Mahdi Daliri - Faculty of Electrical Engineering, K.N.Toosi University, Tehran, Iran
Hamid Abrishami-Moghaddam - Faculty of Electrical Engineering, K.N.Toosi University, Tehran, Iran
Maryam Momeni - Faculty of Electrical Engineering, K.N.Toosi University, Tehran, Iran
Sona Ghadimi

This paper presents an automatic segmentation method for neonatal skull, scalp and brain tissues using Magnetic Resonance (MR) images. The analysis and study ofneonatal skull, scalp and brain tissues in MR images is of great interest due to its potential for studying early growth patterns,morphological changes in neurodevelopmental disorders and EEG/MEG source modeling. Our method beneficiated from fuzzy support vector machine (SVM) using gray level features and probabilistic model parameter; first, a probabilistic model was generated using a set of accurately manually segmentedtissues. Then for each voxel the gray level features were extracted by the voxel intensity and its 3D neighborhoods.Finally, a Fuzzy SVM algorithm was applied for the segmentation using training sample, gray level and probabilistic model features. Considering chemical shift artifactin this golden standard and utilizing 6 newborn MR images, the segmentation algorithm was done and its results werecompared with manual segmented data. The average similarity indices for the scalp, skull and brain segmented regions wereequal to 84%, 63% and 88% for the test data, respectively.

Fuzzy Support Vector Machine (FSVM),Region of Interest, Probabilistic model feature, Gray- Level feature, Magnetic Resonance Image