An AI algorithm anticipates whether a COVID-19 test will be positive or negative.
A study demonstrates that COVID-19 infections may be predicted using machine learning models that were trained using basic symptoms and demographic characteristics.
Both domestically and internationally, infections are still being caused by COVID-19 and its most recent Omicron strains. The two most popular techniques for quick COVID-19 testing are serology (blood) and molecular tests. The COVID-19 tests differ greatly because they employ various processes. While serology tests look for the presence of antibodies brought on by the SARS-CoV-2 virus, molecular testing evaluates the presence of viral SARS-CoV-2 RNA.
There is currently no research on the relationship between serology and molecular testing or which COVID-19 symptoms are crucial for a positive test result. The College of Engineering and Computer Science at Florida Atlantic University conducted a study utilizing machine learning to better understand the correlation between molecular testing and serology tests as well as the attributes that are most beneficial in identifying COVID-19 positive test results.
Five classification algorithms were trained by researchers from the College of Engineering and Computer Science to forecast COVID-19 exam results. They used easily accessible symptom characteristics, combined with demographic information like age, gender, fever, and the number of days from the beginning of the symptoms, to construct an accurate prediction model.
The study shows that machine learning models can predict COVID-19 infections when they are trained using basic symptom and demographic variables. The findings, which were published in the journal Smart Health, pinpoint the major symptoms linked to COVID-19 infection and offer a method for quick screening and affordable infection identification.
The findings of the COVID-19 test show that the duration of symptoms, such as temperature and breathing problems, has a significant impact on the outcomes. Additionally, research indicates that when compared to post-symptom onset days of serology tests, molecular tests have substantially narrower post-symptom onset days (between three and eight days) (between five to 38 days). Since the molecular test monitors an active infection, it has the lowest positive rate.
Additionally, there are substantial differences amongst COVID-19 tests, in part because donors' immune responses and viral loads, which are the focus of many test methodologies, are always changing. It may be possible to see various positive/negative results from the two distinct types of testing, even for the same donor.
Serology tests rely on seroconversion, the time frame during which the body begins to produce detectable quantities of antibodies, while molecular testing depends on viral load. Both of these tests are time-dependent, according to senior author Xingquan "Hill" Zhu, Ph.D., a professor in the Electrical Engineering and Computer Science Department at FAU. The number of days post-symptomatic is extremely significant for a positive COVID-19 test, according to our findings, and should be carefully taken into account when screening patients.
Researchers used test results from 2,467 donors who were each subjected to one or more COVID-19 tests as the testbed for the study. They created a collection of characteristics for predictive modeling utilizing the five different types of machine-learning models by combining symptoms and demographic data. They investigated the association between serology and molecular testing by comparing test types and outcomes. They used the serology or molecular test findings to categorize the 2,467 donors as positive or negative for test outcome prediction, and they developed symptom characteristics to represent each donor for machine learning. "We classified related symptoms into bins since COVID-19 creates a broad variety of symptoms and the data gathering technique is fundamentally error-prone," stated Zhu. Without symptom reporting standards, the range of symptom features significantly expands. In order to overcome this, we used this binning technique, which was successful in reducing the symptom feature space while preserving sample feature data.
These predictive models achieved more than 81 percent AUC scores (Area under the ROC Curve, which provides an aggregate measure of performance across all possible classification thresholds) and more than 76 percent classification accuracy by utilizing created bin features in combination with the five machine-learning algorithms.
One special aspect of our testbed, according to Zhu, is that certain donors may have several test results. This allowed us to compare serology tests and molecular testing and determine the consistency between each type of test.
The researchers employed the Random Forest, XGBoost, Logistic Regression, Support Vector Machine (SVM), and Neural Network machine learning methods. Three performance metrics—Accuracy, F1-score, and AUC—were used to compare results.
Predictive modeling is confounded by a number of perplexing problems that have no conclusive answers. According to Stella Batalama, Ph.D., dean of the FAU College of Engineering and Computer Science, the testbed developed by our team "is certainly new and clearly shows an association between different types of COVID-19 exams." "Our researchers have created a novel technique for clinical interpretation and predictive modeling of noisy symptom characteristics. Infectious disease prevention and many other facets of health challenges are being addressed with the use of such AI-based predictive modeling techniques.
Magdalyn E. Elkin, a Ph.D. candidate in the Electrical Engineering and Computer Science Department at FAU, is a co-author of the study.
The National Science Foundation funded this project.
- FAU -
About FAU’s College of Engineering and Computer Science:
In the fields of computer science, artificial intelligence (AI), computer engineering, electrical engineering, biomedical engineering, civil, environmental, and geomatics engineering, mechanical engineering, and ocean engineering, the FAU College of Engineering and Computer Science is renowned for its cutting-edge research and instruction. Students are exposed to technological advancements through research done by teachers and their teams, pushing the boundaries of the disciplines' existing state-of-the-art. The National Science Foundation (NSF), National Institutes of Health (NIH), Department of Defense (DOD), Department of Transportation (DOT), Department of Education (DOEd), State of Florida, and business fund the research endeavors of the College. The FAU College of Engineering and Computer Science provides degrees with a contemporary twist that have concentrations in fields of national importance such as artificial intelligence, cybersecurity, the internet of things, logistics, and data science. New degree programs include the Professional Master of Science and Ph.D. in computer science for working professionals, the Master of Science and Bachelor in Data Science and Analytics (first in Florida), and the Master of Science in AI (first in Florida).
About Florida Atlantic University:
The sixth public institution in Florida, Florida Atlantic University was founded in 1961 and formally inaugurated its doors in 1964. The institution presently serves over 30,000 undergraduate and graduate students at its six sites around the southeast Florida coast. In recent years, the University has doubled its research expenditures and outpaced its peers in terms of student achievement rates. FAU is a leading example of an innovative approach where traditional success inequalities disappear because of the cohabitation of access and quality. In addition to being recognized as a top public university by U.S. News & World Report and as having a high level of research activity by the Carnegie Foundation for the Advancement of Teaching, FAU is classified as a Hispanic-serving school.
