Smart Point-of-care Optofluidic Imaging System

Abstract

Global healthcare is undergoing a transformation driven by both policy changes and technological advancements, with the intention of delivering quality healthcare to everyone in need. The technology should be accurate, accessible, and affordable to reduce disparities in healthcare. In this context, point-of-care (PoC) testing (PoCT) technology has been explored to address these requirements. Utilizing portable and cost-effective devices, PoCT is used as an alternative outside traditional centralized and advanced laboratory settings, which aim to provide reliable and rapid results to enable immediate clinical decision-making. Recently, integrating optical imaging techniques into PoCT has become a fast-emerging trend that offers robust visualization of samples and the capacity for measuring multiple signals or analytes. However, conventional optical imaging systems are bulky, complex, and costly optoelectronic instruments comprised of related components. In addition, accurate interpretation of acquired images can be challenging and may require specialized training, which prevents their use. Therefore, optical imaging integrated with PoCT, featuring the feasibility of automation and artificial intelligence, remains in strong demand for diagnostic tests.

In this research, we developed a lensless optofluidic imaging platform designed for integration into PoCT devices for healthcare applications. By eliminating conventional free-space optics and applying machine learning-based analytical techniques, this platform combines microfluidics for sample handling with lensless holographic or shadow imaging, as well as multidimensional analysis to enable automatic, label-free diagnostics. First, the portable imaging platform is based on the existing prototype and optimized for fewer calibration requirements. A physics-aware training strategy keeps physics consistency, which is desirable for interpretation. This strategy bridges traditional physics-based approaches and recently developed learning-based approaches. Then, flowing cells in the cartridge are embedded into a frame-level feature that contains morphological information, such as motion patterns. This can be used for identification and motion segmentation in single-cell/particle tracking time series. Third, a data annotating and training pipeline is proposed, which yields high performance and efficiently decreases the annotation burden for implementing learning strategies in image-based cell profiling. This strategy can be extended to a broader spectrum of cell profilers for more biomedical applications, e.g., rare cell detection, label-free species identification, and cell-cycle phase analysis.

The results of our studies demonstrate that developed solutions enhance accuracy, sensitivity, and overall efficiency. This work helps address PoCT challenges, including miniaturization, affordability, and ease of use, while maintaining diagnostic accuracy comparable to that in conventional laboratory settings. It provides a clear potential for future improvements and research directions.