Image based chemical identification of ophthalmic antibiotics

Substandard and falsified medicines remain a significant global health challenge. While conventional analytical quality control methods such as high-performance liquid chromatography are accurate and reliable, they are often costly, infrastructure-dependent, and unsuitable for resource-limited settings. In this study, we present a low-cost analytical framework based on evaporation-induced deposit imaging combined with machine learning for the identification and quantification of five commercial ophthalmic formulations. Droplets of each formulation were deposited on glass slides, dried under controlled environmental conditions, and imaged. A total of 46 quantitative image features were extracted from each dried deposit, capturing texture, morphology, and intensity variations arising from formulation-specific physicochemical behavior. The evaporation process produced distinct crystallization patterns, including coffee-ring structures, dendritic growth, and spherulitic morphologies, allowing the formulations to be clearly distinguished both visually and statistically. Principal component analysis separated all five drugs into distinct clusters, with the first three principal components explaining 65.6% of total variance. For quantitative analysis, we applied XGBoost and EfficientNet-B3 models to standard dilution images of ciprofloxacin and moxifloxacin. Within the tested concentration ranges, both models performed well, yielding regression slopes close to unity (0.89-0.95) and R² values as high as 0.98. Evaluation of models using R², MAE, and MSE metrics confirmed the models' accuracy. Validation on real samples showed that the models generalized effectively, although prediction error increased due to more complex samples. These results offer a proof-of-concept for using macroscopic images to extract chemical information.