Since the quality of underwater images is often compromised by numerous environmental factors, including light absorption, scattering, and the presence of suspended particles, underwater image enhancement has been a key area of investigation in recent years. AI and deep learning have revolutionized underwater image enhancement, surpassing traditional image processing methods. The detailed examination of the most recent deep learning-based underwater image enhancement approaches categorizes the methods into two main types: Convolutional Neural Networks and generative adversarial networks. The latter GAN is further separated into Conditional GAN and Cycle GAN. The work summarizes deep learning approaches in underwater image enhancements, including the methods, datasets, and quantitative determination metrics. The analysis shows an outline of the contributions and quantitative results of existent methods to make possible the study of underwater image enhancement. Moreover, the present study furnishes a detailed analysis of quantitative evaluations of the typical techniques of deep learning approaches.