Related Experiment Video
Updated: Aug 6, 2025
05:34
Author Spotlight: Advancing Pathogen Diagnostics with Standardized LAMP
Published on: September 8, 2023
842
LightR-YOLOv5: A compact rotating detector for SARS-CoV-2 antigen-detection rapid diagnostic test results
Rongsheng Wang1, Yaofei Duan1, Menghan Hu2
1Faculty of Applied Sciences, Macao Polytechnic University, Rua de Luís Gonzaga Gomes, 999078, Macao Special Administrative Region of China.
Displays
|March 20, 2023
Summary
We developed LightR-YOLOv5, a lightweight detector for SARS-CoV-2 antigen tests. This tool efficiently verifies rapid diagnostic test (RDT) results, aiding community screening and reducing manual verification burdens.
Area of Science:
- Medical Diagnostics
- Computer Vision
- Artificial Intelligence
Background:
- Nucleic acid testing is the gold standard for SARS-CoV-2 detection, but antigen-detection rapid diagnostic tests (RDTs) are crucial for community screening.
- Manual verification of RDT results is time-consuming, and existing object detection algorithms are often too complex for widespread deployment.
- There is a need for efficient and deployable automated systems to verify RDT results.
Purpose of the Study:
- To develop a compact and efficient rotating object detection model for automated SARS-CoV-2 antigen RDT result verification.
- To address the limitations of high model complexity and computational cost in existing detection algorithms.
- To create a tool that can be easily deployed for community or regional screening management.
Main Methods:
- Proposed LightR-YOLOv5, a lightweight detector utilizing L-ShuffleNetV2 for feature extraction.
- Combined semantic and texture features using GSConv and depth-wise convolution, incorporating NAM attention for region localization.
- Introduced a novel data augmentation technique, Single-Copy-Paste, to enhance RDT result detection.
- Achieved a model size of only 2.03MB.
Main Results:
- LightR-YOLOv5 demonstrated superior performance compared to mainstream rotating object detection networks.
- Achieved 12.6%, 6.4%, and 7.3% higher mAP@.5:.95 metrics compared to RetianNet, FCOS, and R³Det, respectively.
- The model maintains a slight reduction in recognition accuracy while significantly reducing model complexity and size.
Conclusions:
- LightR-YOLOv5 offers an efficient and deployable solution for automated SARS-CoV-2 antigen RDT result verification.
- The developed model can significantly reduce the burden of manual verification in community and regional screening.
- This technology has the potential to improve the accessibility and efficiency of rapid diagnostic testing.
