Improving The Prognostic Value Of FDG-PET For NSCLC Patients Treated With SABR Using Advanced Computational Modeling

Abstract

suppression weighted (T2FS) sequences. The cohort had a median followup period of 33 months (range: 9-70) during which 19 patients developed lung metastases. Forty-one different textures (e.g., homogeneity, coarseness, large zone emphasis, etc.) were extracted from the tumor region (excluding surrounding edema) of the separate (FDG-PET, T1 and T2FS) and fused (FDG-PET/T1 and FDG-PET/T2FS) PET/MR scans. Fusion of the scans was implemented using the wavelet transform. The construction of prediction models from the combination of different textural features was then performed using logistic regression with feature set reduction, forward feature selection, and correction for small sample size effect. The performance of the resulting models for lung metastases prediction was evaluated using the area under the receiver operating characteristic curve (AUC) on bootstrap resampling in order to maximize their generalizability to out-of-sample STS population. Results: Overall, textures extracted from fused scans outperformed those from separate scans for the prediction of lung metastases by 10%. The best performance was found using a multivariable model with the following 3 texture parameters extracted from fused scans: small zone emphasis, zonesize variance and high gray-level zone emphasis. The average prediction performance of this model using 1000 bootstrapping testing sets yielded an AUC of 0.978 0.001, with a sensitivity and specificity of 0.909 0.006 and 0.899 0.004, respectively. Finally, the uncertainty of the texture model due to contouring variations was analyzed to be 15% using two different contours: one including the visible edema in the vicinity of the tumors, and one excluding it. Conclusions: Our results demonstrate that textural features extracted from fused FDG-PET/MR scans can be used to assess lung metastasis risk at diagnosis of STS. Accurate risk assessment could improve patient outcomes by allowing better treatment adaptation. For future work, we intend to generalize the methodology developed in this study to other cancer types and clinical endpoints. Author Disclosure: M. Vallieres: None. C.R. Freeman: None. S. Skamene: None. I. El Naqa: None.