Authors:

Madison Sarah Loiselle

Date Created:

Not specified

Course Title:

Professor:

Not specified

About Paper:

Analysis of histopathology slides is a time consuming and laborious task for pathologists. Deep learning models have the potential to aid pathologists in making swifter and more accurate diagnoses for patients. However, a roadblock in the field of clinically applicable veterinary deep learning is data scarcity. The purpose of this research is to assess the performance of deep learning models for digital pathology that are trained on human data, tested on veterinary data and vice versa. If successful, this approach could provide resources to the veterinary setting, in which less curated data is available for model development. In addition, our results are expected to have implications for translational research by connecting histomorphological patterns across species. As a proof-of-principle, we will evaluate two different digital pathology use cases: (1) Classification of healthy tissue types from different organs and (2) Classification of cancer entities. Veterinary (canine, murine) and human data were gathered from a wide variety of publicly available datasets. At this point, the data for first and second use cases have been preprocessed using the Solid Tumor Associative Modeling in Pathology (STAMP) pipeline and solely Macenko color normalization. As a next step, the classification models will be trained on human data and validated using 5-fold cross-validation on human data, and subsequently tested on an external set of animal data. This approach will subsequently be repeated in reverse form; with models being trained on veterinary data and tested on both veterinary and human data. The performance of each model will be evaluated with standard classification metrics including AUC, accuracy, specificity, and sensitivity. Overall, we expect this study to provide an understanding of the cross-species applicability of digital pathology models between humans and animals. Keywords: Histopathology; Deep Learning; Pathology

Source:

Purdue University / 2024

Topics:

No topics listed

Co-authors:

Madison Sarah Loiselle