Computational Approach to Define Histological Phenotypes Across Colitis Mouse Models

• Main Author: Kobayashi, Soma
• Format: Dissertation
• Language: English
• Published: ProQuest Dissertations & Theses 01-01-2023
• Subjects: Bioinformatics; Histology; Public health
• ISBN: 9798379768775

Colitis mouse models have played an integral role in advancing our knowledge of human Inflammatory Bowel Disease. Colitis mouse models develop specific histological abnormalities that are likely reflective of the molecular and immune changes occurring during disease processes. Histology captures a visual snapshot in time of the cascade of events that have occurred to that point. However, histological phenotypes are underutilized in colitis mouse model characterizations and assessment of translational potential. This is due in part to the absence of methods that detect and quantify histologic findings across mouse models. Convolutional neural networks learn to associate visual patterns with image labels and are well suited for this task. As such, we developed a pipeline that detects classes of histology, both normal and abnormal, from hematoxylin and eosin (H&E)-stained colons across multiple mouse models. A multi-mouse model approach offers potential to capture a broader range of histological phenotypes to better model phenotypic clinical heterogeneity. Here, we present the initial foundation of a pipeline that maps immune marker positivity through immunohistochemical staining to computationally-defined H&E histological phenotypes. While the colitis mouse models included in this study are insufficient to fully capture the diversity of clinical heterogeneity, they provide distinct, variable enrichment of histological findings. We have thus been able to explore the value of detecting and quantifying histology across mouse models as a characterization and phenotyping method. Our work here advocates for the need and potential of these approaches, while providing a framework for future incorporation of additional mouse models in our pipeline to further augment capacities.