A Comprehensive Review of Informed Machine Learning in Medical Decision Systems

Abstract

Machine learning (ML) has become a transformative technology in healthcare, facilitating improved diagnosis, prognosis, and treatment personalization. However, conventional ML models often struggle with limited data, lack of interpretability, and insufficient robustness, which constrain their clinical applicability. Integrating prior medical domain knowledge into ML – known as medical-informed machine learning (IML) – addresses these challenges by embedding expert insights and clinical guidelines into the learning process. This paper presents a comprehensive review of medical informed machine learning approaches, analysing studies that incorporate domain knowledge at various stages of the ML pipeline, including data pre-processing, feature engineering, and model learning. We discuss different types of domain knowledge such as clinical rules, causal networks, and physiological formulas, and illustrate their integration into ML models. IML approaches improve data efficiency, predictive accuracy, and model interpretability compared to purely data-driven methods. They also enhance model robustness and coherence with clinical practice, particularly in scenarios with limited or noisy data. Case examples from diagnostic imaging and chronic disease prediction underscore these benefits. Medical – informed ML is a critical advancement for reliable and trustworthy clinical decision support systems. Future research should focus on standardizing domain knowledge representation and developing scalable integration techniques to maximize clinical impact.

Ähnliche Arbeiten