This research introduces a hybrid Intrusion Detection System (IDS) tailored for the Internet of Medical Things (IoMT), addressing the critical cybersecurity vulnerabilities inherent in connected healthcare systems. The proposed model integrates both machine learning (Random Forest and XG Boost) and deep learning (Tab Net and CNN-Bi LSTM) techniques to achieve high detection accuracy, interpretability, and generalizability, particularly for multi-class and imbalanced datasets. The system uses a modular pipeline for preprocessing and hyperparameter tuning, with stratified sampling for balanced dataset splitting. Initial experiments show strong performance from traditional models (up to 98.8% accuracy), but a hybrid Tab Net -CNN- Bi LSTM model achieves superior results, reaching 99.0% testing accuracy. Explainable AI tools, SHAP and LIME, are incorporated to ensure transparency and accountability in decision-making. Compared to existing methods, the proposed model surpasses many in accuracy, scalability, and interpretability, offering a trustworthy solution suitable for real-time deployment in smart hospitals.