AI-Based Detection of Droplets and Bubbles in Digital Microfluidic Biochips

Abstract

Digital microfluidic biochips exploit the electrowet-ting on dielectric effect to move and manipulate microliter-sized liquid droplets on a planar surface. This technology has the potential to automate and miniaturize biochemical processes, but reliability is often an issue. The droplets may get temporarily stuck or gas bubbles may impede their movement leading to a disruption of the process being executed. However, if the position and size of the droplets and bubbles are known at run-time, these undesired effects can be easily mitigated by the biochip control system. This paper presents an AI-based computer vision solution for real-time detection of droplets and bubbles in DMF biochips and its implementation that supports cloud-based deployment. The detection is based on the YOLOv5 framework in combination with custom pre and post-processing techniques. The YOLOv5 neural network is trained using our own data set consisting of 5115 images. The solution is able to detect droplets and bubbles with real-time speed and high accuracy and to differentiate between them even in the extreme case where bubbles coexist with transparent droplets.