Label-free SARS-CoV-2 detection and classification using phase imaging with computational specificity.

Area of Science:

• Biophysics
• Medical Diagnostics
• Machine Learning

Background:

• Fast, accurate, and scalable viral detection is critical for managing pandemics.
• Current methods can be time-consuming and require sample preparation.

Purpose of the Study:

• To develop an optical method for direct, label-free imaging and deep learning-based classification of viral particles.
• To assess the accuracy and speed of this novel diagnostic approach.

Main Methods:

• Utilized an ultrasensitive interferometric method for nanoscale optical path-length imaging of viruses.
• Trained U-Net convolutional neural network models using paired interferometric and fluorescence images.
• Applied the trained network to classify four virus types (SARS-CoV-2, H1N1, HAdV, ZIKV) from interferometric images alone.

Main Results:

• Achieved 96% accuracy in identifying SARS-CoV-2 from a mixture of viruses using only interferometric data.
• Demonstrated rapid inference time of 60 milliseconds per image on a standard GPU.
• The method successfully imaged unlabeled viral particles with nanoscale sensitivity.

Conclusions:

• This direct imaging and deep learning approach offers a potentially extremely fast (<1 minute/patient) viral detection test.
• The technology's compatibility with standard glass slides suggests potential application with patient breath condensates.
• High-throughput testing is feasible by adapting digital pathology techniques for automated slide scanning.