Differentiating between inflammatory bowel disease (IBD) and small cell lymphosarcomas (lymphomas) in cats who clinically present with chronic gastrointestinal signs remains challenging. Recently, our lab has pioneered the use of ultrasound-based radiomics coupled with CBC/serum chemistry as a means of discriminating IBD from lymphoma, and other conditions, with an accuracy as high as 0.721, and separating cats that benefit from a biopsy from those who would not with an accuracy of 0.962. Unfortunately, the histologic and ultrasonographic features of IBD and lymphoma may overlap, and distinguishing features may be present for both conditions in the same cat. The ‘gold standard’ for diagnosis -biopsies- may be prone to uncertainties based on the location, type (pinch vs. whole section), timing, and biopsy interpretation. We propose to address challenges associated with discriminating IBD from lymphomas with a multi-modality approach.

First, we will train, validate, and test a deep neural network model based on N=150 digital whole slide images of biopsies from the small intestines of cats for discriminating IBD from lymphoma. In conjunction with the Cornell University Hospital for Animals (CUHA) Information Technology infrastructure and Animal Health and Diagnostic Center, we will retrieve the digital slides of 150 cats to build and evaluate two deep neural network models for computational pathology analysis. We will then quantify the extent of agreement between manual assessments and the deep neural network models.

Second, we will conduct a prospective longitudinal study of N=70 cats with IBD over 1.5 years. Cats recruited in CUHA will obtain endoscopic biopsies, US images, and CBC/serum chemistry data eight months after their diagnosis to determine whether there is a progression of IBD to lymphoma. Finally, with the cats enrolled in the trial, we will obtain US elastography measurements and compare the stiffness of the small intestines of cats presenting with IBD with those who do not.