Scientists use wearable technology to detect stress levels during sleep

UVM study is the first to surface perceived stress using sleep data

UNIVERSITY OF VERMONT

 

IMAGE: 

THE LEMURS PROJECT USES OURA RINGS TO GATHER BIOMETRIC DATA FROM FIRST- AND SECOND-YEAR COLLEGE STUDENTS TO DETERMINE WHICH BIOMARKER SIGNAL POTENTIAL CHANGES IN A PERSON'S HEALTH STATUS.

view more 

CREDIT: ANDY DUBACK

What if changes in a person’s stress levels could be detected while they sleep using wearable devices? A new study by University of Vermont researchers published today in PLOS Digital Health is the first to find changes in perceived stress levels reflected in sleep data—an important step towards identifying biomarkers that may help flag individuals in need of support. 

Given how critical sleep is to physical and mental health, the research team suspected signals might exist in sleep data, says Laura Bloomfield, a research assistant professor of mathematics and statistics and lead author of the study. “Changes in stress are visible.”

When parsing baseline sleep data, the researchers found “consistent associations” between people’s perceived stress scores and factors such as total sleep time, resting heart rate and heart rate variability, and respiratory rate. While it’s no surprise that most participants received less than the recommended 8 to 10 hours of sleep for young adults, the minutes do matter. For every additional hour of sleep recorded, the odds of someone reporting moderate-to-high stress decreased about 38 percent. Nightly resting heart rates offered more clues. For each additional beat per minute, the odds of experiencing stress increased by 3.6 percent.

Bloomfield is a principal investigator of the Lived Experience Measured Using Rings Study (LEMURS)—a longitudinal study started at UVM in 2022 that tracks hundreds of first- and second-year college students 24 hours a day using a wearable Oura ring biosensor and through surveys about their wellbeing. This is the first peer-reviewed paper from LEMURS and shows that data gleaned from wearables can reveal changes in people’s mental health status.

“The study showed that sleep measures from the Oura ring were predictive of participants’ perceived level of stress. If we are able to identify in real-time that someone is experiencing increased stress, there might be an opportunity to offer helpful interventions.,” Bloomfield explains. “There are a lot of ways to implement interventions, but the first step is understanding the connection between sleep measures and mental health measures.”

About LEMURS

The LEMURS project was conceived by Chris Danforth, professor of applied mathematics at UVM’s Vermont Complex Systems Center and fellow of the Gund Institute for Environment and Bloomfield, MD/Ph.D., to determine how wearable technologies could be used to improve young people’s health and well-being with personalized health feedback. LEMURS is supported by a grant from MassMutual.

College students, in general, don’t sleep enough, often feel stressed, and are at greater risk of experiencing mental health issues. The LEMURS research team will also evaluate the effectiveness of interventions such as exercise, excursions into nature, and group therapy—all interventions which have previously shown improvements in health and wellbeing—to understand which work best and how quickly scalable they are for large populations. But to do all of this requires identifying biometric data that provide the clearest signals for addressing changes in physical and mental health—a process that involves gathering and sifting through millions of hours of data each year.

LEMURS participants wear Oura rings that quietly collect measurements including temperature, heart rate, breathing rate, and nightly sleep duration as well as complete routine surveys to collect more subjective responses about potential stressors and their emotions. Location information is also used to calculate the exposure participants have to nature. All this data is then combed by LEMURS researchers like Mikaela Fudolig, research assistant professor of mathematics and statistics, who test specific relationships that could be used to develop health interventions. She co-authored the PLOS Digital Health paper and says there is power in the study’s numbers.

Initially, 600 first-year students aged 18 to 20 enrolled in LEMURS. A second cohort of first-year students was added in fall 2023 with a goal of following these individuals through college and far into the future.

“We have been tracking the same students for almost two years now, and there are very few studies that do that,” says Fudolig, research assistant professor of mathematics and statistics, who co-authored the PLOS Digital Health paper and says there is power in the study’s numbers. “We have several sources of data. Taking these all together—your ring data, your survey data, your nature-dose data, we also have blood work done—we will see a lot of different dimensions from these participants. So, combining them is, to me, the most exciting thing of it all.”

