Stability of dosomics features extraction on grid resolution and algorithm for radiotherapy dose calculation.

Purpose: Dosomics is a novel texture analysis method to parameterize regions of interest and to produce dose features that encode the spatial and statistical distribution of radiotherapy dose at higher resolution than organ-level dose-volume histograms. This study investigates the stability of dosomics features extraction, as their variation due to changes of grid resolution and algorithm dose calculation. Material and Methods: Dataset has been generated considering all the possible combinations of four grid re-solutions and two algorithms dose calculation of 18 clinical delivered dose distributions, leading to a 144 3D dose distributions dataset. Dosomics features extraction has been performed with an in-house developed soft-ware. A total number of 214 dosomics features has been extracted from four different region of interest: PTV, the two closest OARs and a RING structure. Reproducibility and stability of each extracted dosomic feature (Rfe, Sfe), have been analyzed in terms of intraclass correlation coefficient (ICC) and coefficient of variation. Results: Dosomics features extraction was found reproducible (ICC > 0.99). Dosomic features, across the combination of grid resolutions and algorithms dose calculation, are more stable in the RING for all the con-sidered feature's families. Sfe is higher in OARs, in particular for GLSZM features' families. Highest Sfe have been found in the PTV, in particular in the GLCM features' family. Conclusion: Stability and reproducibility of dosomics features have been evaluated for a representative clinical dose distribution case mix. These results suggest that, in terms of stability, dosomic studies should always perform a reporting of grid resolution and algorithm dose calculation.