Contact Us
(910) 584-6193

While the world is in the midst of the largest mass vaccination undertaken by mankind, new strains of SARS-CoV-2, which are 8-10 times more infectious, have been detected in the UK, South Africa, Nigeria, Brazil, and Japan, and have been spreading rampantly across the globe. Given the higher infectivity and transmission of the new strains and the efficacy of the vaccines against the new strains unknown, effective tools that can be employed for mass screening are the need of the hour.

To fulfill this need, RetinaNet is using AI based tools to evaluate ophthalmologic (Retinal, Scleral and Iris) images to create a real-time test for initial screening of suspects. This test procedure could provide results in less than 3 minutes from acquiring the image sample to reporting of results.

Test background​ : ​While literature on incidence and prevalence of ophthalmic manifestations of COVID-19 remains limited, a broad spectrum of ophthalmic manifestations, including ​conjunctivitis, anterior uveitis, retinitis, and optic neuritis have so far been described in COVID-19[2, 3].​ Two separate research teams in Asia reported reproducible ophthalmologic changes in clinical and laboratory confirmed COVID-19 pneumonia. Although there are conflicting reports about the incidence of conjunctivitis in COVID, ranging from 0.8-31%, conjunctivitis is the most common ocular manifestation of COVID-19 [4].

Fortem Genus Labs (FG Labs) in collaboration with FSU used artificial intelligence and machine learning in image processing to train Convolutional Neural Network (CNN) to evaluate scleral images and provide real -time diagnosis of COVID-19.

RN was initially trained to recognize signs in chest X-ray images and distinguish COVID pneumonia from other types of pneumonia or no
pneumonia. RN was also trained to identify diabetic retinopathy and differentiate all 5 stages using fundoscopic retinal images. This experimental stage established that RN could achieve differential diagnoses with AI training and image evaluation.

Current Findings : ​Based on the scleral images alone, RetinalNet CNN was able to identify COVID-19 patients with an accuracy of 82%.
The results obtained are shown in Table 1 and Table 2.

Discussion​ : ​Though the sample size of the test group was small, these results are very encouraging. Early PCR tests had Positive and Negative Predictive values (PPV & NPV respectively) in the 60% range initially. They have subsequently made incremental improvements and have PPV and NPV in the 90% range. A larger test sample would make a very marked difference. For the True Positive group N=7 each subject accounts for more that 10% of the PPV/NPV. There are also a number of test sampling and test subject characteristics that will be examined during further development.

The accuracy of 82% is even more encouraging since RN provides a cost-effective, real-time screening of subjects, and can provide results in less than 3 minutes, from acquisition of image, analysis through CNN to final report. On the other hand, the gold standard for COVID-19 detection-PCR currently takes 24-72 hours for diagnosis, and is limited by the number of kits available and also its cost. The rapid reporting, cost-effectiveness and absence of invasive interventions(such as nasal or pharyngeal swabs) make this test an ideal tool for contact tracing, workforce monitoring and rapid screening.

Scope and Limitations : ​The CNN was trained mostly on “augmented” images,

since original, usable eye scans were limited. Secondly, since the number of retinal and iris images in the database were limited, the CNN was ‘trained’ only on scleral images. Usage of retinal fundus images and iris scans, to supplement the scleral images, should further increase the diagnostic yield.

The preliminary findings of the study provide evidence that the product of this research, namely the trained CNN has high accuracy, precision and sensitivity on the training dataset and is showing increasing performance on the same metrics, as more images are processed. These quantitative results are very encouraging, and they strongly suggest that further research is warranted to determine the feasibility of applying this algorithm in the clinical setting and to determine whether use of this AI based test could

lead to improved care and outcomes compared with current methods of Covid-19 assessment. While more needs to be done, the early success of this research promises a readily available, cost-effective and rapid, non-invasive screening test with a diagnostic yield comparable to PCR, which would be an indispensable tool in our armamentarium to fight the COVID-19 pandemic.


[1] ​WHO Coronavirus Disease (COVID-19) Dashboard
[2] Seah I, Agrawal R. Can the Coronavirus Disease 2019 (COVID-19) Affect the Eyes? A Review of Coronaviruses and Ocular Implications in Humans and Animals. Ocul Immunol Inflamm. 2020;28(3):391-395. doi:10.1080/09273948.2020.1738501
[3] Amesty MA, Alió del Barrio JL, Alió JL. COVID-19 Disease and Ophthalmology: An Update [published online ahead of print, 2020 May 22]. Ophthalmol Ther. 2020;1-12. doi:10.1007/s40123-020-00260-y
[4] Amesty MA, Alió del Barrio JL, Alió JL. COVID-19 Disease and Ophthalmology: An Update [published online ahead of print, 2020 May 22]. Ophthalmol Ther. 2020;1-12. doi:10.1007/s40123-020-00260-y

Pankaj K Jeswani MBBS, DNB Consultant Ophthalmologist, Daga Memorial Hospital, Nagpur, India.

iDetect Covid-19 rapid test
310 Dick St. Suite D
Fayetteville NC 28301

phone-square linkedin facebook pinterest youtube rss twitter instagram facebook-blank rss-blank linkedin-blank pinterest youtube twitter instagram