These potential predictors of stress led to a sleep analysis of LEMURS participants by Fudolig which detected two distinct heart rate curves, particularly among women. We find that those who reported an impairment in their daily life due to anxiety or depression had heart rates that dropped later in the night, she explains.

“A high burden of stress”

The COVID-19 pandemic worsened mental health problems for an already vulnerable population. In the decade before COVID, the Center for Disease Control and Prevention found the percentage of high school students nationwide experiencing persistent feelings of sadness or hopelessness jumped from 26.1 percent to 36.7 percent. The CDC’s 2021 Youth Risk Behavior Survey showed another jump—42 percent of students reported feeling persistently sad.

This is one reason Danforth and Bloomfield launched LEMURS in the first place.

“There is a high burden of stress in this population,” Bloomfield says, “College is seen as a very carefree period of time where you are coming into your own, but it’s also a period with a lot of transition and a lot of additional stressors. There needs to be better, accessible support systems for young adults during this time.

She wasn’t surprised to learn that perceived stress scores of LEMURS participants were high—64% of responses were considered moderate-to-highly stressed. These are personal assessments of how individuals feel about problems they encounter and their ability to manage them and responses vary depending on one’s life experiences, personality, support, and coping skills. Part of the challenge with interpreting stress signals using biometric data is figuring out when deviations from someone’s baseline are problematic and concerning, Bloomfield explains.

“This is a resilient population, they are young and healthy,” she continues. “But I think this study is bringing to light important issues facing this population. The ultimate goal with our research is that you can help support people in times of decreased mental health or physical health status.”

Additional UVM researchers involved in this study include Julia Kim, Jordan Llorin, Juniper Lovato, Matt Price, Taylor H. Ricketts, Peter Sheridan Dodds, Kathryn Stanton, and Christopher M. Danforth, working with Ellen McGinnis and Ryan McGinnis (Wake Forest University). This is the first peer-reviewed paper from UVM’s Lived Experience Measured Using Rings Study, a longitudinal investigation using wearable technologies to detect and incentivize positive changes in physical and mental health. The study is funded by a grant from MassMutual.

---

JOURNAL

PLOS Digital Health

DOI

10.1371/journal.pdig.0000473 

METHOD OF RESEARCH

Experimental study

SUBJECT OF RESEARCH

People

ARTICLE TITLE

Predicting stress in first-year college students using sleep data from wearable devices

ARTICLE PUBLICATION DATE

11-Apr-2024

 

Thin oil films enable stable oil and water mixtures sans surfactant

(SURFACTANTS ARE DETERGENTS)

AMERICAN ASSOCIATION FOR THE ADVANCEMENT OF SCIENCE (AAAS)

Thin oil films absorbed onto the surface of water droplets lead to anomalously stable, surfactant-free oil and water mixtures, according to a new study. The findings demonstrate a mechanism for stabilizing water droplets in a water-oil emulsification without the need for a surfactant, which could have important technological applications, including the creation of very pure and controlled materials. Oil and water cannot form homogenous mixtures. Instead, when combined, droplets of one fluid will disperse inside the other, forming an emulsion. However, when two like droplets approach one another, the thin film separating them becomes unstable, causing them to coalesce into a single, larger droplet. Droplets will continue to merge until the fluids fully separate. The only known way to stabilize droplets and prevent them from spontaneously coalescing is by adding a surfactant to the mix. Claire Nannette and colleagues discovered an exception to this rule and report the spontaneous formation of stable water droplets within a polymeric oil. Nannette et al. show that when a thin layer of oil absorbs onto the surface of a water droplet, the oil at the interface becomes far more viscous than the surrounding bulk oil. This results in an attractive interaction between the two water droplets that drives a dramatic change in the dynamics of the oil molecules within the thin film separating the water droplets. This in turn leads to an extreme slowing of the formation of holes in the oil film separating two drops, which cause coalescence in other oil-water mixtures. According to the authors, water droplets in the polymeric oil can remain stable over timescales of several weeks, without the use of any surfactant or solvent.

