Program Information
Earlier Prediction of Treatment Efficacy in Neoadjuvant Chemoradiation of Sarcomas Using Radiomics Analysis From Texture Features in DCE MRI
L Xu1 , P Yang1 , W Huang2 , T Niu1*, Y Kuang3 (1) Zhejiang University, Hangzhou, Zhejiang (2) Oregon Health & Science University, Portland, OR (3) University of Nevada, Las Vegas, Las Vegas, NV
Presentations
TU-L-GePD-JT-5 (Tuesday, August 1, 2017) 1:15 PM - 1:45 PM Room: Joint Imaging-Therapy ePoster Theater
Purpose: To investigate the feasibility of radiomics analysis from texture features in DCE-MRI as an early predictive biomarker for efficacy of neoadjuvant chemoradiation therapy in sarcoma.
Methods: Serial DCE-MRI scans were performed on days before therapy (time point 1, TP1) and after 2 weeks of chemoradiation therapy (time point 2, TP2) in fourteen patients with high-risk extremity soft tissue sarcoma. Fifty-two voxel-based texture features were extracted within the tumor regions from T1-weighted and DCE-MRI images at TP1 and TP2. The variance of texture features between TP1 and TP2 were calculated, resulting in 104 imaging feature differences as the parameters for the predictive model constructed. Univariate Analyses were employed to evaluate the capability of each individual parameter in predicting pathologically optimal response or sub-optimal responses. The significantly different individual parameters (p<0.05) calculated from univariate analyses were incorporated into artificial neural network (ANN) with 4-fold validation technique to obtain a best predictive model, which can classify the patients into different response groups. The predictive performance of the model was calculated using receiver operating characteristic (ROC) curves.
Results: Eight parameters including gray level co-occurrence matrix (contrast cluster prominence, dissimilarity, difference variance, difference entropy, inverse difference normalized), gray level size zone matrix (small zone emphasis), neighborhood gray-tone difference matrix (contrast), showed a strong capability to predict the treatment efficacy in the Univariate Analyses (p<0.05). The best predictive performance was obtained using an ANN model of incorporating these eight parameters. The model reached an area under ROC (AUC) of 0.85 ± 0.0029.
Conclusion: The promising results demonstrated the ANN model with the eight parameters derived from DCE-MRI could provide an improved predictive value over conventional imaging metrics, which would offer the potential to tailor treatment to individual patients nearly in real time, thereby substantially improving patient survival and quality of life.
Funding Support, Disclosures, and Conflict of Interest: Research Supported by National Institutes of Health: eRA Person ID for the Principal(s) Investigator: 10567553 Funding: NIH/NIGMS P20 GM103440; NIH U01 CA154602 Zhejiang Provincial Natural Science Foundation of China (Grant No. LR16F010001) National High-tech R&D Program for Young Scientists by Ministry of Science and Technology of China (863 Program, 2015AA020917)
Contact Email: