Automatic segmentation of pelvic floor ring on 3D ultrasound

A 3-dimensional ultrasound image of a woman’s pelvic floor region, in which a pelvic floor ring is present, is analysed for the automatic detection and segmentation of the ring. The segmentation comprises image pre-processing, position analysis and rotation analysis. Several approaches are presented and discussed. The overall best performing result is achieved using a filter chain comprising a 1-dimensional Gaussian blur, LoG filter and a morphological opening operation to pre-process the images, blob detection and analysis to determine the position, and a cross-correlation based rotational analysis. While the location of the ring crosscut can be accurately determined in approximately 70% of cases, the orientation can not be determined with sufficient confidence