Author: David L. Barbee, David Byun, Ting Chen, Paulina E. Galavis, Siming Lu, Sarah Rosemary Morris, Hesheng Wang, Michael J Zelefsky 👨🔬
Affiliation: NYU Langone Health 🌍
Purpose: MR-Linac enables dose escalation in prostate SBRT on accurately defined dominant intra-prostate lesion (DIL) on daily MR images. This study aims to evaluate inter-fraction changes in the radiomics features of T2-weighted planning MR.
Methods: Twenty-two patients undergoing 5-fraction SBRT for prostate cancer (Prostate: 40Gy; DIL: 45Gy) on a Unity MR-Linac were included. The prostate and DIL were contoured on daily T2w MR images, along with diffusion-weighted and other sequence MR. The T2w images were intensity normalized, and 2D radiomics feature maps were calculated on the axial slices of the prostate using pyRadiomcs. Mean feature values in the volumes of DIL, DIL_margin (DIL_PTV-DIL), and Prostate (Prostate–DIL_PTV) were calculated. Linear mixed models are used to assess fraction-dose responses of the features with random effects on the intercepts and slopes to account for patient-specific differences in the baseline values and response rates.
Results: Average doses in the prostate, DIL_margin and DIL were 41.8, 45.3, and 46.6Gy, respectively, over 5 fractions. For the 1st, 3rd and 5th fraction, the average DIL volumes were 1.73±0.31, 1.95±0.28, and 2.08±0.30cc, and the prostate volumes were 53.6±5.2, 60.8±6.0 and 60.6±6.1cc, respectively. No significant differences were observed in radiomics features between the prostate and DIL_margin. Feature glrm_GrayLevelNonUniformity (p=0.004) and others (p<0.05) of the prostate had significant linear relations with fraction numbers with random effects on the baseline values. With random effects on both baseline and response rates, glrlm_RunLengthNonUniformityNormalized (p=0.002) and others in the DIL showed significant linear dose responses.
Conclusion: Selected radiomics features from planning MR images show significant responses to inter-fraction treatments in both prostate and DIL. The patient-specific inter-fraction dynamics of the features may correlate with the tumor responses, providing early predictions for treatment outcomes of prostate SBRT on MR-Linac.