Keywords:
Lung, CT-Quantitative, Radiation therapy / Oncology, Radiotherapy techniques
Authors:
Q. Van Den Blink1, P. Ramachandran1, R. Ladwa1, A. Bernard2, M. Lehman1; 1Woolloongabba, QLD/AU, 2St Lucia, QLD/AU
DOI:
10.26044/ranzcr2021/R-0485
Conclusion
Models utilizing radiomic features have the potential to predict treatment outcomes using diagnostic image datasets. These models may have a role, in combination with clinicopathological and genetic features, in personalising treatments in Stage III NSCLC to improve outcomes. Our study identified higher clinical stage and radiomic feature of kurtosis, to have a significant association with failure free survival. Age, stage and features of range and GLCM autocorrelation were significant for influencing overall survival. The accuracy of machine learning models was only modestly affected by using limited versus all features for predicting any failure and increased or maintained high predictive values for 3-year survival across all models. This provides an opportunity to formulating future models which are simpler and more accessible using selected features rather than a larger complement.
Further work will involve exploring radiomic feature selection for machine learning models to predict for specific types of recurrence and creating models combining clinical and radiomic features to increase accuracy in predicting outcomes in this cohort. Models will be validated on independent cohorts of patients with the aim of providing a useful clinical tool in assessing and prognosticating patients undergoing definitive chemoradiation.