Analytics For Progression Free Survival And Distant Metastasis Prediction Of Anal Cancer Patients After Chemoradiation Therapy Using Spatial Temporal Fdg-Pet/Ct Features

M.D. Chuong,Hao Howard Zhang,Jian Z. Wang,K. Latifi,N. Saeed,S Tan,Wook-Jin Choi,S.E. Hoffe,R. Shridhar,Wei Lu

DOI: https://doi.org/10.1016/j.ijrobp.2015.07.954

2015-01-01

Abstract:To evaluate the role of advanced analytics in predicting progression-free survival (PFS) and distant metastasis (DM) of anal cancer patients treated with chemoradiation therapy (CRT). Seventeen anal cancer patients (3T1, 7T2, 5T3, 2T4 and 12N0, 1N1, 3N2 1N3) treated with CRT were retrospectively studied. The median prescription dose was 56 Gy (range, 50-62.5 Gy). All patients underwent FDG-PET/CT scans before and after CRT. Three types of variables were examined for their ability to predict PFS and DM: (1) 26 clinical parameters (age, gender, TNM stage, histology, GTV dose, etc.); (2) 15 traditional PET/CT measures (SUVmax, SUVpeak, tumor diameters, etc.); (3) 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 methods to select an optimal set of predictors and supervised machine learning (ML) for predictive analysis was evaluated. ML models, including logistic regression (LR), decision tree (DT), neural network (NN) and support vector machine (SVM), were tested. Leave-one-out cross-validation was used to assess the reliability and accuracy of each model. The optimal predictor set contained 14 variables: 3 clinical parameters (tumor size, T stage and whether 5-FU was held during any cycle of chemotherapy), 2 traditional PET/CT measures (pre-CRT SUVmin and SUVmedian), and 9 spatial-temporal PET/CT features. Using this optimal set of 14 predictors, all 13 PFS patients (who did not have recurrence) were correctly predicted and only 1 local recurrence patient was misclassified. Overall, 94% prediction accuracy was achieved for predicting PFS. All 3 DM patients were correctly classified resulting 100% prediction accuracy (no other patients were misclassified). The high accuracy was achieved by LR, NN and SVM, except DT (71%). 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.