Program Information
Shell Feature: A New Descriptor for Predicting Distant Failure in Lung SBRT
H Hao1,2,3 Z Zhou3, S Li3,4, M Folkert3 , K Westover3, P Iyengar3, L Yang5,J Wang3, 1. School of Computer Science and Technology, Xidian University, Xi'an,710071, China, 2. Key Laboratory of Intelligent Perception and Image Understanding of Ministry of Education, International Research Center for Intelligent Perception and Computation, Joint International Research Laboratory of Intelligent Perception and Computation, Xidian University, Xi'an,710071, China, 3. University of Texas Southwestern Medical Center, Dallas, TX, 75390, United States, 4. School of Biomedical Engineering,Southern medical university, Guangzhou, 510515, China, 5. Department of Pathology, National Cancer Center/Cancer Hospital, Chinese Academy of Medical Sciences, Beijing, 100021, China.
Presentations
WE-F-605-8 (Wednesday, August 2, 2017) 1:45 PM - 3:45 PM Room: 605
Purpose: To develop and demonstrate a novel tumor shell feature for predicting distant failure in early stage non-small cell lung cancer (NSCLC) patients treated with stereotactic body radiation therapy (SBRT).
Methods: A total of 48 stage IA and IB NSCLC patients underwent stereotactic body radiation therapy (SBRT) from 2006 to 2012 were used. A tumor shell feature, consisted of the outer voxels around the tumor boundary, was extracted from a serial of axial 3D positron emission tomography (PET) slices. The hypothesis behind this feature is that noninvasive lesion and invasive tumor may differ in their morphologic patterns in the surrounding area of tumor edge, which in turn reflect the differences of radiological presentations in PET imaging. The utility of proposed feature was evaluated through three predictive models: 1) support vector machine (SVM); 2) dictionary learning (DL); and 3) DL and SVM combined method (DL_SVM), where the vectorized shell feature was used as the input for these models. Meanwhile, the same three classifiers using a collection of 34 handcrafted features commonly applied for PET were used for comparison. In all predictive models, a synthetic minority over-sampling technique (SMOTE) is performed to handle the class-imbalanced data. The performance of the shell feature was assessed in terms of the area under the characteristic curve (AUC), sensitivity, specificity and accuracy.
Results: The tumor shell feature showed better discrimination than the handcrafted features in all three classifiers. The AUC, sensitivity, specificity and accuracy of the shell further via the DL_SVM classifier are 84%, 81%, 85% and 83%, while for the handcrafted features are 75%, 75%, 77% and 75%, respectively.
Conclusion: The proposed tumor shell feature may be associated with the invasive properties of the primary tumor and it can be used to predict the distant failure for early stage NSCLC patients after SBRT.
Contact Email: