CT male pelvic organ segmentation using fully convolutional networks with boundary sensitive representation

Area of Science:

• Medical Imaging
• Radiotherapy
• Computational Anatomy

Background:

• Accurate segmentation of pelvic organs in CT images is vital for effective prostate cancer radiation therapy.
• Challenges include unclear boundaries, patient variability, and imaging artifacts like bowel gas.
• Existing methods struggle with the complexity of male pelvic anatomy segmentation.

Purpose of the Study:

• To develop a novel automatic segmentation framework for prostate and surrounding organs in CT images.
• To enhance segmentation accuracy by incorporating boundary-sensitive representations.
• To improve the robustness and precision of organ segmentation for radiotherapy planning.

Main Methods:

• A three-module framework using fully convolutional networks (FCNs) was proposed.
• An organ localization model identifies candidate regions for segmentation.
• A boundary-sensitive representation model and multi-label cross-entropy loss were employed for robust segmentation.

Main Results:

• The proposed method demonstrated superior performance compared to baseline FCNs and other state-of-the-art techniques.
• Evaluated on a diverse dataset of 313 CT scans from prostate cancer patients.
• Achieved higher accuracy in segmenting the prostate, bladder, and rectum.

Conclusions:

• The novel segmentation framework effectively addresses challenges in male pelvic organ segmentation.
• The boundary-sensitive representation significantly improves segmentation accuracy and robustness.
• This method offers a promising advancement for automated radiotherapy planning in prostate cancer treatment.