A new variational method for selective segmentation of medical images

Abstract Selective segmentation aims to separate a subset of target objects or regions of interests in an image. It is widely used in medical image analysis for some specific tasks such as extracting anatomic organs or lesions. However, selective segmentation of medical images is usually challenged by their limited imaging quality. In this paper, we propose a two-phase selective segmentation method. The first phase is a preprocessing step, which aims to reduce influence of noise or cluttered background on segmentation. The second phase performs selective segmentation on the preprocessed image. For the first phase, we propose a new image smoothing model which can effectively reduce noise or intensity inhomogeneity inside objects while retain edges of the original image. Moreover, the proposed model has attractive mathematical and physical properties, in that it has one single optimal solution. For the second phase, we propose a modified Gout’s active contour method, which can obtain targeted objects more efficiently and accurately. Our main contribution is the new image smoothing model, which can effectively attenuate complicated background but preserve edges of targeted object. Extensive experiments on real medical images show that, our smoothing model can greatly facilitate the second phase, and our method can significantly improve some existing related methods in terms of either visual assessment or quantitative evaluation.