Use of Artificial Intelligence on spatio-temporal data to generate insights during COVID-19 pandemic: A Review
Summary (4 min read)
1. Introduction
- The COVID-19 pandemic caused by the novel coronavirus (SARS-CoV-2) exemplifies the vulnerabilities of healthcare systems in the face of unforeseen adversity.
- In fact, AI systems assist healthcare by performing tasks such as diagnosis, risk assessment, forecasting and surveillance, and health policy and planning.
- To assist further research in the field, this plethora of articles have been summarized and evaluated in several review articles.
- These review articles discuss the impact of AI in a general sense i.e. without further categorization into specific topics.
2. Search Strategy and Selection Criteria
- The literature survey for research papers on existing work on COVID-19 (or pandemics in general) and AI was done using IEEE Xplore, PubMed and Google Scholar for articles published between the time period 2019 and 2020 September.
- Only articles written in English were considered in this review.
- The keywords used were [”COVID-19”, ”pandemic” or ”sars-cov-2”] and [”AI” or ”artificial intelligence” or ”machine learning” or ”pattern recognition” or ”prediction” or ”forecasting”] and [”time” or ”location”, ”demography” or ”country” or ”spatio-temporal”].
- Papers were shortlisted from peer reviewed papers published or in press for journals indexed in Scimago.
- The texts were examined to pick 49 research papers that matched the scope of this survey.
3. Results
- The authors literature survey showed that spatio-temporal forecasting was performed either as a time series forecasting problem or as a diversity aware analysis by taking the diversity of populations into account via one/multiple parameter(s).
- The parameters being used are split into groups based on whether the parameter is inherent to the individuals , the environment or an activity performed by the individuals .
- The papers are grouped into sections depending on the parameters used to model the affected parties as given in Table 1.
3.1. Spatio-Temporal modeling and forecasting
- The first cluster of research papers on modeling and forecasting was based on spatial and temporal data alone.
- 15 Several methods including statistical approaches, machine learning techniques, deep learning models, and time series models have been evaluated in this landscape.
- The various AI models used and example papers are listed in Fig. 3.
Temporal modeling:
- Temporal forecasting of COVID-19 infection cases have been evaluated using multiple deep learning models.
- The Bi-LSTM model and the ConvLSTM models are shown to outperform the other LSTM variants with a Mean Absolute Percentage Error (MAPE) of 3% for day ahead predictions.
- A novel statistical method has been developed to forecast the spread of COVID-19 globally.
- This method considers lead-lag effects between different time series using the dynamic time warping technique to analyze non linear relationships among nations.
- The study concludes that the best parameter to assess the effectiveness of confinement and risk of virus diffusion, is the average number of daily contacts in a population (c).
Spatio-temporal modeling:
- Researchers have focused on modeling and forecasting the spread of COVID-19 in the spatio-temporal domain to identify not only ‘‘how much” the disease has spread but also ‘‘where” it has spread to.
- The model takes into account effects of different containment strategies, event and contact restrictions, and different courses the infection may take; which is not possible when dealing with traditional SIR models.
- A novel approach is taken using Moran’s I test to find regional correlations of COVID-19 cases in China.
- The spatial association between states are modelled into the following six types:30 1) Sharing of borders 2) Euclidean distance 3) Population 4) Population density 5) Number of doctors and hospitals 6) Number of medical beds.
- This shows that similar to attributes such as population, underlying spatio-temporal attributes such as inter-state travel can be employed to predict the dynamic spread of the pandemic.
3.2. Modeling the impact on diverse populations
- The authors review methods that group the population into clusters based on different factors such as occupation, age and other such parameters and analyze COVID-19 spread in those demographics.
- An overview of what characterizes the population diversity is given in Fig.
- The importance of this diversity/context awareness of information for handling the pandemic across the globe has been studied.
- Key challenges in this domain are handling non-uniformly sampled data, incomplete data and figuring out the context without it being readily fed into the models.
- A summary of this section is given in Table 2.
Medical Conditions
- A large-scale analysis of the severity of COVID-19 among the population, given the existing statistics of the medical conditions as per the Global Burden of Disease and Population census among the world, is done using statistical analysis coupled with resampling techniques.
- This study shows the importance of utilizing larger data sources to generate insights during the pandemic.
- Another study aiming to uncover a similar correlation between pre-existing conditions and COVID-19 mortality was conducted on individual patient data from Wuhan using linear regression and statistical techniques.
