Related Experiment Video
Updated: Jun 8, 2026
Methylation Specific Multiplex Droplet PCR using Polymer Droplet Generator Device for Hematological Diagnostics
Published on: June 29, 2020
YOLO-Drop: A Deep Learning Model Enabling Accurate, High-Throughput Image Analysis for Droplet Digital Immunoassay at
Jianglian Xu1, Fuliang Huang1, Nanchi Jiang1
1Marshall Laboratory of Biomedical Engineering, Shenzhen Key Laboratory for Nano-Biosensing Technology, School of Biomedical Engineering, Shenzhen University Medical School, Shenzhen University, Shenzhen, Guangdong 518060, People's Republic of China.
We developed YOLO-Drop, a deep learning model for analyzing droplet digital enzyme-linked immunosorbent assay (ddELISA) images, significantly improving biomarker detection accuracy and speed for early disease diagnosis.
Area of Science:
- Biomedical Engineering
- Artificial Intelligence
- Biotechnology
Background:
- Ultrasensitive detection of low-abundance protein biomarkers is crucial for early disease diagnosis and monitoring treatment efficacy.
- Droplet digital enzyme-linked immunosorbent assay (ddELISA) offers attomolar sensitivity but is limited by conventional image analysis methods for high-throughput applications.
Purpose of the Study:
- To develop a custom deep learning model, YOLO-Drop, for accurate and high-throughput analysis of ddELISA droplet images.
- To deploy YOLO-Drop on an embedded platform with a graphical user interface for real-time, user-friendly operation.
- To enhance the speed, automation, and accuracy of ddELISA.
Main Methods:
- Developed YOLO-Drop based on the YOLOv8 architecture, incorporating deformable convolution, BiFormer modules, a high-resolution feature pyramid network, small-object prior, and class-aware nonmaximum suppression.
- Trained the model on a dataset of 5,574 droplet images containing approximately 750,000 droplets with diverse signal patterns.
- Deployed the model on an NVIDIA Jetson Orin Nano embedded platform with a graphical user interface.
Main Results:
- YOLO-Drop achieved 99.69% detection accuracy on heterogeneous ddELISA droplet images.
- The model demonstrated fast inference speeds on the Jetson platform, enabling real-time, on-device analysis.
- When applied to ddELISA, YOLO-Drop enabled the detection of interleukin-6 (IL-6) down to 9.57 aM in complex biological matrices.
Conclusions:
- YOLO-Drop significantly enhances the speed, automation, and accuracy of ddELISA for biomarker quantification.
- Deep-learning-assisted analysis holds transformative potential for next-generation biosensing platforms in clinical settings.
- The developed system supports accurate, automated, and high-throughput biomarker quantification for disease diagnosis and monitoring.
