Cross-Section and Top-View Image-Analysis Pipelines for Automated Characterization of 3D-Printed Microfluidic Channels and Valves

Abstract

Printing a microfluidic device in one piece encloses its channels and valve membranes inside a solid body, where inspection from the outside cannot easily tell whether a channel formed or how far a membrane moves. In a companion study we built a design framework for printed membrane valves that rests on measurements taken from micrographs of purpose-built test devices. Here we report the imaging methods and the three open-source analysis programs that produce those measurements. Two of the programs analyze cross-section views, in which the microscope looks along each channel through its open end while light shines through the device from behind. In that view an open channel appears as a bright window, and a blocked channel or a deflected membrane darkens it. A general-purpose segmentation model (the Segment Anything Model) outlines candidate channel regions without task-specific training. A sequence of classical image tests then decides which candidate is the channel. The first cross-section program classifies each channel as open, occluded, or not formed and measures how far the channel roof sags. The second compares images of each valve before and after pressurization and fits a circular arc to the edge of the membrane to measure how far the membrane reaches into the channel. A set of quality checks turns an unreliable reach measurement into missing data rather than a false zero. The third program, still under development, analyzes top views taken through the transparent resin. The top-view program tracks both walls of every channel along its full length and classifies each channel as formed, narrowed, pinched closed, or not formed. Applied to 112 top views, it located narrowings and closures partway along channels, which a cross-section at the channel end cannot show, and it agreed with an expert on 17 of 24 reviewed channels. Each program also writes a diagnostic figure for every automated decision.