Electrocardiogram (ECG) signal denoising plays a critical role in enhancing the accuracy and reliability of cardiac diagnostics and monitoring systems. However, conventional statistical denoising methods often fail to achieve efficient noise removal while preserving essential signal features, especially in real-time applications. This paper introduces ECGD-Net, a novel two-stage ECG denoising network that integrates Multivariate Intrinsic Empirical Mode Decomposition (MIEMD) for filtering and a Convolutional Neural Network (CNN) for deep denoising. In the preprocessing stage, the ECG signal is decomposed into multiple frequency bands using MIEMD, preserving critical components of the signal while isolating noise. The decomposed signal is then fed into the CNN, designed to capture both local and global signal features, ensuring effective noise reduction. The performance of ECGD-Net is evaluated using metrics such as Signal-to-Noise Ratio (SNR), Root Mean Square Error (RMSE), Structural Similarity Index (SSIM), and Quality Factor (QF). Results demonstrate that ECGD-Net significantly outperforms existing state-of-the-art methods, providing superior noise reduction and preserving vital ECG features necessary for accurate cardiac diagnostics. Future work involves optimizing the model for real-time implementation and expanding its application to wearable devices and telemedicine.