International Congress Series 1256 (2003) 1309
An accurate and efficient hybrid approach for near real time 3D brain segmentation Lixu Gu a,b,*, Terry Peters b a
Department of Computer Science, Shanghai Jiao Tong University, 1954 Huashang Road, Shanghai 200030, PR China b Imaging Laboratory, Robarts Research Institute, London, Canada
Received 21 March 2003; received in revised form 21 March 2003; accepted 21 March 2003
1. Introduction Image Segmentation is one of the key areas in image processing and computer vision, with numerous applications including image-guided and robotically assisted surgery, etc. [1,2]. Real time 3D segmentation of the brain from MR scans is a challenging problem due to the huge size of source datasets. Here we propose a new 3D hybrid segmentation algorithm, which performs rapid brain segmentation using a multistage approach. 2. Approach Our 3D segmentation algorithm includes four steps: 1. 2. 3. 4.
Reduce the connectivity between regions by morphological recursive erosion. Perform initial evolution of a front by a fast marching method. Refine the contours by morphological reconstruction algorithm. Recover the lost data elements from stage 1 by the recursive dilation method.
3. Experimental results The approach is tested on 10 T1- or T2-weighted neurological MRI images, and validated by a simulated 181217181 voxel T1-weighted MRI volume (from MNI’s ‘‘Brainweb’’ site http://www. bic.mni.mcgill.ca/brainweb/) for which ‘‘truth’’ is known with respect to its WM, GM, CSF components. The other nine T1 or T2 weighted neurological MRI datasets were 256256124 volumes. The mean execution time for this algorithm, when run on a 550 MHz Dual PIII-based PC, was 38 s. 4. Conclusion The robustness of the hybrid segmentation approach was tested by 10 MRI brain datasets. Apart from the model-brain image all of the employed datasets were typical diagnostic-quality clinical images. These measurements revealed that the algorithm achieved a ‘‘similarity index’’ of 0.963. The average computing time was 38 s. References [1] A.F. Frangi, W.J. Niessen, M.A. Viergever, Three-dimensional modeling for functional analysis of cardiac images: a review, IEEE Trans. Med. Imag. 20-1 (2001) 2 – 25. [2] M.A. Audette, T.M. Peters, Level-set segmentation and registration for computing intrasurgical deformations, Proc. SPIE 3661 Medical Imaging, (1999) 110 – 121. * Corresponding author. Department of Computer Science, Shanghai Jiao Tong University, 1954 Huashang Road, Shanghai 200030, PR China. Tel./fax: +86-21-62932902. E-mail address:
[email protected] (L. Gu). 0531-5131/03 D 2003 Published by Elsevier Science B.V. doi:10.1016/S0531-5131(03)00340-6