Boundary Coding Representation for Organ Segmentation in Prostate Cancer Radiotherapy

Area of Science:

• Medical Imaging
• Radiotherapy
• Computational Anatomy

Background:

• Accurate segmentation of prostate and organs at risk (OARs) in male pelvic CT images is crucial for effective prostate cancer radiotherapy.
• Challenges include unclear organ boundaries and significant shape variations, hindering precise segmentation.
• Existing methods using direct boundary representations offer limited performance improvements.

Purpose of the Study:

• To develop a novel Boundary Coding Network (BCnet) for learning discriminative organ boundary representations.
• To utilize these learned representations as context information to enhance organ segmentation accuracy.
• To improve the segmentation of prostate and OARs in male pelvic CT images for radiotherapy planning.

Main Methods:

• A two-stage learning strategy is employed within BCnet.
• Stage 1: Boundary coding representation learning using sub-networks trained on dilation/erosion masks to capture spatial-semantic context near boundaries.
• Stage 2: Organ segmentation utilizing the learned boundary coding representation as context alongside image patches.

Main Results:

• The BCnet method demonstrated superior performance on a large, diverse male pelvic CT dataset.
• Achieved improved accuracy in segmenting the prostate and OARs compared to state-of-the-art methods.
• The learned boundary coding representation proved effective in guiding the segmentation process.

Conclusions:

• The proposed BCnet effectively learns discriminative boundary representations for improved organ segmentation.
• This approach enhances the accuracy of prostate and OAR segmentation in male pelvic CT images.
• BCnet offers a promising advancement for radiotherapy planning and treatment delivery.