Background and purpose: Prediction of local failure in radiotherapy patients with non-small cell lung cancer (NSCLC) remains a challenging task. Recent evidence suggests that FDG-PET images can be used to predict outcomes. We investigate an alternative multimodality image-feature approach for predicting post-radiotherapy tumor progression in NSCLC.
Material and methods: We analyzed pre-treatment FDG-PET/CT studies of twenty-seven NSCLC patients for local and loco-regional failures. Thirty-two tumor region features based on SUV or HU, intensity-volume-histogram (IVH) and texture characteristics were extracted. Statistical analysis was performed using Spearman's correlation (rs) and multivariable logistic regression.
Results: For loco-regional recurrence, IVH variables had the highest univariate association. In PET, IVH-slope reached rs=0.3426 (p=0.0403). Motion correction slightly improved correlation of texture features. In CT, coefficient of variation had the highest association rs=-0.2665 (p=0.0871). Similarly for local failure, a CT-IVH parameter reached rs=0.4530 (p=0.0105). For loco-regional and local failures, a 2-parameter model of PET-V(80) and CT-V(70) yielded rs=0.4854 (p=0.0067) and rs=0.5908 (p=0.0013), respectively. Addition of dosimetric variables provided improvement in cases of loco-regional but not local failures.
Conclusions: We proposed a feature-based approach to evaluate radiation tumor response. Our study demonstrates that multimodality image-feature modeling provides better performance compared to existing metrics and holds promise for individualizing radiotherapy planning.
Copyright © 2011 Elsevier Ireland Ltd. All rights reserved.