3D prostate TRUS segmentation using globally optimized volume-preserving prior.

Area of Science:

• Medical Imaging
• Computational Anatomy
• Biomedical Engineering

Background:

• Accurate segmentation of 3D transrectal ultrasound (TRUS) images is crucial for 3D TRUS-guided prostate biopsy.
• Challenges in 3D TRUS image segmentation include US speckles, shadowing, and missing edges, hindering accurate prostate boundary delineation.

Purpose of the Study:

• To propose a novel convex optimization-based approach for extracting prostate surfaces from 3D TRUS images.
• To preserve a global volume-size prior during prostate segmentation.
• To develop an efficient numerical solver for the segmentation task.

Main Methods:

• A convex optimization framework was developed for prostate surface extraction.
• The combinatorial optimization problem was addressed using convex relaxation.
• A dual continuous max-flow formulation with a bounded flow conservation constraint was introduced.
• An efficient numerical solver was implemented on Graphics Processing Units (GPUs).

Main Results:

• The proposed method achieved a mean Dice Similarity Coefficient (DSC) of 89.5% ± 2.4%.
• Mean Absolute Distance (MAD) was 1.4 ± 0.6 mm, and Maximum Distance (MAXD) was 5.2 ± 3.2 mm.
• Volumetric Difference (VD) was 7.5% ± 6.2%, with segmentation completed in under one minute.
• Low standard deviation in accuracy metrics indicates high reliability.

Conclusions:

• The novel convex optimization approach provides accurate and efficient prostate segmentation from 3D TRUS images.
• The method effectively preserves volume-size priors, crucial for clinical applications.
• The GPU-implemented solver demonstrates significant speed advantages for practical use in prostate biopsy planning.