| Medical image segmentation is a complicated and challenging task in medicalimage processing, especially for human brain magnetic resonance image (MRI). Theadvantage of MRI imaging technology is noninvasive to human body detection, aswell as the high resolution of the soft tissue. Therefore, MRI imaging is a keytechnology to the study of the brain function, pathological and anatomic.In recent years, many scholars are studying the segmentation technology of thebrain MR image. But the MR images are affected by partial volume effect, noise andbias field easily, which bring a lot of difficulties to the segmentation of brain MRimages. Among the brain image segmentation algorithms, the fuzzy clusteringtechnique has been widely applied. The FCM algorithm is a kind of fuzzy clusteringalgorithms. It obtains better results on brain image segmentation.In this paper, the bias field and the noise in the MR image are both taken intoaccount. We introduce the bias field model to deal with the bias field in the image.The bias field in the MR image are smooth and changing slowly. The non-uniformfield is modeled as a multiplicative gain, and we transform it into additive bias fieldby logarithmic transformation. In order to correct the bias field, we use this model tocompensate the bias field in the image. For the noise in the image, the local andnon-local neighborhood information are taken into account. The objective function ismodified to be affected by the bias field, local and non-local information. Theexperiments prove that the non-local distance function keeps the image edgeinformation completely, at the same time, the local distance function makes thestructure and the detail information in the image preserved. We conduct extensiveexperiments and compare our method with different types of FCM extension methodson simulated MR images. The results show that our proposed method can deal withthe bias field and noise effectively and outperforms other methods. |
PDF Full Text Request