Plasmonic immunosensor optical microscopy and machine vision to detect SARS-CoV-2 virus

1. Overview and Data Source

This dataset contains optical microscopy images of plasmonic biosensors based on gold nanoislands (AuNI) on glass substrates, used for detecting SARS-CoV-2 virus particles. It also includes extracted image features for machine learning-based detection. The dataset was generated for the study by Oiticica P. R. A. et al. (2025) (https://doi.org/10.1021/acssensors.4c03451). The Images were acquired using a Zeiss Axio Lab.A1 microscope with a 40x objective (400X magnification). Various handcrafted and deep-learning-based Convolutional Neural Networks (CNN) were used for feature extraction. Machine learning models, including LDA, KNN, SVM, and RF, were trained to classify images based on SARS-CoV-2 virus concentration. The highest classification accuracy (91.6%) was achieved using the MobileNetV3_small feature extractor combined with an SVM classifier, to detect the SARS-CoV-2 virus with concentrations as low as 1 PFU/mL. This approach has potential applications for detecting other viruses and analytes.
2. Data Structure and Description

This dataset contains 858 optical microscopy images in .TIF format (RGB, 1920×2560 resolution), captured under standardized conditions. The test categories are:

Positive tests: SARS-CoV-2 virus at dilutions from 1×10⁻⁴ to 1×10⁵ PFU/mL (10 classes).
Negative tests: RSV virus (1×10³ to 1×10⁵ PFU/mL), blank tests with PBS/MgCl₂, and probe images (immunosensors before testing).

File Naming Format: {sensor_number}_{label_name}.tif
Example: s012_cov_03.tif refers to an image of sensor s012 after testing with SARS-CoV-2 at dilution 03. The correspondence between dilution numbers and PFU/mL is in imageinfo.csv table.

The Features folder contains tables (.h5 format) with extracted image features for all images, with a table for each computer vision method.

Feature Table Columns:

filename: Image filename.
label_name: Test label combining analyte and dilution code.
concentration: PFU/mL concentration.
analyte: Analyte type ('CoV inat', 'RSV inat', 'PBSMgCl2', and 'No' for immunosensor before tests.)
vision_type: Feature extraction method (Handcrafted or CNN).
feature_vector_rgb: Feature vector extracted from RGB images.
feature_vector_gray: Feature vector extracted from grayscale images.

The table imageinfo.csv contains the first 4 columns of the feature tables.

3. Citation

If you use this dataset, please cite our paper:
Oiticica, Pedro R. A. and Angelim, Monara K. S. C. and Soares, Juliana C. and Soares, Andrey C. and Proença-Módena, José L. and Bruno, Odemir M. and Oliveira Jr, Osvaldo N. (2025). "Using Machine Learning and Optical Microscopy Image Analysis of Immunosensors Made on Plasmonic Substrates: Application to Detect the SARS-CoV‑2 Virus". ACS Sensors. DOI: 10.1021/acssensors.4c03451.

5. External Sources

GitHub: https://github.com/praoiticica/COVID-plasmonic-sensor-ML.