Abstract:
Breast cancer is one of the most common and deadly cancers in women. Furthermore,
dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) plays a key role in its diagnosis. However, the images produced by this latter medical imaging modality are often affected
by Rician noise, which badly affects the performance of segmentation models. This manuscript
reviews and discusses recent deep learning methods for DCE-MRI breast tumor segmentation.
Then, it introduces RicIU-Net, a U-Net variant that injects Rician noise into encoder layers
during training to improve robustness. Tested on the public BreastDM dataset, RicIU-Net
outperforms U-Net and U-Net 2.1D in terms of ice sore and yields a satisfactory IoU score,
showing better adaptation to real-world imaging conditions. Hence, the proposed approach offers
a simple yet effective way to enhance the reliability of the segmentation without external denoising
