In the heyday of artificial intelligence (AI), many were afraid to lose their jobs to something initially designed to make lives better. Amid the COVID-19 outbreak, AI has come to their rescue, and has been tremendously used to protect the whole world for a better and safer tomorrow.
by Dr Allen Lai, Dr Ying-Ja (Inca) Chen, and Jen-Hao Cheng
AI as a Whistle-Blower
Believe it or not, the first global alarm on COVID-19 was in fact triggered by an AI tool called HealthMap. HealthMap is designed to scan online media and detect for signs of disease outbreaks.1 But how did HealthMap blow the whistle?
It was on 30 December 2019 when HealthMap incidentally captured information related to a new type of pneumonia, and its development in China’s Wuhan City, one of the early epicentres of COVID-19 pandemic.
Soon after this online take-up, the system circulated email bulletins to its host located in the Boston Children’s Hospital, as well as to HealthMap’s subscribers. Many were later independently alerted.
The next day, 31 December 2019, the message travelled to reach Dr Philips Lo, Deputy Director General of Taiwan’s Center for Disease Control and Prevention (CDC). He was alerted through an online bulletin board (i.e., PTT, similar to Reddit), regarding the occurrence of atypical pneumonia in the city of Wuhan. The World Health Organization was subsequently informed by Taiwan’s CDC on the same day.
The aforementioned example, stress the importance of scanning online (social) media, in which AI has remarkably held an edge. While it is unfortunate that reaction to the early warnings of COVID-19 may have been slow. Demonstrating the use of AI in identifying potential disease outbreaks through strengthening of surveillance systems, and the monitoring of disease outbreaks should complement conventional approaches.
AI as a Screening Tool
The COVID-19 outbreak escalated tremendous surge in demand for mass screening, which has revealed how conventional diagnostic methods are unable to provide sufficient capacity.
While each country is poised to strengthen its multi-tiered screening approach, AI-based automated classification of disease in contrast to normal status is able to swiftly assist physicians in making judgements on a patient’s condition.
Two advantages of AI-based models are leveraged in developing screening and diagnostic tools:
- First, deep convolutional neural networks can make accurate classifications from image data and separate infected patients from others by training them with a large enough number of medical imaging data.
- Second, AI is able to integrate information from multiple sources, and of various types, into one model for prediction. Smartphone apps in a similar vein are developed to collect data of symptoms, demographics, geographic and contact information. All this in turn collectively provides a powerful prediction within the reach of the general public. The study below illustrates AI’s power in prediction.
In the United Kingdom (UK), a COVID-19 Symptom Study application was developed to collect symptom data from 2.5 million British participants. Using a regression model, researchers found that a combination of loss of smell and taste, fatigue, persistent cough and loss of appetite was the best to predict COVID-19 infection.2
As a result, the model could reach a sensitivity of 65 percent and specificity of 78 percent in screening. Although such predictive analytics has room for improvement, it current functionality is able to accurately diagnose and allows app users to self-identify whether they need to be tested based on their symptoms.
AI as a Diagnostic Tool
The presence of COVID-19 in a patient is confirmed by molecular tests which includes RT-PCR, antibody, or serology test, and imaging modalities such as chest X-ray and computed tomography (CT). The shortage of test kits however has forced medical practitioners to seek alternative screening methods. This is where AI models can be used to train and improve the efficiency and performance of COVID-19 diagnosis.
To make this a reality, several research groups have leveraged on patients’ X-ray and/or CT images to train deep convolutional neural networks, or machine learning classifiers for example, support vector machine or random forest model. These efforts have amazingly reached up to 99 percent accuracy in classification of COVID-19 patients.3-6
In addition to conventional methods, diagnostic devices originally developed to screen for environmental contaminants have been revamped to screen for the novel coronavirus. Botanisol Analystics, a company in the US, has developed a tool that shines a laser to a patient’s sample and using light scattering patterns to detect pathogens.
By training machine learning models on the light scattering patterns, this device is designed to also be used for screening of COVID-19, potentially offering a new way of screening that can be performed in real-time.7
AI as a Contact Tracer
After one being diagnosed with COVID-19, the next crucial step is to trace the root of transmission, and prevent the contagion from spreading further.
To achieve this, health authorities need to identify those who are recently in contact with the index case, a process called contact tracing. Generally, the process traces up to a period of 14 days of close contact with the confirmed case.
What AI does is process the information accumulated by various sources, such as GPS (Global Positioning System), mobile data and credit card transaction data. Through graph theory algorithms, AI-powered analytics is capable of capturing potential origins of the transmission, and locating possible vulnerable individuals. Till now, up to 35 countries have to some extent employed such AI enabled applications to contact trace confirmed and suspected cases.2
Undoubtedly, the use of AI-facilitated contact tracing has been criticized as to breach data security and privacy. The trade off, however, is clear. Whether the laws would allow the public authority to access personal data, especially those related to disease control measures, in order to maintain health security for the whole society is subject to each government’s policies. For example, in Israel “passed an emergency law to use mobile phone data” to control the COVID-19 crisis.8 Some countries like in Taiwan have taken early actions due to past experiences.
