This week’s list of data news highlights covers April 10 – April 16, 2021 and includes articles about automating the initial processes for immigration cases in the United States and discovering signs of life on Venus.
Scientists at the NASA Jet Propulsion Laboratory have collaborated with researchers from the University of Alaska to predict volcanic eruptions using satellite data of heat wave emissions from areas surrounding volcanoes. By analyzing data on volcanoes that erupted within the last 16 years, the team discovered that in the years leading up to eruptions, surface temperatures above volcanoes increased one degree Celsius more than normal and decreased after eruptions.
The U.S. Citizenship and Immigration Services (USCIS) is automating the pre-processing of immigration cases that require adjudication using three tools. The first tool uses natural language processing to read through immigration documents and flag potentially fraudulent applications where multiple applicants’ stories do not align. The second tool uses machine learning to comb through biographic and biometric data on applications to identify existing ties to the country applicants may have, such as a marriage or permanent residency status, which changes how their applications are considered. The third tool uses computer vision to sort, classify, and validate submitted evidence by type so that immigration judges do not have to physically sort through papers during hearings.
Researchers at McMaster University in Canada have created machine learning models that predict the type of criminal offense people with psychiatric disorders are likely to commit. The team applied the models to 1,240 patients in the Canadian forensic mental health system database and used 30 variables, such as patient sociodemographic and previous clinical treatments, to predict whether an individual will commit a nonviolent, violent, or sexual offense with 80 percent specificity. With the models, clinicians can adjust risk assessment strategies and implement therapeutic interventions early-on.
Epidemiologists from the University of California Los Angeles and Boston University have partnered with NASA to use satellite data to map different types of air pollutants that are most harmful to human health. The team used a light capturing tool on satellites to measure the optical properties of pollutants based on the polarization of sunlight scattered by their particles, which allowed them to infer the composition of pollutants. Using this data, the team was able to identify regions in 12 cities where there are high concentrations of particles smaller than 2.5 micrometers, which are particles that pose the most severe health risks such as premature death from heart and lung conditions.
Physicians at the Veterans Affairs Boston Healthcare System in Massachusetts have used genetic data to identify existing drugs that can be repurposed to mitigate severe symptoms of COVID-19 that require hospitalization. The team compared the genetic profiles of 7,500 patients hospitalized with COVID-19 to a control group to understand how the medications they prescribed affected their immune response. Through their analysis, they discovered that drugs targeting two proteins called the IFNAR2 and ACE2 proteins could reduce inflammatory responses in patients with severe respiratory disorders and improve the functions of the central nervous system.
The U.S. Food and Drug Administration (FDA) has approved GI Genius, the first AI-based device that uses machine learning to identify polyps or suspected tumors from microscopic videos of routine colonoscopies. GI Genius works by highlighting lesions on the video screen that may require further assessment, and offers suggestions to physicians on what to do next, such as a closer inspection, tissue sampling, or testing of the lesion. When tested, colonoscopies that used the GI Genius identified 55 percent of patients with lab-confirmed cancerous polyps and tumors, 13 percent more than standard colonoscopies alone.
Researchers at Cal Poly Pomona in California have analyzed archived data from 1978 on the composition and abundance of atmospheric gases on Venus. The team’s analysis found that chemical compounds in clouds on the planet, such as phosphine, hydrogen sulfide, and nitric acid, indicate that some clouds are not in a state of equilibrium. This imbalance suggests that life-related chemical processes may be occurring, which would create a habitable environment for microorganisms.
Canon Medical Research Europe, a medical imaging software company in Scotland, has partnered with the University of Glasgow to develop an AI assessment tool that measures tumors that cause mesothelioma, a cancer that develops from inhaling asbestos fibers. Even though Scotland has the highest incidence of mesothelioma globally, treating the tumors has been difficult because of how the cancer grows and its response to chemotherapy. The team tested the prototype on 100 CT scans where clinicians already identified the tumors, and discovered that the tool could identify and measure the size of tumors without human input.
The National Institutes of Health Helping to End Addiction Long-term (NIH HEAL) Initiative is working with the University of North Carolina to create a cloud-based data platform where researchers, policymakers, and other stakeholders can access data to inform their research and policies around the opioid crisis. The platform contains a broad-range of data from 12 health centers. To ensure data is easily discoverable, the team is working on making data in various formats interoperable.
Researchers at the Massachusetts Institute of Technology Data to AI Lab have developed Cardea, an open-source system that hospitals can use to share existing machine learning tools with each other rather than coding models from scratch. To use Cardea, users input a dataset of electronic health records and tell the system what they want to find out, such as how long a patient may stay hospitalized. Cardea then applies a relevant tool to the dataset and predicts what could happen based on patterns found in previous inquiries made from other hospitals. When researchers tested the accuracy of Cardea against other data science platforms, Cardea outperformed them by 90 percent.
Image credit: Ronile