An approach to eliminate false positive lung cancers based on state variance

This paper presents a new approach to eliminate false positive lung cancers based on state variance, according to the variance of morphologic features between adjacent CT slices. First, the regions of interest (ROI) in CT images are selected by dual fast marching method, and the morphologic features of ROI like area and centroid are extracted; then, the layer features are extracted with morphologic features by state variance, and another new method is used for modeling for the cancers and non-cancers respectively based on layer features, eventually, false positive lung cancers are eliminated by calculating the maximal similarity between uncertain sample and models. We take the experiment by actual chest CT images, eliminate 68 false positive lung cancers in 100 samples which include 72 suspect lung cancers, obtain a detection rate of 94% for false positive lung cancers, and validate the effectivity of this method.