Histopathological Classification of Canine Cutaneous Round Cell Tumors Using Deep Learning: A Multi-Center Study

Area of Science:

• Veterinary Pathology
• Computational Pathology
• Artificial Intelligence in Medicine

Background:

• Canine cutaneous round cell tumors (RCT) present diagnostic challenges in veterinary pathology.
• Computer-aided diagnostic systems can enhance accuracy and consistency, especially for high-volume screening.
• Automated analysis is crucial for improving efficiency and reducing errors in veterinary diagnostics.

Purpose of the Study:

• To develop and validate ARCTA (Automated Round Cell Tumors Assessment), a fully automated deep learning algorithm for classifying canine cutaneous RCT and grading mast cell tumors.
• To assess the performance of ARCTA in terms of accuracy and speed for histopathological image analysis.
• To establish a novel, automated tool for veterinary diagnostic applications.

Main Methods:

• Development of a deep learning algorithm (ARCTA) trained on 416 canine RCT and 213 mast cell tumor histopathological images.
• Utilized a fully automated approach for image classification and tumor grading.
• Evaluated algorithm performance on a dedicated test set.

Main Results:

• ARCTA achieved high accuracy in RCT classification (91.66%) and mast cell tumor grading (100%) on the test set.
• The algorithm demonstrated a fast average computational time of 2.63 seconds per image.
• Identified specific misclassifications, including histiocytomas (training set) and melanomas (test set).

Conclusions:

• ARCTA is the first fully automated algorithm for canine tumor classification and grading in veterinary histopathology.
• The algorithm provides a rapid, accurate, and consistent diagnostic aid for canine cutaneous tumors.
• This technology has the potential to significantly improve the efficiency and reliability of veterinary tumor diagnostics.