关键词: 疑似肺结节, 法向量叠加, 局部形状约束, 凸凹性特征, 医学图像处理

Abstract: An improved boundary normal vector overlap algorithm based on local shape constraint and adaptively defined normal vector magnitude is proposed to detect suspected pulmonary nodules. The original region of interest(ROI) is segmented with adaptive thresholding method in pulmonary parenchyma, then the local convex-concave shape feature is gained for every pixel of initial ROI boundaries to satisfy the calculations of normal vector direction and adaptively defined normal vector magnitude of every convex pixel on the initial ROI boundaries. Overlapping the normal vectors, the locally maximum overlap can be chosen so as to detect the suspected round pulmonary nodules of different sizes. The restriction due to the local convex-concave shape feature should be taken into account before overlapping the normal vectors, thus simplifying the calculation of overlap. The adaptively defined normal vector magnitude is available to get rid of the limitation of fixed-size nodules as the result of detection. Experiment results indicated that the improved algorithm can detect suspected pulmonary nodules effectively and applicable to the nodules of any sizes.