PT  - JOURNAL ARTICLE
AU  - Zhang, Hao
AU  - Wang, Jiahui
AU  - Chuong, Michael
AU  - Latifi, Kujtim
AU  - Tan, Shan
AU  - Choi, Wookjin
AU  - Hoffe, Sarah
AU  - Shridhar, Ravi
AU  - Lu, Wei
TI  - Analytics in predicting progression-free survival and distant metastasis after chemoradiotherapy using spatial-temporal FDG-PET/CT features
DP  - 2015 May 01
TA  - Journal of Nuclear Medicine
PG  - 1414--1414
VI  - 56
IP  - supplement 3
4099  - http://jnm.snmjournals.org/content/56/supplement_3/1414.short
4100  - http://jnm.snmjournals.org/content/56/supplement_3/1414.full
SO  - J Nucl Med2015 May 01; 56
AB  - 1414 Objectives Using spatial-temporal FDG-PET/CT features to assess the accuracy of advanced analytics in predicting progression-free survival (PFS) and distant metastasis (DM) of anal cancer patients treated with chemoradiotherapy (CRT).Methods 17 patients underwent FDG-PET/CT scans before and after CRT. 3 types of features were examined: 15 traditional PET/CT measures (SUVmax, etc.); 26 clinical parameters (TNM stage, etc.); 253 spatial-temporal PET/CT features, which comprehensively quantify a tumor’s FDG uptake intensity and distribution, spatial variation (texture), geometric property and their changes. Advanced analytics including feature selection method and supervised machine learning models (logistic regression (LR), decision tree (DT), neural network (NN), support vector machine (SVM)) was applied to achieve an optimal set of features and to improve prediction accuracy. Leave-one-out cross-validation was used to assess the accuracy.Results The optimal set contains 14 features: 2 traditional PET/CT measures (pre-CRT SUVmin and SUVmedian), 3 clinical parameters (tumor size, T stage, 5-FU held during CRT), and 9 spatial-temporal PET/CT features (see table). Using this optimal set, 94% prediction accuracy was achieved for predicting PFS: only 1 local recurrence patient was misclassified. All 3 DM patients were correctly classified resulting 100% prediction accuracy (no other misclassifications). None of the local recurrence patients or DM patients could be classified correctly via traditional measures or clinical parameters alone. The high accuracy was achieved by LR, NN and SVM, except DT (71%).Conclusions Advanced analytics combining spatial-temporal PET/CT features, traditional measures and clinical parameters achieved high accuracy in predicting PFS and DM of anal cancer patients treated with CRT. Selected spatial-temporal PET/CT features