医学
无线电技术
子宫颈
核医学
小波
放射治疗
放射科
肿瘤科
人工智能
内科学
癌症
计算机科学
作者
Roxolyana Abdah-Bortnyak,Zohar Keidar,Salem Billan,Ramandeep Gill,U. Bar-Peled,Ella Kuptzov,Myroslav Lutsyk
标识
DOI:10.1016/j.ijrobp.2020.07.751
摘要
The purpose of this study was to test a predictive model for complete metabolic response of primary cervical tumor using radiomics wavelet transform analysis derived from pretreatment 18FDG PET-CT data. The study population included newly diagnosed patients with locally advanced Squamous cell carcinoma of cervix uteri (FIGO, 2009), who were treated with definitive chemo-radiotherapy. For radiomics analysis of primary tumor, metabolic tumor volume (MTV) was defined on pretreatment 18FDG PET-CT using a SUV threshold of 40%. Complete metabolic response (CMR) was defined as absence of pathologic FDG uptake on PET/CT, performed three months after completion of treatment, using the same MTV threshold of 40%. Radiomics software package with wavelet feature extraction module was used for image texture analysis, while analysis of 850 radiomics features were included. Descriptive and frequency statistics were used. For evaluation of prediction performance, multilayer perceptron algorithm with two hidden layers was used. TwoStep cluster analysis was used for patient classification. 86 patients were enrolled, and their data allocated for training (56 pts) and testing (22 pts) sets. CMR was detected in 65 (75.6%) patients. Multilayer perceptron model showed high predictive importance of wavelet in the model, with sensitivity of 0.95 of area under the receiving operator characteristics curve. Applying TwoStep cluster analysis for classifying the texture and wavelet parameters, two clusters were defined with predominant wavelet features importance. This study demonstrated that PET-based radiomics features in combination with wavelets features analysis can be used for pretreatment prediction of CMR of the primary tumor in locally advanced cervical cancer. The model validation is needed.
科研通智能强力驱动
Strongly Powered by AbleSci AI