---

JOURNAL

Science

DOI

10.1126/science.adj6728 

ARTICLE TITLE

Thin adhesive oil films lead to anomalously stable mixtures of water in oil

ARTICLE PUBLICATION DATE

12-Apr-2024

 

*FREE* Growing tribal clean energy in the US

AMERICAN ASSOCIATION FOR THE ADVANCEMENT OF SCIENCE (AAAS)

New US federal legislation sets aside nearly $14 billion for 574 federally recognized indigenous nations and villages, which can be used to support tribal climate responsiveness and energy sovereignty. In a Policy Forum, Kimberly Yazzie and colleagues present a roadmap for designing, implementing, and funding projects and people to accelerate the renewable energy transition while also benefiting the indigenous entities involved. According to the authors, this opportunity positions indigenous communities to develop their economies and energy projects on their own terms to rectify many decades of energy and economic poverty and injustice. While the 2021 Infrastructure Investment and Jobs Act and the 2022 Inflation Reduction Act provide unprecedented opportunities to expand tribal renewable energy, indigenous groups face several barriers to accessing these funds. Here, Yaszzie et al. discuss these challenges and recommend strategies that could improve on funding opportunities that include indigenous partnerships. These include increasing flexibility in the deadlines for federal funding programs and loans, providing technical assistance for clean energy projects, and supporting the continuity of the work conducted by tribal energy projects, including workforce development. “Implementation will require vigilance to ensure that integrating justice throughout agency programs generates tribal wealth without recreating historical injustices. The issues we highlight and the recommendations we make are not all surprising, but they require bold action.”

---

JOURNAL

Science

DOI

10.1126/science.adk8298 

ARTICLE TITLE

Opportunities to grow tribal clean energy in the US

ARTICLE PUBLICATION DATE

12-Apr-2024

Disclaimer: AAAS

 

Study reveals giant store of global soil carbon

CHINESE ACADEMY OF SCIENCES HEADQUARTERS

 

IMAGE: 

THE UPPER SOLUM (MOLLIC EPIPEDON) SHOWS A DARK COLOR BECAUSE OF THE HEAVY ACCUMULATION OF ORGANIC MATTER, WHILE THE LOWER SOLUM (CALCIC HORIZON) IS WHITISH DUE TO THE PRESENCE OF CALCIUM CARBONATE. THE SOIL TYPE IS CALCIC MOLLI-USTIC CAMBOSOLS ACCORDING TO CHINESE SOIL TAXONOMY.

view more 

CREDIT: ZHANG GANLIN

Soil carbon usually refers only to the organic matter component of soils, known as soil organic carbon (SOC). However, soil carbon also has an inorganic component, known as soil inorganic carbon (SIC). Solid SIC, often calcium carbonate, tends to accumulate more in arid regions with infertile soils, which has led many to believe it is not important.

In a study published in Science, researchers led by Prof. HUANG Yuanyuan from the Institute of Geographic Sciences and Natural Resources Research of the Chinese Academy of Sciences (CAS) and Prof. ZHANG Ganlin from the Institute of Soil Science of CAS, together with collaborators, have quantified the global store of SIC, challenging this long-held view.

The researchers found a whopping 2,305 billion tons of carbon stored as SIC in the top two meters of soil worldwide, which is more than five times the carbon found in all of the world's vegetation combined. This hidden pool of soil carbon could be key to understanding how carbon moves around the globe.

"But here's the thing: This huge carbon pool is vulnerable to changes in the environment, especially soil acidification. Acids dissolve calcium carbonate and remove it either as carbon dioxide gas or directly into the water," said Prof. HUANG.

"Many regions in countries like China and India are experiencing soil acidification due to industrial activities and intense farming. Without remedial actions and better soil practices, the world is likely to face a disturbance of SIC in the next thirty years," she added.

