A Fusion Biopsy Framework for Prostate Cancer Based on Deformable Superellipses and nnU-Net

Area of Science:

• Medical Imaging
• Urology
• Computer-Aided Diagnosis

Background:

• Fusion biopsy combines MRI and TRUS for targeted prostate cancer biopsies.
• Accurate prostate segmentation from both imaging modalities is crucial for fusion.
• Current automatic segmentation methods, especially for TRUS, face challenges like noise and unclear boundaries, often requiring large datasets.

Purpose of the Study:

• To develop a novel, data-efficient method for automatic prostate segmentation from TRUS images.
• To integrate this method with MRI segmentation for a comprehensive image fusion tool.
• To facilitate accurate, targeted prostate biopsies through improved image analysis.

Main Methods:

• Proposed a novel optimization formulation using superellipses for accurate prostate shape representation and segmentation from TRUS images, minimizing annotation requirements.
• Utilized the nnU-Net framework for automatic prostate gland segmentation from MRI images.
• Developed a customized registration algorithm to fuse segmented contours from both MRI and TRUS.
• Integrated these components into a graphical user interface for clinical application.

Main Results:

• The superellipse-based method provides accurate prostate segmentation from TRUS without extensive training data.
• The integrated system successfully fuses MRI and TRUS prostate contours.
• The tool aids physicians in performing targeted prostate biopsies.

Conclusions:

• The proposed superellipse formulation offers a robust and data-efficient solution for TRUS prostate segmentation.
• The developed image fusion tool enhances the precision of targeted prostate biopsies.
• This approach represents a significant advancement in computer-aided diagnosis for prostate cancer.