<p>In this study, we address these challenges by introducing common multimodal image representation, a novel neural network-based model designed for multimodal image registration. Common multimodal image representation network utilizes a densely connected convolutional network architecture, inspired by the Tiramisu model, to transform diverse imaging modalities into a common latent space. This latent space, termed the common multimodal image representation, facilitates the integration and alignment of images by abstracting and preserving critical features across different modalities. We employed the publicly available multimodal biomedical dataset for evaluating registration methods, which includes magnetic resonance imaging, computed tomography, positron emission tomography, and single photon emission computed tomography images covering a variety of anatomical regions. The dataset provided a robust foundation for training and testing common multimodal image representation network, ensuring the model’s applicability across various clinical scenarios. Our experimental results demonstrate the superior performance of common multimodal image representation network in terms of accuracy and efficiency. These results outperform existing models such as modality fusion net, registration network, Fusion convolutional neural network, and MultiModNet, both in accuracy and computational efficiency. Common multimodal image representation network exhibited rapid computation times, emphasizing its suitability for clinical use.</p>