Abstract: To solve the problem of difficulty in segmentation of weak edge ultrasound images, an improved CNN (Convolutional Neural Networks) based weak edge ultrasound image segmentation method is proposed. The method first uses stationary wavelet transform to remove the noise in the image, and then uses weighted least square filter to enhance the image edge details. Then, an improved convolutional attention module is added to the residual network model to extract image features. Finally, the image segmentation accuracy is improved by optimizing the loss function. The experimental results show that the proposed method has good performance in processing weak edge details of ultrasound images and can improve the segmentation accuracy of medical ultrasound images. 

Key words: ultrasound image segmentation, image preprocessing, convolutional neural network ( CNN), stationary wavelet transform, weighted least squares filter