Reverberation artifacts in medical ultrasound B-mode images, caused by multiple repetitive reflections of the echo signal, significantly degrade image quality and hinder accurate diagnoses. Various studies, particularly those focusing on deep learning, have proposed techniques to mitigate these artifacts. Deep learning-based methods face a primary challenge in training strategy, which is categorized into supervised, semi-supervised, and unsupervised approaches. While supervised learning is simple and effective when input and ground-truth data are available, it is often impractical for artifact reduction due to the difficulty of obtaining suitable data. Conversely, unsupervised learning presents a promising alternative to overcome these data acquisition challenges. A recent advancement is deep coherence learning (DCL), an unsupervised technique specifically for enhancing ultrasound imaging quality. In this thesis, a custom phantom is designed to leverage DCL for suppressing reverberation artifacts. The effectiveness of the developed deep coherence learning with reverberation dataset (DCL-Reverb) was assessed using real-world experimental data. Quantitatively, DCL-Reverb demonstrated higher contrast-to-noise ratio (CNR) and generalized contrast-to-noise ratio (gCNR) compared to conventional methods. Qualitatively, it also achieved clearer B-mode images and superior artifact suppression in axial profiles.