principal component analysis.Fourthly, based on the8d characteristics which are extracted, Ground Glass Opacityare classified by momentum BP neural network. Recognition rate of the method is88%. In this paper, a lot of lung medical image are experimented and the final results showthat the algorithm can increase the sensitivity of detecting pulmonary nodules, andreducing the false positive rate.