Semi-automated whole-body texture analysis may improve predictive performance of FDG PET-CT for patients with differentiated thyroid carcinoma

• Published in: The Journal of nuclear medicine (1978) Vol. 58; p. 107
• Main Authors: Hirata, Kenji, Shiga, Tohru, Uchiyama, Yuko, Watanabe, Shiro, Toyonaga, Takuya, Manabe, Osamu, Kobayashi, Kentaro, Kikuchi, Hisaya, Magota, Keiichi, Tamaki, Nagara
• Format: Journal Article
• Language: English
• Published: New York Society of Nuclear Medicine 01.05.2017
• Subjects: Cluster analysis; Clustering; Computed tomography; Correlation analysis; Glycolysis; Heterogeneity; Image analysis; Image processing; Inflammation; Iodine; Iodine 131; Iodine radioisotopes; Lesions; Mathematical analysis; Medical imaging; Medical prognosis; Metastases; Nuclear medicine; Parameters; Patients; Performance prediction; Population (statistical); Population studies; Positron emission; Positron emission tomography; Prognosis; Rank tests; Risk analysis; Risk factors; Scanners; Statistical tests; Therapy; Thyroid; Thyroid cancer; Thyroid carcinoma; Thyroidectomy; Tomography; Tumors
• ISSN: 0161-5505; 1535-5667

Objectives: Although patients with differentiated thyroid carcinoma generally have good prognosis, several risk factors are known to indicate poor prognosis. Recent advances in molecularly-targeted therapies make it more important to find patients with poor prognosis despite repeated radioiodine therapy. To date, FDG PET-CT is known to provide useful information for risk stratification. SUVmax, metabolic tumor volume (MTV), and total lesion glycolysis (TLG) are used to semi-quantitatively analyze the FDG PET images. More recently, texture analysis, a group of methods to quantify image heterogeneity, has been reported to be useful for other malignancies. As an advantage, texture parameters can be calculated semi-automatically and thus can provide objective information. In this study, we aimed to propose a method of whole-body texture analysis for FDG PET-CT, and to clarify its usefulness for predicting overall survival (OS) in patients with differentiated thyroid carcinoma. Methods: In this retrospective study, we included patients with differentiated thyroid carcinoma who underwent FDG PET/CT after total thyroidectomy before receiving I-131 radioactive iodine therapy. Inclusion criteria were as follows: the patients were examined using a single PET-CT scanner (Siemens Biograph64) because different scanners may affect texture features; the patients had MTV greater than 3 ml (=90 voxels) to allow texture analysis. The final population of the study consisted of 54 patients (M:F = 19:35, 65.6±9.7 years old). After >6 hours fasting, the patients were injected with FDG (4.5MBq/kg), followed by whole-body scanning 1 hour later. In image analysis, all the uptake masses with cut-off of SUV...3.0 were extracted automatically. Then, a nuclear medicine physician manually removed physiological or inflammatory uptake masses. All the metastatic lesions were considered for the following calculation of texture features. Histogram and 4 texture matrices (co-occurrence matrix, gray-level run length matrix, gray-level zone length matrix, and neighborhood gray-level different matrix) were generated to calculate a total of 36 texture parameters. The optimum cut-off was determined for each parameter by testing any considerable cut-off values using Kaplan-Meier method with log-rank test. The best cut-off producing the maximum chi-square value was adopted. The predictive performance was compared among texture parameters based on the maximized chi-square value. P<0.00125 by Bonferroni's correction was considered significant. Hierarchical clustering was applied to categorize the texture parameters. Results: During the follow-up period, 12 of 54 patients died. The median follow-up period for surviving patients was 29 months. Fourteen of the 36 parameters (HGZE, SZHGE, HGRE, SRHGE, LRHGE, GLNUr, contrast, LZHGE, dissimilarity, SRE, LRE, RP, LZE, ZP) successfully distinguished patients between longer vs. shorter OS at the best cut-off (P<0.00125), whereas SUVmax, MTV, and TLG did not reach statistical significance (p=0.24, 0.02, and 0.01, respectively). Hierarchical clustering showed that all the 36 parameters could be categorized into roughly 4 groups. In addition, Pearson's correlation analysis revealed that strong correlations between some parameters such as ContrastCM and Dissimilarity (R>0.9), indicating that the number of parameters can be reduced. Conclusion: These results suggested that the semi-automated whole-body texture analysis for FDG PET-CT can provide objective and operator-independent classifiers and also may have additional value to conventional parameters (SUVmax, MTV, and TLG) in predicting prognosis of patients with differentiated thyroid carcinoma. (ProQuest: ... denotes formulae/symbols omitted.)