After the Severe Acute Respiratory Syndrome (SARS) outbreak in 2003, Taiwan’s Communicable Disease Control Act9 allowed the collection of personal location data for disease prevention. During the current COVID-19 outbreak, Taiwanese health authorities were able to trace the roadmap of the infected individual and send self-health management notices to others who may be in proximity of the infected individual. Prevention and diagnosis are not the whole picture of pandemics management. The next step would be to assess the role of AI in vaccine development and drug discovery against COVID-19.
AI as a Catalyst in Vaccine Discovery and Drug Development
Prevention of COVID-19 requires the development of vaccines. Development of a vaccine typically involves the identification of protein or genetic material of the virus to stimulate an immune response. Identification of viral genetic make-up traditionally takes months. AI, on the other hand, is able to accelerate this process. How is AI able to make this possible? To explain how it works, we need to understand the workings of AlphaFold.
AlphaFold was introduced by Google DeepMind. As a deep learning-based AI, AlphaFold can predict the 3D structure of a protein based on nucleic acid sequence. DeepMind was able to predict and release multiple protein structures of the COVID-19 virus. In addition, natural language processing methods that treat protein or genetic sequences as codes have also identified multiple vaccine candidates.
AI has shortened the identification process and facilitated vaccine discoveries. As for drug development, what’s the role of AI? The answer lies in drug reposition.
A repositioned drug goes directly into pre-clinical testing and clinical trials, bypassing many initials, making it an appealing approach in the COVID-19 situation. The main hurdle in drug repositioning is identifying candidates that can target COVID-19. This is where AI plays a part.
AI expedites this process by speedily screening through approved drugs that have the potential to fight the virus. Similarity between COVID-19 and SARS allowed the use of supervised learning, drugs successful against SARS are used to train models for predicting the efficacy of the drug on COVID-19.
Unsupervised models such as Generative Adversarial Network (GAN) models have been used to generate and simulate novel interaction of the drugs with COVID-19, which mends the differences between COVID-19 and other known viruses.
Conclusion
AI is reliant on the input of an ample amount of data. In this article, it is clear the use of AI is very promising to combat against the COVID-19 pandemic at various stages of pandemic management. From early warnings, screening and diagnostics, to contact tracing, AI has shown more efficacious when the back-end data analytics is deployed timely for decision making. AI-facilitated vaccine discoveries have already identified multiple candidates which would otherwise takes months to pinpoint. The AI-powered drug repositioning is promising to overcome the main hurdle of screening potential drugs, adding alternative treatment options for COVID-19 patients. Though at its preliminary stages, AI sheds light to make us ready for future pandemics. [APBN]
References
- Cho, A. (2020). Artificial intelligence systems aim to sniff out signs of COVID-19 outbreaks. Science. https://doi.org/10.1126/science.abc7698
- Menni, C. et al. (2020). Real-time tracking of self-reported symptoms to predict potential COVID-19. Nature Medicine, 26, pp.1037–1040.
- Lalmuanawma, S. et al. (2020). Applications of machine learning and artificial intelligence for Covid-19 (SARS-CoV-2) pandemic: A review. Review Chaos Solitons Fractals, 139, 110059.
- Ardakani, AA. et al. (2020). Application of deep learning technique to manage COVID-19 in routine clinical practice using CT images: results of 10 convolutional neural networks. Comput Biol Med, 121, 103795
- Ozturk, T. et al. (2020). Automated detection of COVID-19 cases using deep neural networks with X-ray images. Comput Biol Med, 121, 103792.
- Mei, X. et al. (2020). Artificial intelligence–enabled rapid diagnosis of patients with COVID-19. Nature Medicine, 26, 1224–1228.
- Burke, C.W. (2020 August 26). Detecting Coronavirus’ Structural Fingerprint as a Screening Method. BioSpace. https://www.biospace.com/article/detecting-coronavirus-structural-fingerprint-as-a-screening-method/
- Tidy, J. (2020 September 4). Coronavirus: Israel enables emergency spy. BBC. https://www.bbc.com/news/technology-51930681.
- Taiwan. Ministry of Health and Welfare. Communicable Disease Control Act, Article 48 and 58. Taiwan. National Development Council. Personal Data Protection Act, Article 15.
About the Authors
Dr. Allen Lai, Medical Director, Southeast Asia, Ferring Pharmaceuticals.
Dr. Ying-Ja (Inca) Chen, Associate Director,
Bioinformatics and Artificial Intelligence
Division, ACT Genomics.
Jen-Hao Cheng, Manager, Artificial
Intelligence Department, ACT Genomics.