Disturbances to SIC accumulated over Earth's history have a profound impact on soil health. This disruption compromises the soil's ability to neutralize acidity, regulate nutrient levels, foster plant growth, and stabilize organic carbon. Essentially, SIC plays a critical dual role in storing carbon and supporting ecosystem functions that depend on it.

The researchers revealed that approximately 1.13 billion tons of inorganic carbon are lost from soils to inland waters each year. This loss has profound but often overlooked implications for carbon transport between the land, atmosphere, freshwater, and ocean.

While society has recognized the importance of soils as a fundamental part of nature-based solutions to combat climate change, much of the focus has been on SOC. It is now clear that inorganic carbon deserves equal attention.

This study underscores the urgency of incorporating inorganic carbon into climate change mitigation strategies as an additional lever for maintaining and enhancing carbon sequestration. International programs such as the "4 per mille initiative," which aims to increase (mostly) SOC by 0.4% annually, should also consider the critical role of inorganic carbon in achieving sustainable soil management and climate mitigation goals.

By broadening the understanding of soil carbon dynamics to include both organic and inorganic carbon, the researchers hope to develop more effective strategies for maintaining soil health, enhancing ecosystem services, and mitigating climate change.

---

JOURNAL

Science

DOI

10.1126/science.adi7918 

ARTICLE TITLE

Size, distribution, and vulnerability of the global soil inorganic carbon

ARTICLE PUBLICATION DATE

12-Apr-2024

 

Wired to learn and remember

New research unveils how spinal cord nerve cells can learn and remember—completely independent of the brain 

VLAAMS INSTITUUT VOOR BIOTECHNOLOGIE

Leuven (Belgium), 11 April 2024 — The role of the spinal cord is often simplified to that of a simple relay station, carrying messages between the brain and the body. However, the spinal cord can actually learn and remember movements on its own. A team of researchers at the Leuven-based Neuro-Electronics Research Flanders (NERF) details how two different neuronal populations enable the spinal cord to adapt and recall learned behavior in a way that is completely independent of the brain. These remarkable findings, published today in Science, shed new light on how spinal circuits might contribute to mastering and automating movement. The insights could prove relevant in the rehabilitation of people with spinal injuries.  

The spinal cord’s puzzling plasticity 

The spinal cord modulates and finetunes our actions and movements by integrating different sources of sensory information, and it can do so without input from the brain. What’s more, nerve cells in the spinal cord can learn to adjust various tasks autonomously, given sufficient repetitive practice. How the spinal cord achieves this remarkable plasticity, however, has puzzled neuroscientists for decades.  

One such neuroscientist is Professor Aya Takeoka. Her team at Neuro-Electronics Research Flanders (NERF, a research institute backed by imec, KU Leuven and VIB) studies how the spinal cord recovers from injuries by exploring how the nerve connections are wired, and how they function and change when we learn new movements. 

 “Although we have evidence of ‘learning’ within the spinal cord from experiments dating back as early as the beginning of the 20th century, the question of which neurons are involved and how they encode this learning experience has remained unanswered,” says Prof. Takeoka. 

Part of the problem is the difficulty in directly measuring the activity of individual neurons in the spinal cord in animals that are not sedated but awake and moving. Takeoka’s team took advantage of a model in which animals train specific movements within minutes. In doing so, the team uncovered a cell type-specific mechanism of spinal cord learning. 

Two specific neuronal cell types 

To check how the spinal cord learns, doctoral researcher Simon Lavaud and his colleagues at the Takeoka lab built an experimental setup to measure changes in movement in mice, inspired by methods used in insect studies. “We evaluated the contribution of six different neuronal populations and identified two groups of neurons, one dorsal and one ventral, that mediate motor learning.” 

"These two sets of neurons take turns," explains Lavaud. "The dorsal neurons help the spinal cord learn a new movement, while the ventral neurons help it remember and perform the movement later."  

"You can compare it to a relay race within the spinal cord. The dorsal neurons act like the first runner, passing on the critical sensory information for learning. Then, the ventral cells take the baton, ensuring the learned movement is remembered and executed smoothly." 

