Lung Lesion Extraction Using Histogram Binning Based Automatic Segmentation Approach

Abstract

Lung Lesion Extraction becomes the crucial part in the lung cancer diagnosis. The accurate segmentation of lung lesion from computerized axial tomography (CAT) scans is important for lung cancer diagnosis and research. A novel toboggan based growing automatic segmentation approach (TBGA) with a three-step framework is used for lung lesion segmentation. The initial seed point in the lung lesion was first automatically selected using an improved toboggan method for the subsequent 3D lesion segmentation. Then, the lesion was extracted by an automatic growing algorithm with multi constraints. Finally, the segmentation result was optimized by a lung lesion refining method. By using this lung lesion segmentation algorithm better performance will be obtained. The combination of TBGA and adaptive histogram binning, have similar or slightly better accuracy than previously obtained TBGA results on same-center training and evaluation. In conclusion, we believe that the novel HBBAS can achieve robust, efficient and accurate lung lesion segmentation in CT images automatically.