Using AI to assess Tertiary Lymphoid Structures in H&E and Multiplex IF stained Liver tissue
Following on from our previous article on Tertiary Lymphoid Structures (TLS), in which OracleBio clinical pathologist Gabriele Kohnen discussed the potential role of Tertiary Lymphoid Structures in normal and diseased tissues, this article describes some recent work where we utilized quantitative digital pathology techniques to assess TLS in haematoxylin and eosin (H&E) and multiplex immunofluorescence (mIF) stained liver tissue.
The first objective of this study was to detect and count TLS present in each liver section utilizing image analysis. To do this, we developed an AI deep learning algorithm within Visiopharm (digital pathology software).
Under guidance from our clinical pathologists, image analysis scientists digitally annotated visible TLS present across a group of training sample images stained with H&E. The number of sections and the number of training regions within sections used to train the algorithm can depend on the heterogeneity of TLS and level of general disease or tissue inflammation observed across the whole study sample set.
In the current study, TLS were evident within liver tissue, ranging from simple T and B cell aggregates through to lymphoid follicle formation. Input from our clinical pathologists on the range of TLS structures to include within the training set was crucial to the performance of the deep learning App. Examples of TLS detected by the AI App are shown in Figure 1.
[Above - Figure 1]: Examples of TLS within H&E-stained liver sections and the corresponding deep learning classification overlays. Panel 1 – Examples of H&E sections containing TLS. Panel 2 – Examples of the corresponding deep learning classification of TLS and liver tissue.?
Once TLS have been identified, the classification can then be used to generate specific image analysis data, for example:
? Number of TLS per mm2 of tissue
? Area of TLS (as % of whole tissue area)
? Average size of detected TLS (mm2)
? TLS Spatial distribution:
The next objective would be to quantify the cellular content of detected TLS, using the multiplex-IF marker staining panel to support segmentation and classification of the various cell types present. TLS can contain a variety of cell types including B and T lymphocytes, macrophages, endothelial cells, histiocytes, and follicular dendritic cells. In addition to quantifying the number of cells, multiplex staining combined with digital image analysis can enable a detailed assessment of the functional/activation status and spatial resolution of key cell types.
In the current study, a mIF assay was developed to highlight the expression of multiple antigens within liver tissue. The assay included markers for CD3, CD4, CD8, CD20, and CD68. Algorithms developed within Visiopharm can segment, phenotype, count, and spatially resolve individual cells present within the detected TLS, taking into account the following considerations:
Figure 2 highlights a TLS from one liver sample containing a large CD20 positive B-cell core with CD3/CD4 and CD3/CD8 T-cells around the periphery.
领英推荐
[Above - Figure 2]: Example of a TLS within an mIF stained Liver section. Panel 1 – Magnified areas capturing a single TLS showing examples of CD3 (red), CD4 (purple) and CD8 (green). Nuclei are shown in blue (DAPI). Panel 2 - Magnified areas capturing a single TLS showing examples of CD3 (red), CD20 (cyan), CD4 (purple) and CD8 (green). Nuclei are shown in blue (DAPI). Panel 3 – Examples of the formed positive cells created using the image analysis algorithm.
Apps can then be applied to generate specific image analysis data including, for example:
Summary & Conclusion
Image analysis can be used to characterize TLS and classify the expression of established TLS-defining cellular components within liver tissue samples. ?
Looking forward, image analysis of TLS within other diseases can also support a better understanding of the underlying mechanisms contributing to disease progression.
For example, in Oncology, image analysis techniques can evaluate the presence of TLS within a tumour and determine the average distance of TLS to the invasive tumour front. It may also be possible to determine TLS distribution in metastatic deposits. Studies evaluating therapeutic response within a tissue may also include the effect of the therapy directly on TLS content.
In conclusion, the combination of clinical pathologist expertise and quantitative digital pathology techniques can enable a deeper understanding of the role of TLS within various diseases and provide cell-level data that helps deliver detailed insights into the biological activities and therapeutic potential of targeting TLS to deliver new treatment paradigms.
We hope you enjoyed both perspectives on TLS — Part 1 from a clinical pathologist PoV, and this latest Part 2 from an image analysis scientist's PoV.
We'd love to hear your thoughts and questions in the comments!
This post is written by OracleBio’s Senior Image Analysis Scientist, Karen McClymont.
With a PhD in Biochemistry and her active involvement & support in the analysis of some of OracleBio’s most complex studies, Karen plays a key role within the Image Analysis team and its continued development.