Utilization of deep learning to accurately determine cell of origin in canine follicular and medullary thyroid carcinomas on routinely processed, H&E-stained tissue sections

Canine thyroid carcinomas (CTCs) are a common endocrine malignancy that requires histopathologic examination with costly and time-consuming immunohistochemistries for definitive diagnosis, which can delay the time to treatment beyond surgical excision. A significant diagnostic challenge arises in differentiating compact follicular thyroid carcinomas (FTCs, derived from follicular cells) and medullary thyroid carcinomas (MTCs, derived from medullary cells) with routine hematoxylin and eosin (H&E) staining. Literature suggests these have similar clinical outcomes; however, publications often do not distinguish between compact FTCs and MTCs. The primary objective of this project is to develop and validate an artificial intelligence (AI) deep learning algorithm that can accurately determine the cell of origin (follicular or medullary) in CTCs without the use of ancillary immunohistochemical (IHC) stains. The primary hypothesis is that the algorithm can accurately determine the cell of origin in CTCs on routine H&E-stained histopathology slides. A secondary objective includes reviewing and comparing demographic information between follicular-derived or medullary-derived CTCs and between several follicular subtypes and medullary carcinomas, while tertiary objectives include evaluating the ability of pathologists to correctly identify compact follicular thyroid carcinomas from medullary thyroid carcinomas on H&E alone and comparing their diagnoses to the interpretations of the algorithm’s output. This study confirmed the primary hypothesis that it is feasible to determine the cell of origin for CTCs by an AI model. Additional demographic information with comparisons between the different types of CTCs is provided, and the need for ancillary diagnostics in differentiating compact FTCs and MTCs is re-iterated. For this model, most of the convoluted neural nets are ready for use in conjunction with interpretation by a pathologist. Additional work is needed on the convoluted neural net that is for differentiating between FTC subtypes and MTCs. The use of this AI model could expedite the workflow for the pathologist and allow for rapid definitive diagnosis between compact FTCs or MTCs in dogs on routine H&E-stained slides which would translate to a decreased financial burden for the client, decreased time to diagnosis for the patient, and ultimately decreased costs of reagents and supplies for the diagnostic lab in future cases. Additionally, a successful algorithm could be applied to prospective or, potentially past studies, with whole slide images (WSIs) of CTCs to establish consistent differentiation between FTCs and MTCs when IHCs are not available. This in turn allows for more reliable interpretations of study results (e.g., a response to treatment or the patient outcome) or for re-evaluation of results and conclusions derived from past studies where FTCs and MTCs were not distinguished. These applications could contribute to elucidating previously obscure differences in demographics, prognoses, effective treatment modalities, and factors contributing to tumorigenesis. Furthermore, rapid and inexpensive methods to determine the neoplastic cell of origin in CTCs will assist in paving the way for more swiftly customized and successful therapies (personalized healthcare, precision medicine), as is currently occurring in human medicine.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

VMS-Veterinary Clinical Medcne

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2022

Author dc:creator

• Athey, Jillian M.

Contributors dc:contributor

• Bailey, Keith
• Rudmann, Dan
• Baumgartner, Wes
• Selting, Kim
• Vieson, Miranda

• canine
• thyroid carcinoma
• artificial intelligience
• histopathology
• thyroid
• oncology
• follicular carcinomas
• medullary carcinoma
• supervised learning

Statement dc:rights

• Copyright 2022 Jillian Athey

Language dc:language

en, eng

Handle dc:identifier

https://hdl.handle.net/2142/116040

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F116040