UV spectroscopy associated with machine learning as a complementary tool for the diagnosis of Ehrlichia canis in dogs

Abstract

Canine ehrlichiosis, caused by Ehrlichia canis, is a tick-borne disease with a global distribution that significantly affects the clinical and epidemiological landscape for dogs. Although laboratory diagnosis is essential, current diagnostic methods exhibit limitations in sensitivity, specificity, and cost-effectiveness. This study evaluated the efficacy of ultraviolet (UV) spectroscopy combined with machine learning techniques to differentiate serological samples that are positive and negative for E. canis. A total of 46 canine serum samples, classified via Enzyme-Linked Immunosorbent Assay/Dot-Enzyme-Linked Immunosorbent Assay, were analyzed. Spectra (200–300 nm) underwent dimensionality reduction through principal component analysis, and classification using supervised algorithms, specifically support vector machine. In the binary classification of E. canis positive and negative sera, a linear SVM model (5 PCs) achieved 89.3% accuracy via Leave-One-Out Cross-Validation (LOOCV), yielding 85.7% sensitivity and 92.9% specificity. In the independent test set, the model reached 100% accuracy, sensitivity, and specificity, demonstrating high robustness and potential for diagnostic screening. The findings indicate that UV spectroscopy, in conjunction with machine learning, may serve as an effective complementary tool for diagnosing E. canis.