ESTRO 2023 - Abstract Book

S1320

Digital Posters

ESTRO 2023

Purpose or Objective Neurovascular-sparing external beam radiotherapy (RT) has gained more interest in patients with localized prostate cancer (PCa) since the introduction of MR-guided RT. Manual contouring of small neurovascular structures is a labor and time intensive process and is prone to considerable interrater variations. Our aim is to delineate neurovascular structures automatically on prostate MRI and to reach consistent segmentations by deep learning. Materials and Methods Pretreatment 3.0 Tesla T2-weighted offline planning MRI data (Ingenia MR-RT, Philips Healthcare, The Netherlands) of 134 patients with localized PCa were used for the annotation of conventional structures and neurovascular structures. The neurovascular structures include the penile bulb (PB), corpora cavernosa (CCs), internal pudendal arteries (IPAs), and the neurovascular bundles (NVBs). The three-dimensional convolutional neural network DeepMedic was used to delineate neurovascular structures automatically on prostate MRI. The included patients were randomly split into a training cohort (n = 87), validation cohort (n = 20), and test cohort (n = 27). The quantitative performance of the deep learning-based delineations was evaluated on the test cohort in terms of volumetric Dice similarity coefficient (DSC), mean surface distances (MSD), Hausdorff distances, and the surface DSC within a clinically relevant tolerance threshold of 1 mm (SDSC1mm) and 2 mm (SDSC2mm). The deep learning-based delineations were evaluated on the same axial slices as the study delineations. In addition, the NVBs were evaluated at prostate level and at the exact inferior half (i.e., prostate midgland to apex level) where the NVB is in closest proximity to the clinical target volume and lies within the high dose gradient region. Results Median DSC for the PB was 0.92 (IQR: 0.91 – 0.94), for the CCs 0.91 (IQR: 0.89 – 0.92), for the IPAs 0.76 (IQR: 0.71 – 0.79), for the NVBs at prostate level 0.69 (IQR: 0.64 – 0.74), and for the inferior half of the NVBs 0.71 (IQR: 0.64 – 0.76). Median MSD for the IPAs was 0.34 mm (IQR: 0.26 – 0.45 mm), for the NVBs at prostate level 0.72 mm (IQR: 0.62 mm – 0.77 mm), and for the inferior half of the NVBs 0.52 mm (IQR: 0.42 mm – 0.62 mm). Median SDSC1mm for the IPAs was 0.90 (IQR: 0.86 – 0.93), for the NVBs at prostate level 0.72 (IQR: 0.65 - 0.78), and for the inferior half of the NVBs 0.83 (IQR: 0.75 – 0.90).

Conclusion The use of deep learning for automatic delineations of neurovascular structures is feasible on pretreatment prostate MRI data for MR-guided RT. A good performance was reached for the PB, CCs, IPAs, and the inferior half of the NVBs. The use of deep learning shows potential to significantly reduce the workflow load for radiation oncologists compared to manual contouring.

PO-1625 Reproducibility of AI-based contour generation on synthetic CT

I. Coric 1 , N. Wernlein 1 , M. Nachbar 1 , D. Wegener 2 , S. Boeke 2 , D. Zips 2 , C. Gani 2 , D. Thorwarth 1

1 Section for Biomedical Physics. Department of Radiation Oncology, University Hospital and Medical Faculty. Eberhard Karls University Tuebingen, Tuebingen, Germany; 2 Department of Radiation Oncology, University Hospital and Medical Faculty. Eberhard Karls University Tuebingen, Tuebingen, Germany Purpose or Objective Automatic segmentation based on artificial intelligence (AI) offers a time-saving solution for delineation of organs at risk (OARs), which is essential for an online adaptive workflow at the MR-Linac. With the introduction of synthetic CTs (sCT) in radiotherapy, the need for registration of a planning CT to the daily MRI can be bypassed. However, the use of sCTs can be extended for delineation of OARs in the daily adaptation routine. While the segmentation of bony structures is dosimetrically relevant but proves to be difficult on MRI, the sCT provides remediation. In this work, the reproducibility of OAR segmentations generated by AI-based MRI and CT delineation algorithms on MRI and sCT was compared.

Materials and Methods