Wednesday, September 30, 2020 Home   | Search
Viewing Abstract

Page: 422-432 ------------------------------> Last_modified :7/11/2017 12:11:00 PM

Title: A Comparative Study on the Automatic Segmentation of Colon in Ct Colonography Using K-Means and Fuzy C-Means Clustering
Authors: K. Gayathri Devi ; R. Radhakrishnan
Aff: Anna University, Dr.N G P Institute of Technology, Department of Electronics and Communication, Associate Professor, Coimbatore, India.
Author Email:
Keywords: Computed Tomography; Image Segmentation; Histogram; Colonography; Unsupervised Learning; K-Nearest Neighbor; Clustering; Histogram; ITK Snap ,Sensitivity and Specificity
Abstract:Background: An approach for the automatic segmentation of colon which is filled with air and opacified fluid in CT colonography is presented. The presence of an air and partial volume effect is a challenging factor for the segmentation of colon. Therefore we have developed a new approach for the automatic segmentation of colon.Objective: To extract the segment of colon from all the slices of a particular dataset more precisely and to perform 3D rendering of segmented colon structures in order to improve the accuracy and sensitivity. Methods Used: In this paper, we use K-means clustering and Fuzzy C-means clustering, an unsupervised methods to identify and group the colon segments as clusters. Automatic colon segmentation from abdomen CT data starts with removal of the air portions that is outside the body of the subject, masking of the lungs should be done because they have the same intensity values as the colon segments which is a very difficult task, segmentation of colon segments from all the slices and finally 3D rendering of all the segmented colon segments. Conclusion: The above proposed methods were tested with 5 dataset of abdominal CT images downloaded from TCIA cancer imaging archived. The segmentation was validated by comparing the segmented output with manually segmented output using ITK snap. The three metrics accuracy, sensitivity and specificity were calculated. The proposed method shows an improvement in accuracy and sensitivity compared to the existing graph cut and level set methods.8>