Conditional superiorities and unaddressed bottlenecks: a critical review of artificial intelligence for waterborne microbial detection

Although conventional microbial detection approaches for water samples are widely applied, they still suffer from prolonged assay durations (24-72 h), low sensitivity, and the absence of real-time monitoring capacity. Artificial intelligence (AI) has demonstrated conditional advantages in specific experimental environments, such as achieving a sensitivity of 99% for detecting Cryptosporidium and Giardia in low turbidity water (based on approximately 12,000 annotated images, using fivefold cross validation, completed under laboratory conditions); however, such advantages tend to diminish or vanish in high-turbidity water matrices or when training datasets are insufficient. This review critically evaluates four categories of AI-driven approaches: image-based analysis, spectroscopic techniques, genome, and metagenomic sequencing, as well as predictive pollution modeling. While AI helps boost detection efficiency, precision, and analytical capacity, a set of long-standing obstacles restrict its real-world deployment. The main issues involve non-standardized datasets, low model interpretability, weak generalization over various water substrates, and a substantial gap between lab-based performance and on-site operational outcomes. In summary, to fully exploit the capabilities of AI in aquatic microbial detection, greater emphasis should be placed on on-site validation, unified data specifications, and practical performance benchmarks, rather than further algorithmic innovation. This review seeks to provide practical references for scholars and practitioners working in the fields of microbiology, AI and water quality monitoring and management. KEY POINTS: • AI shows favorable performance for microbial detection under lab conditions. • Model performance declines greatly in complex water with many practical barriers. • Standardized data and validation will advance real-world application.