Deep Learning-Based CT Segmentation of a Non-radiopaque Hydrogel Rectal Spacer in Prostate Radiotherapy

Area of Science:

• Radiotherapy and Medical Imaging
• Artificial Intelligence in Healthcare
• Oncology

Background:

• Accurate delineation of organs and implants is crucial for effective radiotherapy planning.
• Non-radiopaque hydrogel rectal spacers are used to reduce rectal dose in prostate cancer treatment.
• Current methods may involve MRI for precise contouring, adding complexity and potential registration errors.

Purpose of the Study:

• To evaluate the feasibility and geometric accuracy of CT-based automatic segmentation for hydrogel rectal spacers.
• To determine if accurate segmentation can be achieved without MRI, preserving adjacent organ contouring quality.
• To assess the performance of a deep learning platform (OncoStudio) for this task.

Main Methods:

• Retrospective analysis of 21 prostate cancer patients treated with radiotherapy and a hydrogel rectal spacer.
• CT-based automatic segmentation of the spacer, prostate, rectum, and bladder using OncoStudio.
• Manual delineation on CT with reference to co-registered MRI for comparison.
• Geometric accuracy assessed using Dice Similarity Coefficient (DSC), Hausdorff Distance (HD₉₅), and Mean Surface Distance (MSD).

Main Results:

• The mean DSC for the rectal spacer was 0.787, with mean HD₉₅ of 4.03 mm and mean MSD of 1.18 mm.
• High geometric agreement was observed for adjacent organs: prostate (DSC 0.876), rectum (DSC 0.890), and bladder (DSC 0.963).
• No systematic over- or under-segmentation occurred, with all MSD values below 2.5 mm.

Conclusions:

• CT-based automatic segmentation using OncoStudio provides clinically acceptable delineation of rectal spacers without MRI.
• The method maintains contouring quality of surrounding organs, comparable to MRI-based references.
• CT-only automatic contouring can potentially reduce MRI use and registration uncertainties in prostate radiotherapy.