Prostate boundary segmentation from ultrasound images using 2D active shape models: optimisation and extension to 3D

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Biomedical Engineering

Background:

• Prostate segmentation is crucial for prostate cancer diagnosis and treatment planning.
• Accurate delineation of the prostate boundary from medical images is challenging.

Purpose of the Study:

• To develop and evaluate a semi-automatic algorithm for prostate boundary segmentation using 2D and 3D active shape models (ASM).
• To optimize ASM parameters for prostatic ultrasound and assess the accuracy and speed of the 3D segmentation algorithm.

Main Methods:

• Developed a 2D active shape model (ASM) algorithm for semi-automatic prostate segmentation from ultrasound images.
• Optimized ASM construction and image search parameters using minimum description length landmark placement.
• Extended the algorithm to 3D segmentation using rotational-based slicing.
• Evaluated the 3D segmentation accuracy using distance and volume error metrics against manual segmentation.

Main Results:

• Optimized ASM construction and image search parameters were identified.
• The 3D segmentation algorithm achieved an average mean absolute distance of 1.09+/-0.49 mm.
• An average percent absolute volume difference of 3.28+/-3.16% was observed.
• The algorithm demonstrated a 5x speed increase compared to manual segmentation.

Conclusions:

• The developed semi-automatic active shape model algorithm provides accurate and efficient prostate segmentation from ultrasound images.
• The optimized algorithm offers a significant improvement in speed and comparable accuracy to manual segmentation for clinical applications.
• This method holds promise for enhancing prostate cancer diagnosis and treatment planning.