Cell counting by Hough transform and SVM classifier

Description

Analyzing liver cross-section images is a manual and laborious task, slowing down critical research toward finding alternative cures for patients with end-stage liver disease.

In this project, we present methods to automatically count hepatocytes cells, count nuclei and classify liver vessel types, using input images of cell boundary and cell nuclei based on a dataset of 21 images.

Compared to a trained researcher, the methods are able count cells, segment overlapping nuclei, and classify vessel types, including portal vein, central vein, and bile duct with reasonable precision and accuracy.

Image name:1
Number of 1 nuclei cells: 611
Number of 2 nuclei cells: 96
Number of 3 nuclei cells: 13
Number of nucleis: 850

Image name:2
Number of 1 nuclei cells: 622
Number of 2 nuclei cells: 109
Number of 3 nuclei cells: 15
Number of nucleis: 885

Spatial transformer networks

Rigid motion transformation