Medical Ultrasound Image Improvement Based on Denoising Convolutional Autoencoder

The challenge of enhancing ultrasound images while retaining essential features is vital in the medical field. Speckle noise significantly degrades image quality and can obscure important details. In this paper, we propose a new enhancement approach that integrates a Non-Subsampled Shearlet Transform (NSST) with a denoising autoencoder. By leveraging the capabilities of NSST for multi-scale and directional decomposition, our method effectively separates fine anatomical features from speckle noise. Our approach results in substantial speckle reduction while preserving critical details, thus enhancing the visual quality of ultrasound images. Experimental evaluations reveal the effectiveness of the proposed method in noise suppression and feature preservation, measured through various performance metrics. This innovative strategy represents a significant advancement in ultrasound image processing, with potential benefits for clinical diagnostics.