Skip to main content
. 2019 Jan 11;5(1):14. doi: 10.3390/jimaging5010014

Figure 5.

Figure 5

(A) Readily implemented segmentation methods of the Fiji image analysis software were applied to images after pre-processing using GF. (B) It was found that intensity-based Otsu-thresholding delivered the most robust results in comparison to other methods, such as k-means clustering ((C); red asterisks: vessels not extracted), SRM (D), or fast-marching level set (E). (F) The same segmentation methods were applied to images after vessel enhancement with TF. Again, it was found that intensity-based Otsu-thresholding delivered the most reliable results (G). While k-means clustering (H), SRM (I), and the tested fast-marching level set implementation (J) resulted in an unsatisfactory segmentation. (K) Vascular volume quantification after GF showed no statistically significant difference between the tested segmentation methods. (L) Quantification of the vascular volume after TF showed a statistically significant difference between the assessed segmentation methods. Namely, TF Otsu vs. TF k-means p 0.0293, TF Otsu vs. TF SRM 0.0058, and TF k-means vs. TF SRM p > 0.9999. CoV was found to be lower after TF with 15.24% k-means and 30.00% SRM and only 10.71% Otsu-thresholding (n = 7, Mann-Whitney U test; representative images).