Machine learning-driven colorimetric sensing and monitoring of optically-engineered liquid marbles

Liquid marbles represent a breakthrough class of autonomous microfluidic platforms featuring dynamic self-healing capabilities. Particularly, optically-engineered liquid marbles (OLMs) uniquely integrate microreactor functionality with inherent colorimetric sensing properties, enabling precise fluid manipulation coupled with full optical accessibility. While digital image colorimetry (DIC) can be regarded as an optical sensing modality that offers rapid, instrument-free analysis, its accuracy remains fundamentally constrained in complex multivariate systems due to ambient interference and interfacial artifacts. To address these limitations, we develop a machine learning-enhanced DIC platform leveraging OLMs as optimized sensing cells. Our integrated approach combines convenient OLM fabrication with stable optical properties, multilayer perceptron (MLP)-driven RGB-to-concentration transformation that eliminates physical modeling requirements, and adaptive calibration protocols that robustly compensate for lighting variations. The MLP-DIC sensor system demonstrates significant advantages in precision and versatility, reliably quantifying chemical diffusion profiles across nonplanar interfaces while maintaining accuracy under challenging conditions. Validation through real-time monitoring of coalescence dynamics and evaporative processes confirms exceptional operational stability in biosimulative environments. This work establishes a new paradigm for intelligent micro-reactor or sensor integration, providing high-fidelity, non-invasive chemical tracking with transformative potential for point-of-care diagnostics and combinatorial chemistry platforms.