Highly Flexible Deep-Learning-Based Automatic Analysis for Graphically Encoded Hydrogel Microparticles.

Area of Science:

• Biomedical Engineering
• Nanotechnology
• Artificial Intelligence

Background:

• Hydrogel microparticle (HMP)-based bioassays offer high multiplex detectability, sensitivity, and specificity.
• Deep learning enhances HMP analysis, but manual annotation and plain particle data limit accuracy with functional nanomaterials.
• Existing manual data annotation is labor-intensive and time-consuming.

Purpose of the Study:

• To develop an efficient deep learning-based analysis for encoded HMPs with diverse graphical codes and functional nanomaterials.
• To overcome limitations of manual data annotation and plain particle data in HMP analysis.
• To improve the speed and accuracy of HMP bioassay data processing.

Main Methods:

• Utilized auto-annotation for rapid dataset preparation, achieving 0.11 s/image throughput.
• Employed synthetic data mixing for model training, enhancing analysis of HMPs with magnetic nanoparticles.
• Developed a deep learning model for analyzing diverse graphical codes and functional nanomaterials.

Main Results:

• Achieved a mean average precision of 0.88 using synthetic data mixing, a twofold improvement over standard methods.
• Demonstrated practical applicability with a triplex immunoassay for preeclampsia biomarkers.
• Reached a processing throughput of 0.353 s per sample for analyzing result images.

Conclusions:

• The proposed auto-annotation and synthetic data mixing strategy significantly enhances the efficiency and accuracy of deep learning-based HMP bioassay analysis.
• This automated approach addresses the limitations of manual annotation and plain particle data, enabling robust analysis of complex HMPs.
• The method shows strong potential for rapid and reliable diagnostic applications, as evidenced by the preeclampsia biomarker assay.