Abstract: BACKGROUND: Dogs with inflammatory protein-losing enteropathy have a guarded prognosis, with death occurring as a result of the disease in approximately 50% of cases. Although dogs treated with dietary therapy alone are significantly associated with a positive outcome, there is limited ability to differentiate between food-responsive protein-losing enteropathy (FR-PLE) and steroid-responsive protein-losing enteropathy (SR-PLE) at diagnosis. OBJECTIVE: To determine if a transfer learning computational approach to image classification on intestinal biopsy specimens collected at diagnosis is able to differentiate FR-PLE from SR-PLE. ANIMALS: Nine client-owned dogs diagnosed with inflammatory protein-losing enteropathy that were subsequently classified based on treatment response into FR-PLE (n=3) or SR-PLE (n=6). METHODS: A retrospective study using formalin-fixed, paraffin-embedded intestinal biopsy specimens collected during diagnostic investigations for protein-losing enteropathy at a referral veterinary teaching hospital. A machine-based algorithm was used on 20 and 51 images of intestinal biopsy specimens from dogs with FR-PLE and SR-PLE, respectively. RESULTS: Using the pre-trained Convolutional Neural Network (CNN) model with 70/30 training/test ratio for images, the model was able to differentiate intestinal biopsy images from dogs with FR-PLE and SR-PLE with an accuracy of 81.82%. CONCLUSIONS AND CLINICAL IMPORTANCE: Our results suggest that dogs with inflammatory PLE can be predicted as FR-PLE or SR-PLE at histopathologic diagnosis through computational approaches. This will help to ensure dogs with inflammatory PLE are prescribed the most appropriate treatment at diagnosis to ensure optimal response and outcome.
