Detection of Suspected Lung Nodular Lesions Based on Boundary Normal Overlap Method in Thoracic CT Images

摘要

An improved boundary normal overlap algorithm based on local shape constraints and adaptive distance projection is proposed for detection of suspected pulmonary nodular lesions in this paper. First, the initial region of interest is segmented from the lung parenchyma image using an adaptive threshold method. Then, the local shape constraint is calculated for each pixel on the boundaries of initial ROI. If the pixel local shape is convexity, normal of this pixel is computed, and line is projected along the direction of this normal. In the course of projection, the projection distance is confirmed adaptively. Last, suspected lung nodules can be detected with local maximum method. Local shape constraints enhance the capability of circle selection, and increase the effect of projection overlap. Adaptive distance projection overcomes the limitation of detection of fixed-size nodular lesions. Experiments has been done for synthetic images and clinical pulmonary CT images by the improved algorithm, experiment results indicated that the improved algorithm have higher sensitivity, and can detect suspected pulmonary nodules effectively.