Speckle Noise Removal in Ultrasound Images Using a Deep Convolutional Neural Network and a Specially Designed Loss Function

The removal of speckle noise in ultrasound images has been the focus of a number of researches. Meanwhile, deep convolutional neural networks (DCNN) has been proved effective for various computer vision tasks, including image classification, segmentation and denoising. In this paper, we apply deep convolutional neural network to remove speckle noise in ultrasound images. Besides, a new hybrid loss function is specially designed for speckle noise removal, which can result in faster and more stable convergence during training. Experiments on synthetic and real Ultrasound images show that the proposed model outperforms other speckle reduction methods.