Author: Ryan Alden, Tithi Biswas, Kaushik Halder, Felix Maria-Joseph, Michael Mix, Rihan Podder, Tarun Kanti Podder 👨🔬
Affiliation: SUNY Upstate Medical University, IIT-Roorkee, University of Florida 🌍
Purpose: Radiomics feature-based model for predicting distant recurrence can potentially provide critical insight for clinical decision-making and assistance in treatment strategies. This study focuses on determining the importance of target volumes for machine learning (ML)-based prediction model for the distant recurrence of early-stage non-small cell lung cancer (NSCLC) after two years of stereotactic body radiation therapy (SBRT).
Methods: Data from a cohort of 141 node negative T-stage I-II NSCLC patients treated with SBRT (48-60 Gy in 3-5 fractions) were analyzed; 23 of the patients had distant recurrence. The prediction model was designed based on radiomics features extracted from two volumetric zones: planning target volume (PTV) and gross tumor volume (GTV). Total 47 features were extracted based on histogram analysis, texture, and geometry-based matrices for each of the PTV, and GTV. Initially, multiple classification algorithms were employed for the distant recurrence prediction while considering the inputs as the radiomics features to the ML models. Three most suitable models (Support Vector Machine, Gradient Boosting, and Random Forest algorithms) were selected for designing the novel ensemble model. Model training utilized data from 105 patients, where 36 patients were used for validation.
Results: The designed ensemble model with PTV zone-based radiomics features attained the performance metric ROC-AUC of 0.66 with a sensitivity of 0.43 and specificity of 0.90. However, the model with GTV zone-based radiomics features achieved better performance as ROC-AUC of 0.76, sensitivity of 0.66, and specificity of 0.86.
Conclusion: This study reveals that the GTV is a more impactful region of interest for the radiomics-based distant recurrence prediction model. The variations of radiomics features in PTV that includes microscopic extension of the tumor and partial normal tissues failed to provide with better predictive values for distant recurrence. However, a future study with larger cohort of patients from multiple institutes is reasonable.