(Sunday, 7/29/2018) 3:00 PM - 6:00 PM
Room: Exhibit Hall
Purpose: To develop an automatic segmentation algorithm based on clustered texture images and assess the accuracy.
Methods: In this study, we proposed an automatic segmentation method based on clustered texture image. First, the algorithm computes the texture image from original images, including Gray level co-occurrence matrix-based Contrast images, Correlation images, Energy images and Homogeneity images. The value of pixel in four images is the value of texture feature calculated from its 5×5 or 7×7 sliding window. Especially, for the pixels in the edge of the original image, the same row or column was added in the periphery of the edge. Then, k-means cluster algorithm was implemented to group the pixels in texture image into a binary mask, including the foreground and the background. Meanwhile, morphological operation, including hole filling and open-close operation, was used to smooth the mask images. Then, the segmentation results were calculated by the mask. The accuracy of this method was evaluated by Dice’s Similarity Coefficient (DSC) which calculated by segmentation result and ground truth.
Results: Five images were used to verify the proposed algorithm. For the texture images computed with 5×5 sliding window, the mean DSC value was 0.963. However, the mean DSC value for 7×7 sliding window was 0.914. The results show that the segmentation results for Homogeneity image have more accuracy results.
Conclusion: Our automatic segmentation algorithm shows good accuracy with 5×5 sliding window and Homogeneity image compared with ground truth. The algorithm maybe able to as an automatic segmentation tool in medical image analysis, such as liver or lung tumor segmentation.
Keywords
Quantitative Imaging, Segmentation, Contour Extraction
Taxonomy
Not Applicable / None Entered.
Contact Email
