This article discusses the integration of medical domain knowledge into machine learning (ML) models for medical decision-making. It emphasizes the importance of incorporating medical knowledge into ML pipelines to enhance model accuracy, interpretability, and alignment with clinical guidelines. The study reviews existing efforts to integrate medical knowledge into ML and maps these strategies onto the phases of the ML pipeline, including data preprocessing, feature engineering, model training, and output evaluation. A case study on diabetes prediction is used to illustrate the benefits of integrating clinical knowledge, such as rules, causal networks, intervals, and formulas, at each stage of the ML pipeline. The results show that integrated models outperform purely data-driven approaches in terms of accuracy, interpretability, and adherence to clinical guidelines. The study also identifies two main challenges to integration: refining the representation of medical domain knowledge and fine-tuning its contribution to the ML model. The article highlights the potential of integrating medical knowledge into ML models to improve their performance in clinical settings, particularly in scenarios with limited data. It also discusses the limitations of ML in medicine, including data efficiency, accuracy, interpretability, and coherence, and proposes integration strategies such as ensemble learning, transfer learning, and informed machine learning to address these challenges. The study concludes that integrating medical knowledge into ML models is essential for building predictive models that are particularly desirable in the healthcare sector, thereby facilitating their adoption in clinical practice.This article discusses the integration of medical domain knowledge into machine learning (ML) models for medical decision-making. It emphasizes the importance of incorporating medical knowledge into ML pipelines to enhance model accuracy, interpretability, and alignment with clinical guidelines. The study reviews existing efforts to integrate medical knowledge into ML and maps these strategies onto the phases of the ML pipeline, including data preprocessing, feature engineering, model training, and output evaluation. A case study on diabetes prediction is used to illustrate the benefits of integrating clinical knowledge, such as rules, causal networks, intervals, and formulas, at each stage of the ML pipeline. The results show that integrated models outperform purely data-driven approaches in terms of accuracy, interpretability, and adherence to clinical guidelines. The study also identifies two main challenges to integration: refining the representation of medical domain knowledge and fine-tuning its contribution to the ML model. The article highlights the potential of integrating medical knowledge into ML models to improve their performance in clinical settings, particularly in scenarios with limited data. It also discusses the limitations of ML in medicine, including data efficiency, accuracy, interpretability, and coherence, and proposes integration strategies such as ensemble learning, transfer learning, and informed machine learning to address these challenges. The study concludes that integrating medical knowledge into ML models is essential for building predictive models that are particularly desirable in the healthcare sector, thereby facilitating their adoption in clinical practice.