Computational diagnosis and risk evaluation for canine lymphoma

Area of Science:

• Veterinary Medicine
• Biomarker Discovery
• Machine Learning in Diagnostics

Background:

• Canine lymphoma diagnosis relies on clinical signs and invasive procedures.
• Biomarker-based blood tests offer a less invasive diagnostic approach.
• Monitoring treatment response and detecting recurrence are critical for canine lymphoma management.

Purpose of the Study:

• To develop and validate a machine learning-based diagnostic system for canine lymphoma using blood biomarkers.
• To evaluate the efficacy of machine learning models for differential diagnosis, screening, and risk estimation.
• To create a user-friendly computational tool for monitoring lymphoma recurrence in dogs.

Main Methods:

• Analysis of clinical data and machine learning model selection (decision trees, kNN, radial basis functions).
• Comparison of various pre-processing techniques for optimal feature selection.
• Development of a web-accessed software for computational diagnosis and risk evaluation.

Main Results:

• Differential diagnosis achieved 83.5% sensitivity and 77% specificity using CRP, Haptoglobin, and clinical symptoms.
• Screening task yielded 81.4% sensitivity and >99% specificity with a decision tree model.
• The system detects lymphoma recurrence up to two months before clinical signs appear.

Conclusions:

• Machine learning models applied to CRP and Haptoglobin levels provide an effective tool for canine lymphoma diagnosis, screening, and risk assessment.
• The developed software aids in early detection of lymphoma recurrence, improving patient monitoring.
• Biomarker-based approaches combined with machine learning show significant promise for veterinary oncology.