Spatially varying accuracy and reproducibility of prostate segmentation in magnetic resonance images using manual and semiautomated methods

Abstract

The main conclusions of this study were that (1) the semiautomated approach reduced interobserver segmentation variability; (2) the segmentation accuracy of the semiautomated approach, as well as the accuracies of recently published methods from other groups, were within the range of observed expert variability in manual prostate segmentation; and (3) further efforts in the development of computer-assisted segmentation would be most productive if focused on improvement of segmentation accuracy and reduction of variability within the prostatic apex and base.

Postediting prostate magnetic resonance imaging segmentation consistency and operator time using manual and computer-assisted segmentation: multiobserver study

“…Then, we used the approach described in detail in Ref. 17 for prostate boundary localization. At a high level, for each slice, we defined 36 equally spaced rays emanating from the center point.…”

“…The training step is identical to the training step used in the semiautomatic method and fully described in Ref. 17. At a high-level, during training, the algorithm learns (1) the local appearance of the prostate border through extracting a set of locally defined mean intensity image patches and (2) the prostate shape variation across the training image set by extracting 2-D statistical shape models for the prostate at each axial cross-section.…”

“…We have previously reported on a semiautomatic segmentation algorithm and this method is based on prostate shape and appearance models learned from a training set. 17 Segmentation is performed in two steps: coarse localization of the prostate, followed by 3-D segmentation boundary detection and refinement. In the semiautomated approach, coarse localization is performed by the operator with four mouse clicks requiring ∼30 s of user interaction time.…”

“…Typically, reported error metrics include the mean absolute distance (MAD) between the boundaries of the automatic and manual segmentations, and/or the Dice similarity coefficient (DSC). Reported MAD values range from 1.5 to 3.4 mm, [17][18][19][20] and reported DSC values range from 82% to 91%. [17][18][19][20][21] Reasons for the range of different error values reported include algorithm design, the use of single-operator manual reference segmentations for validation in most studies, and the use of different imaging datasets.…”

A semiautomatic algorithm for three-dimensional segmentation of the prostate on CT images using shape and local texture characteristics

“…The positive values of Δ V indicate over segmentation, and the negative values of Δ V indicate under-segmentation. For more details regarding the metric calculation see [17-19]…”