Nghiên cứu hồi cứu về ứng dụng mô hình học máy trong chẩn đoán bệnh thận mãn tính ở mèo dựa trên dữ liệu sinh lý và sinh hóa máu

Abstract

The study was conducted from February 2025 to August 2025 at veterinary clinics in Ho Chi Minh City, Vietnam. This retrospective study analyzed hematological and blood biochemistry data from 60 cats (30 healthy and 30 with chronic kidney disease-CKD) to evaluate the efficacy of machine learning (ML) models in diagnosing CKD and identifying biomarkers. Parameters included red blood cell count (RBC), hemoglobin (HGB), hematocrit (HCT), white blood cell count (WBC), neutrophil count (Neu#), eosinophil count (Eos#), lymphocyte count (Lym#), monocyte count (Mon#), platelet count (PLT), alanine aminotransferase (ALT), aspartate aminotransferase (AST), blood urea nitrogen (BUN), creatinine (Crea), total protein (TP), albumin (ALB), and glucose (GLU). Data were preprocessed and split into 80% training and 20% testing sets. Seven ML models [Logistic Regression, K-Nearest Neighbors (KNN), Decision Tree, Random Forest, Gradient Boosting, Support Vector Machine (SVM), and Naive Bayes] were assessed using accuracy, sensitivity, specificity, F1-score, and area under the curve (AUC). The SVM, Logistic Regression, and KNN achieved 92% accuracy, with SVM yielding the highest AUC (0.98). Decision Tree, Random Forest, and Naive Bayes performed less effectively (75 - 83%), often producing false positives. Creatinine was the most influential biomarker (importance score: 1,178), followed by Mon#, AST, Neu#, and WBC. The Crea, WBC, Neu#, ALT, and AST showed significant differences (P < 0.01) between healthy and CKD cats. In brief, ML, particularly SVM, holds promise for supporting the early diagnosis of CKD and enhancing clinical decision-making.