Age
- The authors discuss the research work attempting to uncover the differential impact of COVID-19 on different age groups.
- The age dependence of COVID-19 ’s impact on children is studied using statistical analysis to thereby generate predictions on the population as a whole by considering data from USA.
- This work goes on to generate recommendations based on the predictions on how best to prepare for large scale outbreaks.
- The age stratified CFR rate (case fatality rate -- the ratio of deaths to the number of total cases) is analyzed using basic statistics.
- Another study performs an in-depth analysis on a larger dataset from around the world (China’s statistics coupled with the rest of the world).
Environmental differences
- The spread of COVID-19 has been analyzed with regard to environmental factors such as temperature and humidity.
- One study uses a spatial Seemingly Unrelated Regressions (SUR) to model the COVID-19 spread as an inter-regional contagion process.
- This doesn’t show the effect of temperature and sunshine on the behavior of the virus, but instead analyzes the people’s behavioral changes based on environmental factors, and thereby the spread of the virus as a result of these changes.
- 41 Statistical models are used to show the absence of a correlation between the daily temperature and case counts in Spain.
- 42 Similarly, the performance of ANN based COVID-19 prediction is evaluated to show how absence of weather data does not make a difference.
Social, Economic and Political factors
- This section discusses the effect of social, economic and political dependencies on the spread and severity of COVID-19.
- The study recommends healthcare resources to be directed towards areas where low testing and high positive rates are prevalent.
- 44 Another study explores spatial relationships between socio-demographics and the spread of COVID-19 in three regions of Texas using ordinary least squares (OLS) and geographically weighted regression (GWR).
- This was evidence of how government policies can affect the spread of COVID-19.
- The authors analyze the effects of no intervention, cyclic mitigation or suppression measures followed by relaxation period.
Travelling and Population flow
- In this section the authors focus on travelling and mass population flow between cities during the COVID-19 pandemic prior to the lockdown measures and its impact.
- Using this, a risk analysis was developed using a Bayesian space-time model to quantify the risk of a person migrating from Wuhan spreading COVID-19 to another population.
- 52 While the similar scenario in India’s mass migration has been studied,53 AI is yet to be used in the Indian situation.
- The former was done by a neural network that can absorb time series information of the patient counts and travel statistics, as well as knowledge about the geographical hierarchy of countries and continents.
- The results show that travel restrictions had a minimal effect unless they were supplemented with other behavioral changes to mitigate the disease spread.
3.2.4. Multivariate inputs
- The methods in these articles attempt to model the complex relationships between different parameters and uncover correlations that might not be obvious in traditional statistical analysis.
- The reduction of variables vastly helps real-time forecasting due to simplicity and reduced computational cost.
- 58 Using the information of COVID-19 spread in the city of Lombardy, an individual level model for COVID-19 transmission that depends on geographical location, age, household structure, and current medical conditions was derived.
- The work also shows the importance of understanding the inter-dependence of the input variables to develop deep learning models coupled with human intuition.
4. Discussion and potentials of AI for future avenues.
- Even though medical systems have been computerized in the past decades, the unprecedented nature of COVID-19 has brought forward issues that need to be automated but are still dependent on human intervention.
- The application of AI for such novel issues poses certain challenges, but overcoming these challenges and exploiting the available methodologies could result in a marked improvement in a variety of fields, especially within the medical domain.
- Similarly, traditional techniques do not handle Big Data efficiently.
- It would be interesting to consider these social clustering parameters for this study, as they may have significant correlation with the spread of pandemic.
5. Limitations and Conclusion
- The rapid development and growing research in AI is amassed as preprints rather than peer reviewed work.
- Since a major portion of the AI research and development ends up in commercial deployment, they are not made available to the public.
- Furthermore, India recorded a large number of COVID-19 patients as well and it’s demographics and socio-economics presented a wide range of unique challenges (mass immigration, social inequalities in access to healthcare) in handling the situation.
- Analysis and reports on this matter that were done in local languages are also not considered in this review.
Did you find this useful? Give us your feedback
Citations
4,408 citations
3 citations
2 citations
2 citations
2 citations
References
73,978 citations
52,856 citations
20,189 citations
3,271 citations
2,949 citations
Related Papers (5)
Frequently Asked Questions (20)
Q2. What are the future works mentioned in the paper "Use of artificial intelligence on spatio-temporal data to generate insights during covid-19 pandemic: a review" ?
