College of Health Sciences / The Master Program of AI Application in Health Industry

Figure 1. Graphical abstract

  For newborns, bacteremia is an extremely dangerous silent killer. Because symptoms in neonates are often subtle, traditional blood culture tests can be slow and inefficient. These tests require several days to produce results, which often leads to missing the critical window for treatment. To overcome this medical challenge, the research team collaborated with Kaohsiung Veterans General Hospital to analyze thousands of clinical cases. By using machine learning, they aimed to uncover hidden patterns of infection within basic and cost-effective routine blood test data.

  This study utilized testing across multiple AI models to explore how different clinical indicators impact predictive accuracy. To ensure the AI possessed true generalizability for different patients rather than simply memorizing existing data, the researchers employed a rigorous nested cross-validation technique. This approach significantly reduced the risk of model overfitting and enhanced the credibility of the predictive results.

  The findings demonstrate that AI exhibits exceptional early-warning capabilities when it integrates an infant’s basic physiological information with routine laboratory data. Fluctuations in specific inflammatory markers and metabolic data served as the primary clues for detecting bacteremia.Additionally, changes in blood cell ratios provided critical supplementary information that allowed physicians to conduct a more comprehensive health assessment.

  The breakthrough of this technology lies in its independence from expensive high-end testing. By relying solely on routine reports already available to clinicians, AI can establish an early-warning radar. If implemented clinically, this system will serve as a digital assistant for precision decision-making. It will help medical teams take action before a bacterial infection escalates to safeguard the lives of newborns.

Figure 2. Watercolor portrait of Assistant Professor Cheng-Wei Cheng’s research team

Yao-Jen Dong and Cheng-Wei Cheng, Prediction of Bacteremia in Infants using Machine Learning based on Demographic Data, Complete Blood Cell Count with Differential, Glucose and C-Reactive Protein, Manuscript in Preparation (2025).