Application of Denoising Diffusion Probabilistic Methods in Fetal MRI

Magnetic resonance imaging (MRI) is a common type of medical image acquisition that can also be used to diagnose early diseases. In this sense, fetal MRI is a non-invasive method to generate high-quality fetal volumes and to perform important clinical analysis. However, image denoising is necessary in many situations to ensure accurate evaluations. Thus, approaches such as Denoising Diffusion Probabilistic Methods (DDPM) have emerged and reached great results in this kind of task. In this work, we applied two DDPM-based approaches (an original and an improved one) besides two self-supervised deep models in a fetal MRI dataset. The results showed that, for the used fetal MRI dataset, the improved DDPM, named I-DDPM outperformed the counterparts considering two evaluation metrics for image quality, the Peak Signal-to-Noise Ratio (PSNR) and Root Mean Squared Error (RMSE).