Machine learning-assisted colorimetric serum phosphate detection based on sweet potato-derived carbon dots

A dual-robust, portable colorimetric sensing platform was developed by integrating purple sweet potato-derived carbon dots (PF-CDs) with machine learning-assisted signal processing. Serving as a highly stable, green nano-reductant, the PF-CDs effectively circumvent the autoxidation issues of conventional reagents, efficiently triggering the molybdenum blue reaction to produce a reliable macroscopic colorimetric response. To decouple these signals from environmental and matrix noise, a smartphone-based imaging system coupled with an machine learning algorithm was deployed for precise color recognition and automated quantitative determination. This integrated platform enables rapid phosphate detection within 60 min, exhibiting a broad linear range of 0.1-5.0 mM, a low limit of detection (LOD) of 0.03 mM, and an exceptional prediction accuracy of 99%. Ultimately, by synergizing chemical stability with analytical precision, this strategy offers a highly practical, low-cost, and robust paradigm for POC clinical phosphorus monitoring.