S2451
Physics - Radiomics, functional and biological imaging, and outcome prediction
ESTRO 2026
1 Department of Radiotherapy, Erasmus MC Cancer Institute, University Medical Center Rotterdam, Rotterdam, Netherlands. 2 Department of Radiation Oncology, MAASTRO, GROW School for Oncology and Developmental Biology, Maastricht University Medical Center+, Maastricht, Netherlands Purpose/Objective: From colorectal surgery, it is known that damage to the internal anal sphincter nerves (IASN) is associated with late bowel complaints that disturb defecation, reducing quality-of-life [1]. The IASNs are located in the perirectal fat and the nerve path close to the prostate receives moderate to high-dose levels in prostate radiotherapy. Bowel symptoms are also observed following pelvic radiotherapy. We hypothesized that an NTCP model for late bowel symptoms, based on IASN dose, could be superior to conventional anal canal or rectum-based models. Material/Methods: We evaluated 712 prostate cancer patients treated within two studies with 39x2.0 Gy, 19x3.4 Gy, 20x3.0/3.1 Gy or 7x6.1 Gy (n ±175 each). Patients completed a bowel symptom questionnaire at baseline and at year 1. Patients were scored for late bowel symptoms if the total score for pain, cramps, urge, incontinence, diarrhea, increased stool frequency (scored with 0-none, 1-little, 2- moderate/severe) increased by ≥ 3 points over baseline, while baseline score did not exceed 3 (n=18). Fig. 1 explains how the IASN-region is defined [1,2]. Doses of around 105 Gy in BED ( α / β =2Gy) are considered relevant for nerve damage [3]. Therefore, we considered relative volumes exposed to doses from 80 Gy to 120 Gy as candidate dose predictors for the IASN prediction model. For the rectum and anus models, EUDs with n-values from 0.05 to 1 using α / β =3Gy were considered as candidate predictors. For IASNs, rectum and anus separately, NTCP models were developed by first selecting the candidate dose predictor that best explains the outcome using likelihood estimation, followed by fitting a logistic regression model with backwards elimination using previous abdominal surgery as candidate clinical predictor. For internal validation, bootstrapping was used to assess model performances.
Conclusion: We presented a generalizable ML framework for voxel- wise extraction of microstructural parameters from DWI, leveraging realistic in silico substrates, providing spatially resolved maps of tumor heterogeneity that outperforms conventional approaches and that can be used for tumor characterization and radiotherapy personalization. References: [1] E. Martinez-Heras et al. 2021, Seminars in
Ultrasound, CT and MRI., doi: 10.1053/j.sult.2021.07.006[2]
G. Buizza et al., 2021.
Med Phys., doi: 10.1002/mp.14689[3]
L. Morelli et
al., 2023. Med Phys., doi: 10.1002/mp.16202[4] Kerkelä et al., 2020, J Open Source Softw., doi: 10.21105/joss.02527.[5]
L.
C. Edlund et al., 2021. Nat
Methods., doi: 10.1038/s41592-021-01249-6[6] D. C. Newitt et al., 2021. ACRIN 6698/I-SPY2 Breast DWI,” The Cancer Imaging Archive[7] G. Litjens, et al., 2017. “SPIE-AAPM PROSTATEx Challenge Data,” The Cancer Imaging Archive Keywords: DW-MRI, tumor microstructure, Machine Learning
Poster Discussion 2793
Conventional to ultra-hypofractionated prostate cancer radiotherapy: Does the dose to the internal anal sphincter nerve predict late bowel symptoms? Christian AM Jongen 1 , Ben JM Heijmen 1 , Andras Zolnay 1 , Luca Incrocci 1 , Andre Dekker 2 , Leonard Wee 2 , Wilma D Heemsbergen 1
Made with FlippingBook - Share PDF online