TY - JOUR T1 - Robustness of Textural Features in 11C-choline and 18F-FDG PET/CT Scans of Nasopharyngeal Carcinoma JF - Journal of Nuclear Medicine JO - J Nucl Med SP - 1896 LP - 1896 VL - 57 IS - supplement 2 AU - Lijun Lu AU - Wenbing Lv AU - Jun Jiang AU - Jianhua Ma AU - Qianjin Feng AU - Arman Rahmim AU - Wufan Chen Y1 - 2016/05/01 UR - http://jnm.snmjournals.org/content/57/supplement_2/1896.abstract N2 - 1896Objectives Textural features are increasingly utilized to evaluate tumor heterogeneity and predict therapy response in 18F-FDG PET. Investigations in PET imaging have been primarily limited to FDG PET/CT imaging, but texture analysis in cell proliferation radiotracer 11C-choline PET has not, to our knowledge, been reported. The purpose of the present work is to investigate the impact of segmentation and gray-scale resampling on textural features in 11C-choline and 18F-FDG PET images of nasopharyngeal carcinoma (NPC).Methods Nine patients with nasopharyngeal carcinoma were enrolled in the study. Each patient underwent 11C-choline and 18F-FDG PET/CT scans on two different days. Tumors were delineated using manual segmentation by expert physician, SUV cut-off of 2.5, as well as 15%, 30% and 40% SUVmax thresholding, and then resampled onto 8, 16, 32, 64, 128 gray-scale bins. A total of 31 features based on SUV (intensity), shape and histogram, as well as 57 textural features were extracted from the tumors. The robustness of the features were evaluated by intra-class correlation coefficient (ICC) for different segmentation and resampling methods, respectively. For each feature, inter-segmentation (expert physician, SUV cut-off of 2.5, as well as 15%, 30% and 40% SUVmax thresholding) ICC was computed with fixed resampling 64 gray-scale bins, because, there is no obvious differentce between features extracted with 64 and 128 gray-scale bins. Correspondingly, inter-resampling (8, 16, 32, 64, 128 gray-scale bins) ICC was computed with manual segmentation by expert physician.Results 76 (86%) features extracted from 18F-FDG exhibited ICC higher than 0.8, while only 55 (62%) features obtained from 11C-choline had ICC higher than 0.8 by changing segmentation. 12 (21%) features depicted ICC higher than 0.8 on both 18F-FDG and 11C-choline with respect to resampling. 8 features derived from the gray level cooccurance matrix (GLCM) showed greatest robustness to segmentation and resampling on both 18F-FDG and 11C-choline PET. Specifically, these were Entropy, DifEntropy, Homogeneity1, Homogeneity2, Correlation1, ICM2, IDMN and IDN for segmentation (ICC>0.88) and resampling (ICC>0.83). Besides, Variance1_GLCM, RLN_GLRLM and Coarseness_NGTDM also presented high ICC on 18F-FDG PET for both segmentation (ICC>0.95) and resampling (ICC>0.87).Conclusions Resampling depicted larger effects on the features variability than segmentation in both 18F-FDG and 11C-choline PET images. Features extracted from 18F-FDG PET are more robust than 11C-choline PET with respect to segmentation. Resampling has very similar effects on features computed from 18F-FDG vs. 11C-choline PET. ER -