Generating Realistic Cardiac MR Images Using Diffusion Models

The number of studies in the medical field that uses machine learning and deep learning techniques has been increasing in the last years. To harness the full potential of deep learning for medical imaging, large datasets are required for training. These datasets are difficult to obtain due to privacy concerns, underrepresentation of rare diseases, poor standardization, and lack of diagnostic label quality and availability of experts for annotation. To solve this problem, we aim to explore the models offered by MONAI Generative Models, focusing on the Diffusion Models, to generate realistic synthetic cardiac magnetic resonance (MR) images. The model used has been able to generate very realistic cardiac MR images in a fast and easy-to-implement way. Given that the obtained synthetic images closely resemble the real one to the extent of being difficult to distinguish them, the study has shown its potential utility in augmenting datasets for medical imaging purposes.Clinical Relevance- The current work presents a framework to generate large amounts of synthetic cardiac MR images. Furthermore, the synthetic images obtained by the trained model are extremely realistic, maintaining the same details and characteristics of real cardiac MR images.