(Sunday, 7/29/2018) 4:30 PM - 5:00 PM
Room: Exhibit Hall | Forum 5
Purpose: Currently, there is no reliable method for predicting tumor progression treatment in lung cancer patients undergoing radiotherapy. Our purpose was to evaluate the capacity of the primary tumor standardized uptake values (SUVmax) extracted from pre-treatment thoracic PET/CT images to predict tumor progression in stage II-III locally advanced non-small cell lung cancer (LA-NSCLC) patients undergoing radiotherapy. We hypothesized that SUVs can improve the prediction of tumor progression with respect to the commonly used TNM staging system.
Methods: We analyzed 169 consecutive LA-NSCLC patients treated between 2008 and 2016 with either IMRT or PBT to a median dose of 66.6 Gy in 37 fractions using sequential or concurrent chemoradiation with platinum-based doublet; 14% of these patients developed tumor progression after treatment. Four ordinal features - T, N, and M stages and overall staging group - were used as our baseline model predicting progression/regression status. We also augmented the baseline model by adding the tumor SUVmax extracted from pre-treatment PET/CT scans. Models were built using logistic regression with resubstitution, where the area under the receiver-operating characteristic (ROC) curve (AUC) was used to assess the prediction performance.
Results: The baseline model, solely based on TNM staging, demonstrated predictive performance of AUC = 0.66, p = 0.02. When the primary tumor SUVmax was added to the baseline model, the performance was significantly improved up to AUC = 0.71, p = 0.003.
Conclusion: We found that the capacity of the commonly used tumor TNM staging to predict tumor progression in LA-NSCLC improves significantly when combined with the SUVmax derived from pre-treatment thoracic PET/CT scans. These findings can inform treatment planning for radiation oncologists and may ultimately allow for personalized treatment delivery to reduce patient morbidity or to intensive therapy to reduce treatment failures. Prospective data validating these findings are needed.
Keywords
ROC Analysis, Lung, Pattern Recognition
Taxonomy
IM/TH- Image Analysis (Single modality or Multi-modality): Machine learning
Contact Email
