A Preliminary Study Assessing a Transfer Learning Approach to Intestinal Image Analysis to Help Determine Treatment...

Area of Science:

• Veterinary Medicine
• Computational Pathology
• Canine Gastroenterology

Background:

• Protein-losing enteropathy (PLE) in dogs carries a guarded prognosis, with high mortality rates.
• Differentiating between food-responsive (FR) and immunosuppressant-responsive (IR) PLE at diagnosis is challenging.
• Accurate differentiation is crucial for effective treatment selection in canine PLE.

Purpose of the Study:

• To evaluate a transfer learning computational approach for classifying canine duodenal biopsy images.
• To determine if machine learning can differentiate food-responsive PLE (FR-PLE) from immunosuppressant-responsive PLE (IR-PLE) at diagnosis.
• To assess the potential of AI in guiding initial treatment strategies for canine PLE.

Main Methods:

• Retrospective analysis of duodenal biopsy specimens from 17 dogs with PLE.
• Utilized a pre-trained Convolutional Neural Network model for image classification.
• Employed a 70/30 training/test data split for the machine learning model.

Main Results:

• The machine learning model achieved an accuracy of 83.78% in differentiating FR-PLE from IR-PLE.
• The algorithm analyzed lower magnification, higher resolution images of duodenal biopsy specimens.
• Successful differentiation was based on biopsy images collected at the time of diagnosis.

Conclusions:

• Machine learning holds promise for improving the diagnostic process in canine PLE.
• This computational approach can aid clinicians in selecting appropriate initial treatments for canine PLE.
• Further research into AI applications can enhance decision-making for complex canine gastrointestinal diseases.