The rapid growth of the Internet of Things (IoT) has transformed healthcare through medical IoT devices that enable real-time monitoring and efficient data exchange. However, the MQTT (Message Queuing Telemetry Transport) protocol, widely used in these networks, presents critical security vulnerabilities that expose systems to cyber threats. Traditional intrusion detection systems often struggle with imbalanced datasets and limited generalization across attack types. This study introduces an enhanced deep learning framework that integrates TabNet with Conditional Tabular GAN (CTGAN) to address these challenges. CTGAN plays a central role by generating high-quality synthetic samples to balance class distributions, thereby improving the learning of the model in minority classes. Trained on the CICIoMT2024 dataset, which reflects real-world medical IoT attack scenarios, the proposed model achieves 91. 89% accuracy in multiclass classification. The results highlight the effectiveness of synthetic data augmentation in boosting detection accuracy and robustness. This approach contributes to securing IoMT networks by enabling more resilient and adaptive intrusion detection.