The literature survey has uncovered two key application areas in which AI is used -- predicting the future impact of COVID-19 on populations and uncovering the differential impact of COVID-19 on diverse segments of the population. Table 3 lists a few COVID-19 related issues for which AI may be an ideal candidate. 66 Due to the improvement in processing power of mobile devices, AI algorithms can be adapted to solving problems such as proximity identification and improve the accuracy and reliability of these systems through adaptation and real-time learning. Graph Neural Networks ( GNN ) can be used to model and identify specific events in the spatial and/or temporal domain.
Q3. What are the key challenges in the analysis of COVID-19?
Key challenges in this domain are handling non-uniformly sampled data, incomplete data and figuring out the context without it being readily fed into the models.
Q4. What are the main characteristics of the models developed to handle diversity awareness?
Models developed to handle diversity awareness are mostly mappings between higher dimensional spaces (in cases where the context is passed as an input) with the capability to capture the multi-modal behavioral of data.
Q5. What can be done to model the spatial information of the input data?
The spatial information can be modelled through graphs and the information can be propagated within the nodes through message passing.
Q6. How is the prediction of COVID-19 performed?
In addition, a machine learning based random forest model has been used to forecast the number of COVID-19 cases with a mean correlation coefficient R2 of 0·914.22A novel statistical method has been developed to forecast the spread of COVID-19 globally.
Q7. What databases were used for the literature survey?
The literature survey for research papers on existing work on COVID-19 (or pandemics in general) and AI was done using IEEE Xplore, PubMed and Google Scholar for articles published between the time period 2019 and 2020 September.
Q8. What is the amount of investment in R&D in the intersection of AI and healthcare?
A large portion of annual investment (US$ 1·7 billion in 201869) for R&D in the intersection of AI and healthcare comes from tech corporations such as Google and Facebook.70
Q9. What are the main advantages of using deep clustering techniques?
Deep clustering techniques based on convolution neural networks have outperformed traditional methods64 for such clustering problems.
Q10. What is the effect of the COVID-19 epidemic on the population of the affected country?
Another study developed a real-time risk assessment method for future cases in terms of fatality rate ofan affected country, using population density, healthcare resources, and government enforced preventive measures.
Q11. What is the parameter to assess the effectiveness of COVID-19?
The study concludes that the best parameter to assess the effectiveness ofconfinement and risk of virus diffusion, is the average number of daily contacts in a population (c).
Q12. Why are the preprints not made available to the public?
Since a major portion of the AI research and development ends up in commercial deployment, they are not made available to the public.
Q13. What is the need for more rigorous review?
with further proliferation of articles, a need for more rigorous review has emerged, specifically, reviews that catalogue projects based on their specific utilization of AI i.e. cataloguing based on which tasks are achieved by AI and how.
Q14. How can the authors predict the spread of COVID-19?
This shows that similar to attributes such as population, underlying spatio-temporal attributes such as inter-state travel can be employed to predict the dynamic spread of the pandemic.
Q15. How did the researchers forecast the spread of COVID-19 in Italy?
In an attempt to solve the patient triage problem related to COVID-19, researchers in Italy have formulated a discrete time markov chain model to predict the number of COVID-19 cases, to thereby, efficiently allocate ICU resources across the country.
Q16. What was the effect of the ARIMA model on the number of future cases?
The number of future cases was predicted using an ARIMA model, and a regression tree algorithm was used to validate the argument that the aforementioned variables have the highest impact on fatality rates in those countries.
Q17. How many cases of COVID-19 have been estimated in India?
A data driven forecasting method has been developed to estimate the number of positive cases of COVID-19 in India for the next 30 days.
Q18. What is the effect of climate on the spread of COVID-19?
Another similar study (using ANN based differential equations and particle swarm algorithms) shows that population density (followed by relative humidity) has a higher impact on the spread of COVID-19 in comparison to temperature and wind speed.
Q19. What is the way to model the effect of population flow with consideration on to the geography?
Modeling the effect of population flow with consideration on to the geography has outperformed the naive population flow models in terms of accuracy.
Q20. What is the role of data in the prevention of COVID-19?
This data is also crucial in instructing the authorities involved in policy and decision making, and will thereby ensure efficient containment of the virus spread.