Artificial intelligence and microscopy: A systematic review of deep learning in microalgae and cyanobacteria species identification

This is the authors' abstract. We don't add key points for this paper.

Algal blooms are increasingly frequent and severe, threatening water quality, ecosystems, and public health. Accurate species-level identification is crucial for effective monitoring, yet conventional microscopy is slow, subjective and labour-intensive. Integration of deep learning with microscopy has recently emerged as a transformative solution. This systematic review synthesizes 62 studies that applied deep learning to microscopic images of microalgae and cyanobacteria, offering a comprehensive perspective that spans the entire workflow, from image acquisition and pre-processing to model architectures. A central contribution of this review is the structured synthesis of literature. Studies employing classification models are distinguished from those using detection models, and within each category, subgroups are identified and critically evaluated to clarify their strengths, limitations, and appropriate applications. The analysis shows that while classification models account for most studies overall, their dominance reflects earlier stages of the field. Recently, detection models have become more common and are emerging as preferred choice for addressing the complexity of real-world, multi-species images. Beyond synthesizing existing methods, this review identifies gaps in imaging techniques, dataset availability, and model development. It highlights ongoing challenges, including scarce annotated data and underexplored imaging and modelling approaches, while outlining promising future pathways such as semi-supervised learning, multimodal integration, and advanced detection and segmentation architectures. Finally, publicly available microalgae image datasets are introduced, whose integration with existing methodologies may enable more robust and universal identification models. By combining full workflow coverage, this study clarifies the state of the field and guides subsequent research in deep learning for algal monitoring.