F-18-FDG PET radiomics approaches: comparing and clustering features in cervical cancer

作者：Tsujikawa Tetsuya^*; Rahman Tasmiah; Yamamoto Makoto; Yamada Shizuka; Tsuyoshi Hideaki; Kiyono Yasushi; Kimura Hirohiko; Yoshida Yoshio; Okazawa Hidehiko

来源：Annals of Nuclear Medicine, 2017, 31(9): 678-685.

DOI：10.1007/s12149-017-1199-7

• PET
• Textural feature
• Cervical cancer
• Radiomics

The aims of our study were to find the textural features on F-18-FDG PET/CT which reflect the different histological architectures between cervical cancer subtypes and to make a visual assessment of the association between F-18-FDG PET textural features in cervical cancer.
Eighty-three cervical cancer patients [62 squamous cell carcinomas (SCCs) and 21 non-SCCs (NSCCs)] who had undergone pretreatment F-18-FDG PET/CT were enrolled. A texture analysis was performed on PET/CT images, from which 18 PET radiomics features were extracted including first-order features such as standardized uptake value (SUV), metabolic tumor volume (MTV) and total lesion glycolysis (TLG), second- and high-order textural features using SUV histogram, normalized gray-level co-occurrence matrix (NGLCM), and neighborhood gray-tone difference matrix, respectively. These features were compared between SCC and NSCC using a Bonferroni adjusted P value threshold of 0.0028 (0.05/18). To assess the association between PET features, a heat map analysis with hierarchical clustering, one of the radiomics approaches, was performed.
Among 18 PET features, correlation, a second-order textural feature derived from NGLCM, was a stable parameter and it was the only feature which showed a robust trend toward significant difference between SCC and NSCC. Cervical SCC showed a higher correlation (0.70 +/- 0.07) than NSCC (0.64 +/- 0.07, P = 0.0030). The other PET features did not show any significant differences between SCC and NSCC. A higher correlation in SCC might reflect higher structural integrity and stronger spatial/linear relationship of cancer cells compared with NSCC. A heat map with a PET feature dendrogram clearly showed 5 distinct clusters, where correlation belonged to a cluster including MTV and TLG. However, the association between correlation and MTV/TLG was not strong. Correlation was a relatively independent PET feature in cervical cancer.
F-18-FDG PET textural features might reflect the differences in histological architecture between cervical cancer subtypes. PET radiomics approaches reveal the association between PET features and will be useful for finding a single feature or a combination of features leading to precise diagnoses, potential prognostic models, and effective therapeutic strategies.

• 出版日期2017-11