3D Prostate Boundary Segmentation From Ultrasound Images Using 2D Active Shape Models

Boundary outlining, or segmentation, of the prostate is an important task in diagnosis and treatment planning for prostate cancer. This paper describes an algorithm for semi-automatic, three-dimensional (3D) segmentation of the prostate boundary from ultrasound images based on two-dimensional (2D) active shape models (ASM) and rotation-based slicing. Evaluation of the algorithm used distance- and volume-based error metrics to compare algorithm generated boundary outlines to gold standard (manually generated) boundary outlines. The mean absolute distance between the algorithm and gold standard boundaries was 1.09±0.49 mm, the average percent absolute volume difference was 3.28±3.16%, and a 5x speed increase as compared manual planimetry was achieved.