Cultural heritage embodies profound historical, artistic and scientific significance. Protecting and inheriting cultural heritage as well as uncovering their inherent values, presents a great challenge faced by contemporary cultural relic conservation practitioners. However, owing to the complexity, diversity, and value of cultural heritage, applications such as disease identification, quantitative analysis, deterioration prediction, and virtual restoration often suffer from low efficiency and insufficient levels of automation and intelligence. In addition, conventional data processing techniques struggle to address the nonlinear and heterogeneous characteristics of cultural heritage data, thereby limiting their effectiveness in complex application scenarios. Deep learning, as an emerging technological approach, provides new solutions to these challenges. This paper reviews the application of deep learning in cultural heritage disease research from 2017 to 2024. Using bibliometric methods, it outlines developmental stages and trends and identifies key research hotspots. The results indicate that the period from 2020 to 2024 witnessed rapid growth, with a significant increase in the number of publications on related topics. In terms of geographical contributions, China and Italy have demonstrated particularly strong academic influence and contribution. At the data source level, digital images serve as the core foundation, while spectral analysis and radiographic imaging data, which contain rich material and structural information, have also become research priorities. In terms of model applications, Convolutional Neural Networks (CNNs) and Generative Adversarial Networks (GANs) are the most commonly employed models. Regarding application targets, deep learning technologies have been applied to various types of cultural heritage, including buildings, calligraphy and paintings, murals, earthen painted artifacts, and stone relics. Based on these findings, the main challenges and future development directions in this research field are discussed from four perspectives:1) data—improving the quality and quantity of samples; 2) models—enhancing performance, generalization ability, and interpretability; 3) applications—extending technologies to areas such as disease mechanism analysis and dynamic deterioration prediction; and 4) standards—establishing unified industry standards, data-sharing platforms, and adaptive evaluation metrics. This study provides a theoretical foundation and practical references for future research and contributes to advancing the intelligent development of cultural relic conservation.