Virtual doublestaining using non-linear registration

Virtual doublestaining using non-linear registration
Contact for more information: 

Onno de Boer
Rien van Leeuwen

Immunohistochemical multiple staining is a frequently used technique in (experimental) histopathological studies to investigate the presence of multiple immune cells or tissue components in one and the same tissue section. Because this type of immunestaining requires a lot of expertise and has many limitations, we developed a simplified method, so called ‘virtual doublestainings’.

Virtual doublestaining
Virtual doublestaining using non-linear registration

Using the virtual doublestainings methodology, tissue sections are immunohistochemically stained and a digital scan of the specimen is made using the Philips UFS. Next, the dyes and immunereagents are eluted from the specimen, and the tissue section can be stained again with a different antibody. Using digital techniques these scans can be combined into one image, and co-localisation and/or co-expression of the different markers can be studied. The figure shows an example of such a staining, a human thrombus specimen stained for neutrophils (1A), neutrophil extracellular traps (NETs) (1B), and fibrin (1C).

Virtual doublestaining using non-linear registration

Figure (1D) shows a composite image, combining the above figures 1A, 1B, and 1C: neutrophils in green, NETs in blue, and fibrin in red.

Image registration
Virtual doublestaining using non-linear registration

Image registration is an important and necessary part of the workflow creating virtual doublestaining images. Simple rigid or elastic registration algorithms are not sufficient for studying immunohistochemical co-expression.  Figure 2(A) shows 2 merged images (high power detail from the figures 1(A) and 1(B) above) which were aligned using rigid registration. Note that some cells are properly aligned (arrowhead), while at short distance other cells are not (arrow).

The registration software, which is necessary for improved alignment of these images, is under development in cooperation with Philips. This registration problem demands a non-linear solution. The solution provides a balance between (i) the efficiency of the processing of the large amount of image data, and (ii) the accuracy required for proper alignment of the tissue structure detail. The result of non-linear registration is shown in Figure 2(B). Note that the cells are significantly better aligned (arrowhead).

24/02/2015