Antibody tests are the key -- but only with intelligent guidance.

Rapid antibody tests have the potential to empower frontline healthcare workers. With a simple finger-prick, they have the power to identify immunity in just 15 minutes. And combined with wide-scale deployment, they have the ability to deliver crucial insights to hospitals and governments in the battle against the pandemic.

But without the proper tools, this potential will remain untapped. We must overcome three key barriers to harness the power of immunity testing:

  • Uncertainty - Hospitals must be strategic in their test deployment. Timing is critical -- if the test is performed too soon after exposure to the virus, the results can be completely inaccurate. And even with perfect timing, lateral flow immunoassays have a PPV of as low as 55%— a dangerous shortcoming if not augmented with data on individual exposure risk.

  • Disorganisation - Health care workers will need to be retested at regular intervals. This is essential to combat the scientific uncertainty surrounding sustained immunity and the variable accuracy of current tests. An easy-to-implement, easy-to-use, and interoperable platform is vital for tracking this unprecedented volume of immunity tests.

  • Privacy Concerns - Healthcare workers need to feel confident that their data is secure. Immunity results must not only be FDA and HIPAA-compliant, but stored by an entity that users can trust won’t use it for the wrong reasons — now and forever.

Introducing ImmunoLynk

  • Effortless + Trustless = Fearless. Healthcare workers use our mobile app to take a photo of their immunity test results. Our image classification algorithm, built on Keras over a ResNet50 model, automatically reads the test result as accurately as a physician, removing any possible user interference. The corresponding photo, test result and metadata are stored securely on a corresponding IPFS Blockchain node, guaranteeing the test validity and immutability.

  • The virus can adapt—so can we. Healthcare workers fill out a brief, but thorough, questionnaire about their potential exposure risk factors. Combined with daily symptom surveys and hyper-local prevalence data, our proprietary machine-learning algorithm determines the ideal time to administer an antibody test. The result is fed into a Bayesian regression model, resulting in a single, easy-to-understand “True Immunity” score. This multimodal data integration overcomes the inherent sensitivity and specificity limitations of immunoassays, creating a tailored diagnostic test capable of accurately conveying the amount of uncertainty. Best of all, all logs are immutable, distributed, and instantaneous — so healthcare workers can worry less about the privacy of our data, and more about conquering the crisis.

  • We scale with the pandemic. As our network of healthcare workers and the number of completed antibody tests grow, so does the strength of our algorithms. Our blockchain operating expenses cap at the ultra-low-cost of $25/month per healthcare facility, regardless of the number of employees or tests. IPFS nodes can be deployed and run completely on cloud elastic containers, presenting no additional data transfer, data storage, or uptime cost fees. Every machine learning evaluation is processed through an optimized gateway server so requests are readily processed and delivered to the Blockchain.

  • Connecting the world so we can reconnect. Widespread adoption of the ImmunoLynk platform can effectively construct a distributed worldwide research network, creating the largest-ever study on the time-course of COVID-19 and its respective antibodies. This could provide the key data necessary to intelligently lift quarantine restrictions alongside community immunity testing, as well as prepare for future pandemics.

  • Caring for healthcare. Priority allocation of immunity tests to healthcare workers aligns with European Centre for Disease Prevention and Control (ECDC) and Centers for Disease Control (CDC) recommendations, and with ImmunoLynk's help, it can put health providers' minds at ease. They can care for patients, perform procedures, and return home to their loved ones with less worry about contracting or communicating the virus. Our U.S. Equal Employment Opportunity Commission (EEOC) and American Disabilities Act (ADA)-compliant dashboard software for administrators also permits easy-to-implement tracking of employee symptoms and immunity tests, allowing them to strategically tackle staffing shortages in the wake of looming surges.

What we have done over the past few weeks

By integrating two initially discrete teams based around the globe and collaborating across diverse areas of expertise, we developed a feasible solution for the problems presented above in the context of a realistic business plan. These teams initially connected as participants in the MIT COVID-19 Hackathon, where one was announced a winner of the challenge. This joint team was subsequently named the first prize winner in the Lumiata COVID-19 Global AI Hackathon.

So far, we have created a working prototype of (1) a mobile app dashboard connected to a gateway server that (2) leverages machine learning to (3) combine questionnaire answers, location data and antibody test results to provide users with a "True Immunity" score registered to IPFS Blockchain.

  • (1) - The mobile app uses Expo.io API with React components to interact with users. It consists of a login and registration page, a symptom and risk questionnaire, and finally a test scanning section. This interface is currently functional. It takes a photo of the test placed on a QR barcode, sends it to a Gateway server, and displays the processed test result with an access link to the blockchain data.

