Hybrid Rule-Based and Machine-Learning-Based Post-Processing of Dynamic Contact Angle Measurements

Abstract

Dynamic contact angle measurements provide direct insight into the wettability of biomaterial surfaces, crucial for protein adsorption and cell adhesion. However, in practice, automatic baseline detection, contour fitting, and contact point localization often fail, producing physically implausible frames. Manual correction is time-consuming, subjective, and limits the reproducibility of advancing and receding contact angle (ARCA) measurements. A hybrid postprocessing pipeline is presented, combining deterministic prefiltering of clearly invalid frames with machine-learningbased classification of the remaining data. Using an annotated dataset of 18,868 frames from various substrates, the optimized Naive Bayes classifier with ANOVA-selected features achieved 95.33% accuracy, 90.42% macro-precision, 95.95% macro-recall, and 92.77% macro-F1 score on the test set. The approach reliably identifies severely faulty (ultrabad) frames and significantly smooths ARCA curves, enhancing the objectivity, reproducibility, and efficiency of dynamic contact angle measurements.