Image Credit: Nikkei Asian Review
By Tia Jain
With about 2.5 million confirmed cases and nearly 170,000 deaths world-wide, the COVID-19 pandemic has caused a global catastrophe, as healthcare facilities find themselves struggling to treat overwhelming numbers of patients with insufficient supplies. Nobody would have thought that this year, words like "social distancing" and "quarantine" would become commonplace. In the midst of this global pandemic, many scientists are wondering if and how AI can be applied to help solve COVID-19. However, since AI is still a budding field, much of its potential still remains unexplored, particularly its application to healthcare.
Upon closer inspection of the pandemic's growth, the Harvard Business Review reveals that the main reason why the virus was not contained in its infancy is because our global economy and health care systems are "geared to handle linear, incremental demand," while COVID-19 grows at an exponential rate. Our national health system cannot keep up with this kind of demand, so naturally, many researchers have turned to technology, specifically AI, to find solutions that are both effective and have global coverage.
The first step in handling any virus is being able to diagnose it properly. According to the MIT Technology Review, COVID-Net does exactly that. Developed by Linda Wang and Alexander Wong at the University of Waterloo and the AI firm DarwinAI in Canada, COVID-Net is a convolutional neural network that can help spot COVID-19 in chest x-rays. A convolutional neural network (CNN) passes an image through a series of fully connected layers to get a classification. This makes image classification significantly easier; image classification is a process in computer vision that can classify an image into one of many pre-established categories according to its visual content. For example, a CNN used for image classification may be able to classify an image as a dog if it contains four legs, ears, a tail, fur, and other dog-like characteristics.
In terms of the training database for the model, COVID-Net, specifically, was trained on 6000 images from varying lung diseases. In general, the more data you train your model on or the larger your dataset is, the more accurate your results will be. In terms of the percent of total data that should be used for training versus testing, the ratio is typically 80:20. Reserving 20% of data only for testing prevents an common issue in ML called overfitting, which is when the model reflects the training data too well, consequently performing poorly on the testing data.
Once you identify who has the virus, the second step is stopping the spread of the virus. According to the MIT Technology Review, Andrew Ng's startup called LandingAI can alert people if they are not "social distancing," or staying six feet away from another person. Once embedded in a security camera system, a trained neural network identifies people from a scene in real-time and after being calibrated to real-world dimensions, a second algorithm computes the distances between them.
In addition to addressing COVID-19, AI can also be used for several healthcare-focused solutions. For example, doctors at OrthoAtlanta have recently begun to use Suki, which is "an AI-powered, voice-enabled digital assistant for doctors that is designed to ease the burden of documentation." This enables doctors to focus on treating patients, rather than writing down notes while the patient is speaking. Suki effectively understands voice commands and uses them to create clinically accurate notes that upon confirmation, are inputted into the electronic health record system. Clearly, as AI advances, more healthcare systems are leveraging its power, revolutionizing the way that doctors treat their patients and diagnose illnesses.
That being said, despite the abundance of benefits that the use of AI provides, it is important that consumers are always wary of the accuracy of their AI-generated advice. In fact, the most pervasive issue is that people might take AI-generated advice too seriously. For example, if an AI system tells someone that their symptoms pertain more to the normal influenza rather than the Coronavirus, the individual should seek medical help if they are still concerned or feel otherwise. It should be known that currently, AI is not advanced enough to substitute for a physical checkup by a human doctor. So while AI can help you understand or be a base guideline in interpreting your symptoms, it is in no means a replacement for humans; always take advice with caution.
Second, the use of AI presents ethical issues around storing personal data. Although LandingAI is an incredible application of machine learning to increase safety, it also raises many privacy concerns. Specifically, since LandingAI gathers data real-time from real people, there must be an appropriate consent process. It is unethical to film anyone without their consent, so before being implemented in any public setting, all users must agree to be recorded for safety purposes, or else their personal rights are being violated.
Every day, breaking news stories regarding COVID-19 are on headlines of every news station. With so much information being provided to us, it can be difficult to figure out which sources of media to trust. If you are interested in reading specifically about the latest coverage of the Coronavirus and tech, subscribe to the New York Times Coronavirus Briefing or the Algorithm from MIT Technology Review Coronavirus Newsletter for a reliable rundown of the newest updates! Also, make sure to practice social distancing and to adhere to the safety guidelines given by the CDC and the WHO.
A final question for the readers: If you could apply AI to either the detection, spread, or treatment of COVID-19, which would you pick? What ideas do you have in applying any machine learning principles to increasing social welfare and decreasing the growth in the number of cases? Also, do you think that the pros of using AI outweigh the cons? I'd love to hear from you. Feel free to share your thoughts in the comments below!
Thank you for reading and stay safe!
If you are interested in reading more about the research I mentioned in the article specifically, check out the following links!