Threshold algorithm for pancreas segmentation in Dixon water magnetic resonance images

• Published in: 2017 13th International Conference on Natural Computation, Fuzzy Systems and Knowledge Discovery (ICNC FSKD) pp. 2367 - 2371
• Main Authors: Shan, Xiaoying, Du, Chaolin, Chen, Yufei, Nandi, Asoke, Gong, Xiaoliang, Ma, Chao, Yang, Panpan
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.07.2017
• Subjects: Biomedical imaging; Cancer; component; Dixon water magnetic resonance image; Image segmentation; Level set; Magnetic resonance imaging; Manuals; morphological method; Otsu method; Pancreas; Pancreas segmentation
• DOI: 10.1109/FSKD.2017.8393142

Pancreas segmentation is crucial for a computer-aided diagnosis (CAD) system to provide cancer detection and radiation therapy of pancreatic cancer. Because of anatomically high-variability between subjects, achieving high accuracies in pancreas segmentation remains a challenging task. In this work, based on Otsu threshold method and morphological method, we first proposed a segmentation pipeline for pancreas, using Dixon water magnetic resonance image (MRI) data from five healthy volunteers. The threshold method was used to obtain the approximate outline of the pancreas, and the morphological method was used to separate the pancreas from the surrounding tissues. The segmentation results were compared with manual contours using Dice Index (DI) and we achieved DI: 0.80 ± 0.08 which was better than the level Set Methods (LSMs) DI: 0.64 ± 0.08. The proposed method was simple and easy to integrate with the Medical Imaging Interaction Toolkit (MITK) workbench, so it provided an efficient and simple segmentation method for processing large clinical datasets.