Prospective Evaluation of Prostate and Organs at Risk Segmentation Software for MRI-based Prostate Radiation Therapy.

Area of Science:

• Radiology and Medical Imaging
• Radiation Oncology
• Medical Physics

Background:

• Accurate segmentation of the prostate and organs at risk (OARs) is crucial for dose-volume histogram analysis in radiation therapy.
• Low-dose-rate prostate brachytherapy (LDRPBT) is a common curative treatment, but automated segmentation performance needs clinical evaluation.

Purpose of the Study:

• To assess the clinical performance of fully convolutional networks for prostate and OAR delineation in LDRPBT using clinically relevant metrics.
• To evaluate the impact of physician refinements on automated segmentations and subsequent dosimetry parameters.

Main Methods:

• Prospectively collected MRI data from 30 patients undergoing LDRPBT were analyzed.
• Custom software utilizing fully convolutional networks performed automatic segmentation of the prostate and four OARs.
• Dose-volume histogram analyses were conducted using both original automated and physician-refined contours.

Main Results:

• Automated segmentation software delineated the prostate and OARs on MRI scans.
• Comparison of dosimetry parameters for the prostate, external urinary sphincter, and rectum showed no significant differences between automated and physician-refined contours.
• Physician refinements did not substantially impact key dosimetry outcomes.

Conclusions:

• Fully convolutional networks demonstrate effective performance for prostate and OAR segmentation in LDRPBT.
• Clinical implementation of automated segmentation for LDRPBT is feasible, as physician refinements do not significantly alter dosimetry.
• This study validates the use of automated segmentation in radiation therapy planning.