Related Experiment Video
Updated: May 6, 2026
09:12
High-Throughput Automated Multiplex Immunofluorescence Assays for Translational Research
Published on: June 10, 2025
1.4K
Homogeneous image-based digital immunoassays with high error tolerance
Darren B McAffee1, Qiang Hu2, Assame Arnob2
1ilytica, LLC., San Francisco, CA, USA. darren@ilytica.com.
Npj Imaging
|May 4, 2026
Summary
Machine learning enhances image-based nanoparticle immunoassays for field diagnostics. This approach improves accuracy and sensitivity for detecting biomarkers like C-reactive protein and SARS-CoV-2 antibodies.
Area of Science:
- Biomedical Engineering
- Nanotechnology
- Machine Learning
Background:
- Global health demands field-deployable in vitro diagnostics, requiring smaller sample volumes and tolerance for handling variability.
- Current field diagnostics face limitations in sample processing and measurement capabilities compared to lab-based tests.
- Enhancing error tolerance is crucial for successful design of assays for point-of-care and self-administered use.
Purpose of the Study:
- To investigate machine learning (ML) strategies for improving error tolerance in image-based nanoparticle immunoassays.
- To compare conventional image analysis with ML-enhanced approaches for assay performance.
- To assess the feasibility of using ML for quantitative analyte detection in resource-limited settings.
Main Methods:
- Image-based nanoparticle immunoassays were developed using microliter sample volumes.
- Analyte concentrations were determined by analyzing nanoparticle appearance in images.
- Three analysis methods were compared: conventional image analysis, hybrid ML-conventional segmentation, and end-to-end ML image regression.
Main Results:
- The segmentation-based ML approach achieved 96% specificity and 90% sensitivity for binary classification of SARS-CoV-2 IgG.
- The end-to-end regression ML model provided quantitative performance with a limit of detection of 5.2 ng/mL, comparable to ELISA.
- ML approaches significantly improved dynamic range, sensitivity, and reproducibility compared to conventional methods, with reduced labeling effort.
Conclusions:
- Machine learning significantly enhances the error tolerance and performance of image-based nanoparticle immunoassays for field applications.
- End-to-end ML regression offers a powerful, label-efficient method for quantitative detection, approaching ELISA sensitivity.
- These ML-driven advancements are vital for translating complex diagnostics to accessible point-of-care and self-administered formats.