Learning and memory outside the brain 

The detailed results, published in this week’s edition of Science, illustrate that neuronal activity in the spinal cord resembles various classical types of learning and memory. Further unravelling these learning mechanisms will be crucial, as they likely contribute to different ways in which we learn and automate movement, and may also be relevant in the context of rehabilitation, says Prof. Aya Takeoka: 

“The circuits we described could provide the means for the spinal cord to contribute to movement learning and long-term motor memory, which both help us to move, not only in normal health but especially during recovery from brain or spinal cord injuries.” 

---

Publication

Two inhibitory neuronal classes govern acquisition and recall of spinal sensorimotor adaptation. Lavaud, et al. Science, 2024. DOI: 10.1126/science.adf6801

The research (team) was supported by the Research Foundation Flanders (FWO), Marie Skłodowska-Curie Actions (MSCA), a Taiwan-KU Leuven PhD fellowship (P1040), and the Wings for Life Spinal Cord Research Foundation.  

---

JOURNAL

Science

DOI

10.1126/science.adf6801 

METHOD OF RESEARCH

Experimental study

SUBJECT OF RESEARCH

Animals

ARTICLE TITLE

Two inhibitory neuronal classes govern acquisition and recall of spinal sensorimotor adaptation

ARTICLE PUBLICATION DATE

12-Apr-2024

Brainless memory makes the spinal cord smarter than previously thought

RIKEN

 

IMAGE: 

IN THIS STUDY, SPINAL CORDS THAT ASSOCIATED LIMB POSITION WITH AN UNPLEASANT EXPERIENCE LEARNED TO REPOSITION THE LIMB AFTER ONLY 10 MINUTES, AND RETAINED A MEMORY THE NEXT DAY. SPINAL CORDS THAT RECEIVED RANDOM UNPLEASANTNESS DID NOT LEARN.

view more 

CREDIT: RIKEN

Aya Takeoka at the RIKEN Center for Brain Science (CBS) in Japan and colleagues have discovered the neural circuitry in the spinal cord that allows brain-independent motor learning. Published in Science on April 11, the study found two critical groups of spinal cord neurons, one necessary for new adaptive learning, and another for recalling adaptations once they have been learned. The findings could help scientists develop ways to assist motor recovery after spinal cord injury.

Scientists have known for some time that motor output from the spinal cord can be adjusted through practice even without a brain. This has been shown most dramatically in headless insects, whose legs can still be trained to avoid external cues. Until now, no one has figured out exactly how this is possible, and without this understanding, the phenomenon is not much more than a quirky fact. As Takeoka explains, “Gaining insights into the underlying mechanism is essential if we want to understand the foundations of movement automaticity in healthy people and use this knowledge to improve recovery after spinal cord injury.”

Before jumping into the neural circuitry, the researchers first developed an experimental setup that allowed them to study mouse spinal cord adaptation, both learning and recall, without input from the brain. Each test had an experimental mouse and a control mouse whose hindlegs dangled freely. If the experimental mouse’s hindleg drooped down too much it was electrically stimulated, emulating something a mouse would want to avoid. The control mouse received the same stimulation at the same time, but not linked to its own hindleg position.

After just 10 minutes, they observed motor learning only in the experimental mice; their legs remained high up, avoiding any electrical stimulation. This result showed that the spinal cord can associate an unpleasant feeling with leg position and adapt its motor output so that the leg avoids the unpleasant feeling, all without any need for a brain. Twenty-four hours later, they repeated the 10-minute test but reversed the experimental and control mice. The original experimental mice still kept their legs up, indicating that the spinal cord retained a memory of the past experience, which interfered with new learning.

Having thus established both immediate learning, as well as memory, in the spinal cord, the team then set out to examine the neural circuitry that makes both possible. They used six types of transgenic mice, each with a different set of spinal neurons disabled, and tested them for motor learning and learning reversal. They found that mice hindlimbs did not adapt to avoid the electrical shocks after neurons toward the top of the spinal cord were disabled, particularly those that express the gene Ptf1a.

