PO0408: Artificial Intelligence-Based MRI-Only Post-Planning Workflow for Permanent 125-I Prostate Brachytherapy

• Published in: Brachytherapy Vol. 23; no. 6; p. S137
• Main Authors: Karius, Andre, Grigo, Johanna, Strnad, Vratislav, Schweizer, Claudia, Merten, Ricarda, Putz, Florian, Fietkau, Rainer, Bert, Christoph
• Format: Journal Article
• Language: English
• Published: Elsevier Inc 01.11.2024
• ISSN: 1538-4721
• DOI: 10.1016/j.brachy.2024.08.222

The creation of a post-plan for permanent prostate brachytherapy is nowadays usually based on computed tomography (CT) scans. This enables a distinct detection of the implanted 125-I seeds, but also features drawbacks with respect to a delineation of the prostate and organs at risk compared to magnetic resonance imaging (MRI), due to the low soft tissue contrast provided by CT. For establish a corresponding MRI-only workflow, we developed a seed detection on MRI images based on a Deep-Learning auto-segmentation approach. Six-teen patients treated with 125-I seeds received a CT scan as well as a 1.5 T MRI scan for creating a post-plan at nominal day 30 after implantation. In addition to a T2-SPACE sequence for contouring the prostate, a DIXON sequence as well as a Deep-Learning based quantitative susceptibility map (QSM) based on a 3D gradient-echo sequence were acquired. The CT was fused with the MRI scans and the CT based seed-detection was transferred to the MRI scans to obtain a ground-truth of the implant arrangement. Afterwards, an nnUnet was trained on the QSM and DIXON sequences to learn the detection of seeds considering 12 patients. Finally, a validation of the artificial intelligence-based seed detection was performed for the remaining four patients. The accuracy of seed detections compared to the clinical, CT-based seed identification amounted 89% (considering DIXON only) and 92% (combining DIXON and QSM). The number of false positive detection amounted only up to 3%. The centroids of the seeds detected by means of QSM deviated by only 2.1±1.2 mm from the results obtained via CT. Based on volumetric post-processing, merged seeds could be differentiated as well and in this way altogether 98% of all seeds could be automatically detected with the implemented MRI-only workflow. Regarding dosimetry, the D90 of the prostate differed by a mean of only 0.4 Gy between CT-based and MRI-only based post-planning. The proposed MRI-only workflow represents a promising methodology for an accurate and robust localization of seeds within the prostate, and is considered a valuable alternative to CT-based post-planning.