The advancement of deep learning technology has significantly contributed to the agricultural sector, particularly in image-based rice leaf disease identification. This study aims to analyze the development trends, model performance, and challenges in applying deep learning methods through a Systematic Literature Review (SLR) approach. The review process follows the PRISMA guidelines, analyzing 10 scientific articles published between 2020 and 2026. The results indicate that Convolutional Neural Networks (CNNs) remain the most widely used approach, with various architectures such as ResNet, VGG, and MobileNet demonstrating high classification performance. Furthermore, recent studies show a shift toward more adaptive methods, including object detection models such as YOLO and hybrid approaches that integrate CNNs with Vision Transformers. Most studies report accuracy levels exceeding 90%, highlighting the strong potential of deep learning for automated plant disease detection.However, further analysis reveals a significant gap between model performance on controlled datasets and real-world conditions. The main challenges identified include limited availability of representative datasets, low model generalization capability, high computational complexity, and lack of model interpretability. In addition, practical implementation in real agricultural environments remains limited, indicating the need for more application-oriented research. This study contributes by providing a comprehensive mapping of current research trends and identifying key research gaps, which can serve as a foundation for developing more robust, efficient, and practical rice disease detection systems in the future.