Imaging & BioInformatics

2018

Segmentation of Zebrafish larvea from bright-field microscope images


Yuanhao Guo, Zhan Xiong, and Fons J. Verbeek (2018)
An efficient and robust hybrid method for zebrafish segmentation in the bright-field microscope images.
Machine Vision and Applications. [Doi]



This page illustrates the segmentation method that we have employed for the processing of images from the VAST-BioImager.

1. Pipeline of hybrid method for segmentation VAST images

2018
Pipeline for Hybrid segmentation for zebrafish

2. Results of various segmentation methods for zebrafish in VAST

2018
Illustration of segmentation results

The blue bounding box indicates the location of accurate segmentation;
The red bounding box indicates the location of inaccurate segmentation.
(A) The segmentation obtained by the mean shift algorithm.
This method produces a whole shape representation of the zebrafish but fails in sensitivity to the edges.
(B) The segmentation obtained by the improved level set method.
This method fails to detect the transparent regions mostly found in zebrafish tail area.
(C) An accurate segmentation is obtained from a hybrid approach including the previous two methods with some refined processing.

3. Results of hybrid segmentation method for images from VAST-BioImager

The animations represent the segmentations of images with zebrafish larvae positioned in any axial-view.
The axial-view images are acquired by the VAST-BioImager and a standard bright-field microscope.

Segmentation of images from the VAST camera (ProSilica),
where dfp stands for days post fertilization, an indication of developmental age.

3 dpf

2018
2018

4 dpf

2018
2018

5 dpf

2018
2018

Segmentation of images from the microscope camera (DFC 450c).

4. Configuration for the hybrid method

2018

a. Spatial and color feature space window size for the mean shift algorithm,
where (hr,hs)= (20,20)

2018

b. The improved level set method and its parameters,
where m=150 and μ=5

2018

c. First curve evolution and its parameters,
where Δt1=2 and T1 =9;

2018

d. The curve post-refinement and its parameters,
where, Δt2=0.2 and T2 =4;

A complete elaboration and analysis can be found in the paper.
For additional information, contact Yuanhao Guo or Fons Verbeek,

Imaging & BioInformatics, 2018/2025 fjv