When they examined the mice during learning reversal, they found that silencing the Ptf1a-expressing neurons had no effect. Instead, a group of neurons in the bottom, ventral, part of the spinal cord that express the En1 gene was critical. When these neurons were silenced the day after learning avoidance, the spinal cords acted as if they had never learned anything. The researchers also assessed memory recall on the second day by repeating the initial learning conditions. They found that in wildtype mice, hindlimbs stabilized to reach the avoidance position faster than they did on the first day, indicating recall. Exciting the En1 neurons during recall increased this speed by 80%, indicating enhanced motor recall.

“Not only do these results challenge the prevailing notion that motor learning and memory are solely confined to brain circuits,” says Takeoka, “but we showed that we could manipulate spinal cord motor recall, which has implications for therapies designed to improve recovery after spinal cord damage.”

---

JOURNAL

Science

DOI

10.1126/science.adf6801 

 

Researchers test new behavioral health interventions

The concept of “One Health” – which emphasizes the relationship between human, animal, plant and environmental health – has been gaining ground in scientific discussions in recent years.

FUNDAÇÃO DE AMPARO À PESQUISA DO ESTADO DE SÃO PAULO

 

IMAGE: 

EDUARDO ESTEBAN BUSTAMANTE, PROFESSOR AT THE UNIVERSITY OF ILLINOIS, DURING HIS LECTURE AT FAPESP WEEK ILLINOIS 

view more 

CREDIT: ELTON ALISSON/AGÊNCIA FAPESP

The concept of “One Health” – which emphasizes the relationship between human, animal, plant and environmental health – has been gaining ground in scientific discussions in recent years. Brazilian and North American researchers developing research using this approach presented their work on Tuesday (April 9th), in Chicago (United States), during FAPESP Week Illinois.

One of the panelists was Eduardo Esteban Bustamante, a professor at the University of Illinois in Chicago. He talked about behavioral interventions that have been tested to promote physical activity and healthy eating – practices that, according to the researcher, Americans still do not adhere to, despite the obvious benefits.

“The percentage of Americans meeting nutritional and physical activity recommendations is still very low. Among children aged six to 11, the percentage is 49% for boys and 35% for girls. But as they grow up, these rates get much worse, dropping to 7% and 4%, respectively, in the 16-19 age group, and stagnating at 3% and 2% from the age of 60 onwards," Bustamante said. "When it comes to diet, the reality isn't much better. Just over 10% of American adults over the age of 18 routinely eat fruits and vegetables,” the researcher said.

In an attempt to change this reality, behavioral health intervention programs have been developed and tested across the country. In the last few years alone, more than 3,000 evidence-based physical activity and nutrition interventions have been created in the United States. Of these, around 200 are available in public repositories for use by the public, according to a survey conducted by the researcher.

“These practices are made available on public websites. That way, people can get access to them and follow the instructions correctly to become more active and eat more fruits and vegetables, for example,” Bustamante said.

The problem, however, is that 90% of these scientifically tested physical activity intervention programs in the United States face barriers to dissemination and implementation that limit their potential impact on public health. One of the contributing factors is a lack of alignment with people’s expectations and with the places where they should be implemented, the researcher said.

“I’ve worked with a number of intervention programs, and one of the problems I’ve identified is that we didn’t think about the target audience before we started, and we saw that people weren’t engaged with them. We need to think about how to engage the target audience so that, from the beginning, our interventions fit in and are aligned with their goals,” he said.

“We also need to stop thinking of nutrition and physical activity as medicines that can only benefit health. They’re activities that take place in a context and we can use them to achieve the goals we want, whether they’re health-related or not,” said Esteban.

Based on this finding, the researcher and his collaborators have begun developing and testing new physical activity intervention programs in schools and communities.

For example, one project carried out in collaboration with the University of California Irvine has been using physical activity in schools as a way of learning mathematics. To this end, the basketball court at an educational institution was redesigned to teach children about fractions and decimals.

