Abstract
Image segmentation is an important task involved in different areas from image processing to image analysis. One of the simplest methods for image segmentation is thresholding. However, many thresholding methods are based on a bi-level thresholding procedure. These methods can be extended to form multi-level thresholding, but they become computationally expensive because a large number of iterations would be required for computing the optimum threshold values. In order to overcome this disadvantage, a new method based on a Shrinking Search Space (3S) algorithm is proposed in this paper. The method is applied on statistical bi-level thresholding approaches including Entropy, Cross-entropy, Covariance, and Divergent Based Thresholding (DBT), to achieve multi-level thresholding and used for intracranial segmentation from brain MRI images. The paper demonstrates that the impact of the proposed 3S technique on the DBT method is more significant than the other bi-level thresholding approaches. Comparing the results of using the proposed approach against those of the Fuzzy C-Means (FCM) clustering method demonstrates a better segmentation performance by improving the similarity index from 0.58 in FCM to 0.68 in the 3S method. Also, this method has a lower computation complexity of around 0.37s with respect to 157s processing time in FCM. In addition, the FCM approach does not always guarantee the convergence, whilst the 3S technique always converges to the optimum result.
Highlights
An important component of Computer Aided Detection (CAD) systems based on Magnetic Resonance Imaging (MRI) is segmentation
It still has sensitivity to the number of predefined clusters, and initial values of the center points or membership functions. This dependency causes occasional non-convergence of the algorithm, where it get stuck in local minima points. We introduce another new multi-level thresholding method based on a Shrinking Search Space (3S) scheme, which is implemented on different statistical bi-level thresholding methods including the Entropy method, the Covariance method, the Cross Entropy method, and the Divergent Based Thresholding method (DBT) method
And 9, the 3S+DBT method, as the best choice of the statistical thresholding methods, is compared with the Fuzzy C-Means (FCM) method which is a commonly used clustering algorithm, on one sagittal brain MR image as well, without applying region growing algorithm for non-brain tissue segmentation task. This comparison proves again better segmentation of the 3S+DBT method with respect to the FCM method, and much lower processing time of the 3S method (0.78s) than that of the FCM method (423.99s). (Figure 8a-c) show the segmented images of the original image shown in Figure 8a based on the 3S+DBT method, in a binary format, while, (Figure 9a-c) show the same results from the FCM method
Summary
An important component of Computer Aided Detection (CAD) systems based on Magnetic Resonance Imaging (MRI) is segmentation. Image thresholding is the process of classifying image gray values into two or more levels. This method has been used on T1-weighted MRI images to isolate brain tissues such as skull, Gray Matter (GM), White Matter (WM), and Cerebro Spinal Fluid (CSF). We have evaluated different approaches in the category of histogram entropy class. These methods include Entropy method [2], Covariance method [3], Divergent Based Thresholding method (DBT) [4], and Cross-Entropy method [5], which are investigated here
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
