Segmentation | Algorithm | Description | Performance | Setback |
---|---|---|---|---|
Region growing e.g. GrowCut | Uses region-growing seed points to segment a tumor | Fast, low computational complexity, good reproducibility strong correlation with macroscopic tumor diameter | Segmentation errors due to boundary leakages, unsuitable for highly heterogenous tumors | |
GraphCut | Constructs an image-graph of voxels connected by weighted edges | Can deal with tumors with odd shapes and mosaic intensity | Over segmentation or undesired ROIs when there are artefacts | |
Water shed transformation | Segments tumor from parenchyma based on difference in gray scale intensity | Global segmentation | Over segmentation sensitive to poor tumor margins | |
Active contours, level-set and Live wires | Iteratively marks tumor contour from starting points on tumor edge | Faster than region growing methods | Rely on good initialization points and speed functions, sensitive to noise and poor tumor margins |