Deep learning-based image analysis
Deep learning-based image analysis (artificial intelligence) shows differences in the number and distribution of T lymphocytes in different areas of the colon mucosa in ulcerative colitis and Crohn's disease
Deep learning technology (artificial intelligence) can be used to develop applications for image analysis in pathology, such as the identification of cancer and the quantification of cells in different areas by immunohistochemical examination. This provides fast and reproducible measurements on an almost unlimited number of tissue sections. Findings that prove to have clinical relevance can lay the foundation for the development of new diagnostic tools in pathology.
We have developed deep learning models to distinguish the epithelium from the rest of the tissue in whole-sections of colonic mucosa from patients with inflammatory bowel disease (IBD). We have used openly available and free software; Annotation is done in QuPath, training of deep neural networks in DeepMIB, and visualization of trained models on full sections with FastPathology.
The results show that our deep learning-based models provide accurate segmentation of epithelium and can be used in research on the epithelial microenvironment in tissue sections. All source code, training video, and an annotated dataset of 251 colon sections have been made openly available to contribute to further research. https://pubmed.ncbi.nlm.nih.gov/35155486/
Histological changes in the intestinal mucosa in IBD provide information on disease activity and healing and are predominantly assessed qualitatively by the pathologists. Deep learning-based analyzes make it possible to extract complex quantitative data from tissue sections. Such data can be used to reveal differences in inflammation patterns that are otherwise unavailable to the human eye.
We used our self-developed deep learning model to quantify T lymphocytes in the colon mucosa. In particular, we focused on gammadelta (GD) T lymphocytes that have important immunological functions in mucous membranes. We looked at the number and distribution of these cells in the epithelium and in the rest of the colon mucosa in IBD with active and inactive disease and compared with healthy controls. The results showed that the mucosa in inactive Crohn's disease had far fewer GD T lymphocytes than inactive ulcerative colitis, but at the same time many more T lymphocytes in all parts of the mucosa. In active disease, the number of GD T lymphocytes in the epithelium was reduced in the ulcerative colitis patients, while the Crohn's patients had an unchanged low number. We also found that the number of GD T lymphocytes was lower in those who received treatment with corticosteroids. Age, gender or disease duration had no effect on the number of GD T lymphocytes. We also found no correlation between the relative proportion of GD T lymphocytes in the blood and in the mucosa.
The study reveals significant differences in the number and distribution of T lymphocytes and the subset of GD T lymphocytes in the colonic mucosa between patients with ulcerative colitis, Crohn's disease, and healthy controls. This was especially true in the epithelium itself, which is an important area for immunological response in the mucosa. Our results are an important contribution to a better understanding of the pathology of the two types of IBD.