Globally Optimal Tumor Segmentation in PET-CT Images: A Graph-Based Co-segmentation Method

• Published in: Information Processing in Medical Imaging Vol. 22; pp. 245 - 256
• Main Authors: Han, Dongfeng, Bayouth, John, Song, Qi, Taurani, Aakant, Sonka, Milan, Buatti, John, Wu, Xiaodong
• Format: Book Chapter Journal Article
• Language: English
• Published: Berlin, Heidelberg Springer Berlin Heidelberg 2011
• Series: Lecture Notes in Computer Science
• Subjects: Algorithms; Artificial Intelligence; Compute Tomography Image; Gaussian Mixture Model; Head and Neck Neoplasms - diagnosis; Humans; Image Enhancement - methods; Image Interpretation, Computer-Assisted - methods; Markov Random Field; Pattern Recognition, Automated - methods; Positron Emission Tomography; Positron Emission Tomography Image; Positron-Emission Tomography - methods; Reproducibility of Results; Sensitivity and Specificity; Subtraction Technique; Tomography, X-Ray Computed - methods
• ISBN: 3642220916; 9783642220913
• ISSN: 0302-9743; 1011-2499; 1611-3349
• DOI: 10.1007/978-3-642-22092-0_21

Tumor segmentation in PET and CT images is notoriously challenging due to the low spatial resolution in PET and low contrast in CT images. In this paper, we have proposed a general framework to use both PET and CT images simultaneously for tumor segmentation. Our method utilizes the strength of each imaging modality: the superior contrast of PET and the superior spatial resolution of CT. We formulate this problem as a Markov Random Field (MRF) based segmentation of the image pair with a regularized term that penalizes the segmentation difference between PET and CT. Our method simulates the clinical practice of delineating tumor simultaneously using both PET and CT, and is able to concurrently segment tumor from both modalities, achieving globally optimal solutions in low-order polynomial time by a single maximum flow computation. The method was evaluated on clinically relevant tumor segmentation problems. The results showed that our method can effectively make use of both PET and CT image information, yielding segmentation accuracy of 0.85 in Dice similarity coefficient and the average median hausdorff distance (HD) of 6.4 mm, which is 10 % (resp., 16 %) improvement compared to the graph cuts method solely using the PET (resp., CT) images.