Time-Lapse Quantitative Analysis of Drying Patterns and Machine Learning for Classifying Abnormalities in Sessile Blood Droplets

Abstract

When a colloidal droplet dries on a substrate, a unique pattern results from multi-facet phenomena such as Marangoni convection, capillary flow, mass transport, mechanical stress, colloid-colloid, and colloid-substrate interactions. Even under uniform conditions (surface wettability, humidity, and temperature), slight differences in the initial colloidal composition alter the drying pattern. This paper shows how the evolving patterns during drying in the sessile droplets depend on the initial composition and are crucial for assessing any abnormalities in the blood. To do so, texture statistics are derived from time-lapse images acquired during drying, and different traditional machine learning are applied. In addition, a neural network analysis is performed on both images and their texture statistics. As the drying phenomena are correlated with the varying composition, these methods exhibit excellent performance in distinguishing blood abnormalities with an F1 score of over 97%. This indicates that analysis of time-lapse images during drying and their texture statistics, rather than conventional analysis using images at the final dry state, are crucial for classification. Our results highlight the potential of droplet drying as a low-volume, accurate, and simple screening tool for detecting the type and stage of any disease in bio-fluid samples, such as blood, urine, and saliva.