Skip to Main content Skip to Navigation

From global to local spatial models for improving prediction of urinary toxicity following prostate cancer radiotherapy

Abstract : External beam radiotherapy (EBRT) is a clinical standard for treating prostate cancer. The objective of EBRT is to deliver a high radiation dose to the tumor to maximize the probability of local control while sparing the neighboring organs (mainly the rectum and the bladder) in order to minimize the risk of complications. Developing reliable predictive models of genitourinary (GU) toxicity is of paramount importance to prevent radiation-induced side-effects, and improve treatment reliability. Urinary symptoms may be linked to the irradiation of specific regions of the bladder or the urethra, in which case the dose received by the entire bladder may not be sufficient to explain GU toxicity. Going beyond the global, whole-organ-based models towards more local, sub-organ approaches, this thesis aims to improve our understanding of radiation-induced urinary side-effects and ameliorate the prediction of urinary toxicity following prostate cancer radiotherapy. With the objective to assess the contribution of urethra damage to urinary toxicity, we propose a multi-atlas-based segmentation method to accurately identify this structure on CT images. The second objective is to identify specific symptom-related subregions in the bladder and the urethra predictive of different urinary symptoms. For this purpose, we propose two methodologies for analyzing the spatial dose distribution; one based on the construction of 2D dose-surface maps (DSM) coupled with pixel wise comparisons and another based on 3D dosevolume maps (DVMs) combined with voxel-wise comparisons. Identified subregions are validated in external populations, opening the perspective for patient specific treatment planning. We also implement and compare different machine learning strategies and data augmentation techniques, paving the way to further improve urinary toxicity prediction. We open the perspective of patient-specific treatment planning with reduced risk of complications.
Complete list of metadata

Cited literature [222 references]  Display  Hide  Download
Contributor : ABES STAR :  Contact
Submitted on : Friday, October 2, 2020 - 2:47:20 PM
Last modification on : Wednesday, September 14, 2022 - 10:20:04 AM
Long-term archiving on: : Monday, January 4, 2021 - 8:39:30 AM


Version validated by the jury (STAR)


  • HAL Id : tel-02956181, version 1


Eugenia Mylona. From global to local spatial models for improving prediction of urinary toxicity following prostate cancer radiotherapy. Traitement du signal et de l'image [eess.SP]. Université Rennes 1, 2019. Français. ⟨NNT : 2019REN1S109⟩. ⟨tel-02956181⟩



Record views


Files downloads