Author: Hao Peng, Yajun Yu 👨🔬
Affiliation: Department of Radiation Oncology, UT Southwestern Medical Center, Medical Artificial Intelligence and Automation (MAIA) Lab, Department of Radiation Oncology, UT Southwestern Medical Center 🌍
Purpose: Personalized ultra-fractionated stereotactic adaptive radiotherapy (PULSAR) is a novel ablative radiation dosing scheme developed by our institution. This study aims to establish a regression model to predict gross tumor volume (GTV) after PULSAR, facilitating early and informed decision-making in treatment planning.
Methods: A retrospective study encompassing 69 brain metastasis lesions treated with PULSAR was conducted. Radiomic and dosiomic features were extracted from MRI scans and dose maps, respectively, at pre-treatment and intra-treatment stages. Delta-omics was calculated by the relative change between pre- and intra-treatment features. Redundant and irrelevant features were eliminated by means of variance threshold, linear correlation analysis, and least absolute shrinkage and selection operator (Lasso) algorithm. In order to predict the follow-up GTV, support vector regression (SVR) was utilized and constructed based on multi-omic features. Five-fold cross-validation procedure was applied with 10 repeats to mitigate the limitation of small sample size.
Results: Our study demonstrated that a combination of radiomics and dosiomics outperformed individual-omics model for GTV prediction. Radial basis function (RBF) kernel implemented in SVR was superior compared with linear or polynomial kernel function. The radiomics and dosiomics integrated SVR model with RBF kernel achieved a coefficient of determination (R2) of 0.743 and a relative root mean square error (RRMSE) of 0.022.
Conclusion: The combination of radiomics and dosiomics makes full use of individual-omic features to build a more accurate GTV regression model for PULSAR-treated lesions. Leveraging multi-omics and SVR, it is promising to shift the early decision-making process in PULSAR from empirical judgments to a data-driven strategy. This approach enables more personalized radiation therapy, optimizes patient management and reduces the risks of under- or over-treatment.