Our algorithm was implemented in Matlab 7.0 on a 2.1 GHz Intel Core 2 Duo T6500 PC with 2 G memory. We tested our algorithm on 10 normal T1 MR brain images obtained from Hospital with informed consent from all subjects. Each volume consisted of 176 slices, 448 × 512 pixels per slice. The slice resolution is 0.5 × 0.5 mm2 and the slice thickness is 1 mm. We used the following default setting of the parameters: σ = 1.5, time step Δt = 1, ρ = 1, v0 = 5. Figure shows the final results on four sample images displayed in three orientations. To measure the extraction accuracy of our algorithm, 10 normal MRI brain data sets and the corresponding manual segmentations were obtained from the Internet Brain Segmentation Repository (IBSR) developed by the Centre for Morphometric Analysis (CMA) at Massachusetts General Hospital. Each volume has around 65 coronal slices, with 256 × 256 pixels per slice. The slice resolution is 1.02 × 1.04 mm2, and slice thickness is 3.1 mm. Figure shows the final results of our method on eight sample volumes as well as manual segmentation results displayed in coronal orientation. We also computed the sensitivity, specificity, Jaccard index, Dice index and the FP_Rate of our segmentation results using the manual segmentation results provided by the IBSR (shown in Table ).
Brain extraction results of four sample normal adult datasets downloaded from the IBSR shown in coronal orientation. Columns from left to right: raw image, brain extraction results of our method and manual extraction provided by ISBR.
Performance comparison of BET, MLS and the proposed method for multiple indices using the IBSR data sets
In addition, we compared our algorithm to two popular brain extraction methods: BET and MLS, using the 10 normal data sets from the IBSR, and the segment results are illustrated in Figure . The programs of the BET and MLS were downloaded from their respective WebPages. BET was run on Ubuntu 9.1(a popular Linux distribution). We first ran BET with its default parameters: Fractional Intensity Threshold (FIT, default 0.50) and Vertical Gradient (VG, default 0.0)to segment one training volumes. Unfortunately, with such parameters, BET did not work well, always leading to lots of non-brain tissues(eye, optic nerve, neck.....) being included. Therefore, we did not use the default parameters and by training, the parameters that could provide the best extraction results were applied to all 10 brain volumes (FIT = 0.65, TG = -0.15, robust brain centre estimation). MLS was implemented on the Windows platform using the Java programming language and our algorithm was implemented on the Windows platform using Matlab. The three extraction algorithms were performed on the same hardware platform. Figure shows the comparison results of BET, MLS and the proposed method as well as manual segmentation displayed in coronal orientation. Table lists the comparison result of BET, MLS and the proposed method for multiple indices using the same IBSR data sets. Generally, the larger the sensitivity coefficient, the more accurate the segmentation results. But for a special case, if the segmentation is always conservative, which rather includes lots of non-brain tissues than avoids removing any brain tissue, then FN equals 0 and the sensitivity always equals 1. So an algorithm with larger sensitivity not always have more accurate result. An accurate and robust algorithm must have good performance for multiple indices. Table shows that each algorithm has its advantages and disadvantages. First, the proposed method is superior to BET and MLS with respect to FP_Rate and Specificity. Second, in our experiments, BET is always conservative and often includes some non-brain tissues, leading to the best sensitivity but worst specificity and FP_Rate coefficients. The reason BET had such performance, perhaps because it was more important to avoid removing brain tissue than to remove all non-brain tissues for clinical application. Third, MLS is superior to the our method with regard to sensitivity and has similar performance on Jaccard and Dice indices. Generally speaking, when compared to BET, our method does not include too many non-brain tissues and need not tune many input parameter, so it is accurate and simple to use. Compared to MLS, due to our automatic initialization method, our method is more efficient. So as an automated and simple brain extraction tool, our method can accurately extract brain tissue with high efficiency.
Figure 5 Comparison results of BET, MLS and the proposed method using four slices of normal T1-weighted MR brain images shown in coronal orientation. Columns from left to right: raw image, brain extraction results of BET, MLS, our method and manual extraction (more ...)