Overcoming image co-registration challenges for accurate tissue classification on single IHC-stained samples

In this article, image analyst Hannah Macleod dives deep into the nuances of classifying tissue ROIs on single IHC and explores the transformative potential of image co-registration; from understanding its foundation to evaluating its real-world challenges.

Join us on this microscopic journey, where we unfold the layers of tissue classification, one stain at a time.

What are the issues when trying to classify tissue ROI on single IHC?

Single-immunohistochemistry (IHC) staining defines and allows the visualisation of one cell type or tissue feature such as macrophages or alpha-smooth muscle actin (α-SMA). With single-IHC, as only one component of the tissue is defined, it can be challenging to distinguish regions of interest (ROIs), such as viable tissue from Necrosis, or Tumour and Stroma. In the absence of a Tumour or specific marker to identify one ROI from another, it can be particularly challenging to develop a tissue classifier algorithm on single IHC images.

What is image co-registration and how can it help?

To resolve this challenge, we can apply a technique called ‘image co-registration’, where a serial tissue section cut from the same tissue block as the IHC section is used to create the tissue classification using image analysis software and the classifier overlay is then aligned and applied to the single IHC section. The serial section can be stained with, for example, Haematoxylin and Eosin (H&E) or Pan Cytokeratin (PanCK) IHC. These stains help to better distinguish ROIs, such as Tumour, and are often used in the development of a tissue classifier algorithm for segmentation of ROIs that can be subsequently applied to the IHC images.

When working with serial images, we begin by using image analysis software such as Visiopharm or Indica Labs HALO to co-register the IHC images with their corresponding H&E or PanCK stained serial image. During the co-registration process, the image analysis software will align the images based on the location of the tissue on the scanned image. If the software cannot automatically align the serial sections accurately, the use of manual landmarks can be used to align the tissue on the images. This is achieved by placing ‘pins’ on tissue features, such as Tumour islands, that are present across all serial sections. The software then uses these pins to identify the corresponding areas of tissue on the serial sections and better align the images. The more tissue structures that are morphologically similar across the serial sections, the more landmarks can be used to help achieve the best alignment possible.

Once the images are successfully co-registered, we develop a classifier algorithm on the H&E (or PanCK) stained images, using the difference in colour and texture features of the ROIs to appropriately segment the ROIs. The classifier is then integrated into custom IHC analysis algorithms to produce analysis results such as total positive cells per ROI on the IHC images. When a classifier is incorporated into an analysis algorithm with the appropriate co-registration settings applied, the software will classify areas of tissue on the IHC image based on the classification of the corresponding areas on the H&E image. This means that following successful alignment of the serial sections, there will be accurate segmentation of the desired ROIs (i.e. Tumour/ Stroma) on the IHC images if the tissue structures are in the same location as the corresponding H&E serial section, (Figure 1).

1 – Importance of using serial sections

Accurate segmentation of ROIs on the serial IHC sections requires the H&E and IHC serial sections to have a similar tissue morphology. If, however, the sections collected from a tissue block are not serial, the morphology of the tissue on the IHC images is likely to be different from the corresponding H&E image. Due to the different tissue morphology, when the classifier is incorporated into the analysis algorithms, areas of tissue may be misclassified and lead to inaccurate results, as demonstrated in Figures 2 and 3.

[Figure 3 contd.] In more severe circumstances, the morphology of the two sections is completely different and could not be improved with the use of several landmarks during the co-registration process, resulting in misclassification of the tissue and inaccurate analysis results.

To achieve similar morphology between the tissue of IHC and H&E (or PanCK) serial images, the sectioning and staining of the images to be used for subsequent tissue classification should be planned in advance. Planning ahead to use the middle section of the sections collected to develop the classifier on (i.e., the H&E section), and using the sections on either side of the middle section for the IHC markers will ensure that each IHC marker is close to the H&E section.

Tissues that are sectioned close to one another in the tissue block are more likely to have similar morphology between the sections. Alternatively, after every IHC section a corresponding H&E section could be sectioned, which also ensures that if a H&E section is damaged or unusable, there are other H&E sections close enough in the tissue block to the IHC sections to be used.?

2 – Importance of image alignment for accurate co-registration

Another issue can arise when the serial tissue sections are in different locations on their slide or are at different rotational orientations to each other. This makes it challenging for image analysis software to automatically align the tissues correctly during co-registration. To help the software align the tissue better, we identify morphological structures that are similar across the serial sections and add landmarks to these structures.

Although this is a relatively simple problem to fix, a large quantity of landmarks may be needed to align the tissue which can be time-consuming, especially for larger image sets. However, on occasion, this issue cannot be fixed using landmarks, i.e., if large areas of damage do not allow for sufficient landmarks, accurate co-registration of serial sections may not be possible.

The problem therefore needs to be corrected at the pre-analytical stages of tissue sectioning and slide mounting.

When mounting serial tissue sections onto slides, it is important to mount tissue sections at the same orientation (or within a 45-degree angle) and then scan these tissue sections at the same magnification/resolution settings to capture the same area of tissue. This can help the software automatically align the tissue accurately during co-registration, reducing the need to use landmarks during co-registration.

3 – The importance of choosing the correct stain for serial section

Finally, to develop a classifier to segment desired ROIs, it is important that the correct stain for the serial section is used. Often, tissue sections are stained with H&E as it can distinguish extracellular matrix components and certain cell types, such as Tumour cells. H&E staining is also useful to differentiate areas of artefact to be excluded from analysis, such as Necrosis or red blood cells. A stain that also defines Tumour ROI accurately is PanCK, which clearly distinguishes Tumour cells from Stromal cells. If the desired ROI is composed of collagen, stains such as Picrosirius Red Staining (PSR) or Masson’s Trichrome stain can be used to define ROIs such as fibrotic lesions.

The use of serial section staining not only applies to the identification of the desired ROIs but can also be used to distinguish areas to be excluded from analysis.

In our most recent case study, which can be found here. PSR-stained images were used to exclude areas on corresponding α-SMA serial sections. As prominent vasculature structures can adversely impact the quantification of α-SMA in fibrotic regions, these vascular areas were removed on the PSR-stained images using an AI classifier. The PSR images were then co-registered to serial IHC α-SMA sections, allowing the transfer of the AI classifier to the α-SMA images to exclude prominent vasculature on the IHC image set.

How can OracleBio’s services help?

If you are currently planning a histology study to evaluate single IHC staining in a particular tissue, we can advise on the potential need for a serial section to support more accurate ROI classification.

OracleBio uses both Visiopharm and Indica Labs software, allowing us to support image co-registration and detailed downstream analysis of your histology study samples. Our tips above also provide best practices for sectioning, staining and image capture to ensure your study performs optimally and delivers data to the highest quality.

Get in touch with our experts to discuss how we can help with your research.

This post is written by OracleBio Image Analysis Scientist, Hannah MacLeod.