The increasing deployment of Internet of Medical Things (IoMT) devices in healthcare environments has made these systems particularly vulnerable to malicious cyber activities, threatening both patient safety and the integrity of medical data. One of the major challenges in IoMT attack detection using machine learning and deep learning models lies in the inherent imbalance present in IoMT traffic datasets. Additionally, the dynamic nature of IoMT environments, with devices producing diverse and non-stationary data, complicates the task of developing reliable detection models. This paper proposes an advanced approach to address these challenges by enhancing the security of IoMT networks through synthetic data augmentation and sophisticated deep learning models, Conditional Tabular Generative Adversarial Network (CTGAN) is utilised to balance the IoMT malicious traffic dataset by generating synthetic samples, ensuring a more representative dataset for detection models. Additionally, a Bidirectional Gated Recurrent Unit (Bi-GRU) with an attention mechanism is implemented to accurately classify malicious traffic. We performed binary and multi-class classification on the WUSTL-EHMS-2020 dataset, which comprises internet of medical traffic data. Results demonstrated that the bidirectional GRU achieved 98% accuracy, 96.6% F1-score, and 96.48% geometric mean, respectively.