  • (2) - The gateway server was built using the lightweight Flask framework with an exposed REST API, which receives both images and data from two distinct endpoints. Images are then submitted through the Keras model and, ultimately, uploaded along with the result and metadata to the IPFS Blockchain node through Infura's API.

  • (3) - Our image processing pipeline was built with Tensorflow, Keras and OpenCV2 on the ResNet-50 architectural network. It was extensively optimized to achieve an accuracy rate of 91.8% for reading & recognizing the test result (a rate either very near to or better than the human baseline) for feeding forward to the Bayesian Hierarchical Model.

We have also partnered with healthcare providers to gain access to anonymized symptom and exposure risk questionnaire data linked to antibody test results. This allows us to now predict infection with some degree of certainty purely from questionnaire data.

Our solution’s impact on the crisis

  • Accelerates the antibody testing process. The manufacturing capacity of inexpensive and accurate lateral flow immunoassays is increasing exponentially, shifting the bottleneck from test availability to test administration rate. ImmunoLynk enables decentralized testing, alleviating the burden on hospital departments and research facilities. Our effortless data collection platform is woven into a privacy-focused decentralized storage solution, the nature of which was recently promoted by hundreds of scientists and endorsed by Apple and Google. This data is impossible to alter, hack, or forge by any bad actors.

  • Helps answer the question: Do antibodies provide immunity? Studies of the effects of direct exposure of people who have previously caught COVID-19 are being planned, but are fraught with ethical concerns and high associated costs. By collecting data on healthcare workers, one of the highest risk populations, ImmunoLynk can assess whether there is a substantial reinfection rate and if higher levels of antibodies can reduce that rate. This is a key question that needs to be addressed prior to the implementation of any wide-scale immunity-based solutions.

  • Inform the immunity passport debate and provide solutions. As we assess the protective effect of antibodies against reinfection, we can provide much-needed information on whether immunity passports would have a significant societal benefit. Our decentralized storage of information would also make passport data safe and immutable.

The value of our solution after the crisis

  • Our systemized data from tracking antibody levels & predicting the level of immunity will remain important for possible future waves of COVID-19 over the coming months to years. It will also help to determine if antibody count actually confers immunity and, if so, to what extent. Even as large scale vaccinations are rolled out, our solution could easily be deployed for individuals to assess their immune response.

  • ImmunoLynk is just as applicable to other pandemics for which lateral flow immunoassays are widely used, such as malaria, hepatitis, and dengue. ImmunoLynk can also help aid diagnosis of an active illness if adapted to use antigen immunoassays. Both could even be incorporated into a single test.

  • At the US level, our use of decentralized technology is in line with the nature of the The Cures Act, which "aims to empower Americans with their health data, delivered conveniently to computers, cell phones, and mobile applications". We are actively pursuing integration with the standardized API denoted in The Cures Act. Moreover, it allows a platform for managers to silo their employee's healthcare information, a requirement of the ADA. This easy-to-deploy management of healthcare information is massively scalable to outside industries, such as nursing homes, factories, travel & hospitality, and so much more.

  • At the EU level, our use of decentralized technology will be instrumental in advancing the EU’s health data sharing initiative. The European Commission has previously recommended that, in order to make further progress in interoperability, developments in digital technologies such as artificial intelligence, high-performance computing, decentralized technologies, and cybersecurity solutions should be carried out, noting that it would increase trust and general feelings of accountability in government.

  • Globally, our approach of tapping into a constant stream of data will greatly benefit research specifically linked to antibody testing (e.g. virology) and, more generally, immunity. Considering healthcare workers experience the most regular exposure to COVID-19, they are arguably the best individuals for identifying correlations & trends in the study of immunity. Epidemiologists could also use our data for pandemic modeling (SIR & SEIR), allowing better preparation for future pandemics. Lastly, it could propel future healthcare adoption of blockchain technology — possibly dispelling the popular notion that it is only useful in cryptocurrency, financial technology, & supply chain management.

What's next?

  • We need more images of lateral flow immunoassays to improve the accuracy of our image classification algorithm. We already have thousands of images of lateral flow immunoassays, but more images of positive results and of assays from different manufacturers will help to ensure maximum reliability of our ResNet-50-based CNN classifier.

  • We need more questionnaire answers linked to immunity test results to improve our prediction algorithm. We are in discussions with UC Berkeley (The University of California, Berkeley) and UCSF (The University of California, San Francisco) to utilize our platform in their large scale studies to further improve our prediction algorithm. We have also been collaborating with COV-CLEAR for the integration of our platform into their clinical trials in the UK.

  • We need introductions to hospital administrators and government officials. Our market research demonstrates considerable widespread interest in a platform like ours that streamlines and improves the testing process; we just need the opportunity to prove that our system works and iron out any minor problems. We are currently working with Lumiata to introduce our platform to one of their healthcare partners.

+ 9 more
Share this project: