Veterinary oncology faces unique challenges in detecting and treating cancer in dogs. Traditional diagnostic approaches like biopsies and exploratory surgery are time-consuming and costly, creating a need for non-invasive, inexpensive biomarkers. Effective tumor biomarkers should enable early detection, predict treatment response, and monitor progression.
Thymidine kinase 1 (TK1), a pyrimidine salvage pathway enzyme, has been associated with various cancers including lymphoma, leukemia, and solid tumors, though its role in canine solid tumors remains understudied. Recently developed anti-TK1 antibodies against dog TK1 have shown higher sensitivity in diagnosing solid tumors.
Canine C-reactive protein (cCRP) levels are associated with infection, tissue injury, and cancer. The combination of TK1 activity and cCRP levels forms a neoplastic index that has proven valuable for screening and monitoring malignancies. Machine learning and AI have revolutionized human healthcare by detecting diseases from images, predicting outcomes, and optimizing treatments. This study developed a tool using age, TK1 concentration, and cCRP to detect canine cancers early, classifying dogs as positive or negative for cancer with testing on unused data.
The study included 220 samples from dogs with various tumors (53 lymphoma cases, 167 solid tumors including 26 hemangiosarcoma, 47 histiocytic sarcoma, 26 osteosarcoma, 40 mastocytoma, and 28 mammary tumors) plus 67 healthy controls. Samples came from the Flint Animal Cancer Center and the Swedish University of Agricultural Sciences.
The resulting Alertix-CRI model demonstrated superior predictive capacity with an AUC of 0.97, surpassing previous values of 0.88. It identified 93% of malignant samples and showed exceptional precision for common aggressive tumors including mastocytoma, mammary tumors, lymphoma, and hemangiosarcoma. Compared to alternatives like the Nu.Q® Vet Cancer Test (AUC 69%, sensitivity ~50%), Alertix-CRI achieved 98% AUC with 90% sensitivity while maintaining high specificity – crucial for avoiding false positives and unnecessary diagnostics. The model maintained 90-100% precision across cancer types, accurately predicting malignancy when positive.
Cancer risk increases with age due to cumulative environmental exposures, genetic mutations, and declining cellular repair mechanisms. The Alertix-CRI successfully integrated age alongside TK1 and cCRP biomarkers. This feasibility study demonstrates the potential of combining these factors in a machine-learning model for accurate, non-invasive cancer detection in dogs.