“The result is that the kids, in addition to getting all the health benefits of doing a physical activity, are learning math in a much more engaging way,” he said.

Another project, implemented in the Chicago Park District, one of the largest and oldest park districts in the United States, has been using sports and recreation to develop communication, emotional and conflict resolution skills in at-risk youths.

“The program works with young people who are in high school. We try to get them jobs during the summer so they can stay in the parks and work during that time, and we encourage them to develop behavioral skills through physical activity,” explained Bustamante.

Multifactorial causes

New approaches to behavioral health interventions are also vital to addressing the diabetes epidemic in the United States, said Marck Rosenblatt, dean of the University of Illinois College of Medicine.

“One in ten people in the United States has diabetes. The causes of this disease are multifactorial. It’s not just because people aren’t taking insulin and medications for hyperglycemia, but also because their diet is inadequate and they don’t exercise,” he said.

“It’ll take a multifaceted approach to address this problem, such as interventions in schools. We’re trying to work with local organizations to try to improve diet quality and encourage physical activity, while at the same time studying the molecular underpinnings of diabetes itself,” said Rosenblatt.

According to the researcher, the social determinants of health are an issue that the Chicago institution and health system have been working hard to understand and intervene in.

“It’s humbling to realize that only around 15% to 20% of a person’s health is related to the solutions we develop in our hospitals and clinics. People’s health is more related to their zip code, which correlates with a number of other factors, such as socioeconomic level, social and community context,” he said.

Environmental risks play a fundamental role in the emergence of degenerative diseases and cancer, emphasized Leandro Colli, professor at the Ribeirão Preto Medical School of the University of São Paulo (FMRP-USP), supported by FAPESP.

“We know that the cause of cancer is not only genetic. There are also very strong environmental factors. We can intervene in the genetic risk factors, but we also have to look at the environment,” he emphasized.

The researcher is working with collaborators on a project aimed at identifying mutational signatures in cancer patients – a concept that has emerged in recent years whereby it is possible to look at a cell mutation and try to recapitulate its origin and the agents that caused it.

“We’re starting a project in which we’re following a series of patients to try to better understand the risk factors for mutations that lead to cancer, such as tobacco, exposure to solar radiation and the burning of sugar cane in the Ribeirão Preto region,” said Colli.

More information about FAPESP Week Illinois can be found at: fapesp.br/week/2024/illinois.

 

New computer vision tool wins prize for social impact

DISCount, created at UMass Amherst, derived from two very different needs: Counts of damaged buildings in crisis zones and bird flock sizes

UNIVERSITY OF MASSACHUSETTS AMHERST

 

IMAGE: 

THE PALU TSUNAMI, WHICH STRUCK INDONESIA IN 2018. THE DATA CONSISTED OF 113 HIGH-RESOLUTION SATELLITE IMAGES. COUNTS OF THE BUILDINGS AND THEIR DAMAGE LEVELS WERE COLLECTED PER TILE USING BEFORE- AND AFTER-DISASTER SATELLITE IMAGES. COLORS INDICATE DIFFERENT LEVELS OF DAMAGE (RED = DESTROYED), AND DISCOUNT GAVE AN ESTIMATE OF DAMAGED BUILDINGS PER SUB-REGION.

view more 

CREDIT: PÉREZ ET AL., 10.1609/AAAI.V38I20.30235

AMHERST, Mass. – A team of computer scientists at the University of Massachusetts Amherst working on two different problems—how to quickly detect damaged buildings in crisis zones and how to accurately estimate the size of bird flocks—recently announced an AI framework that can do both. The framework, called DISCount, blends the speed and massive data-crunching power of artificial intelligence with the reliability of human analysis to quickly deliver reliable estimates that can quickly pinpoint and count specific features from very large collections of images. The research, published by the Association for the Advancement of Artificial Intelligence, has been recognized by that association with an award for the best paper on AI for social impact.

“DISCount came together as two very different applications,” says Subhransu Maji, associate professor of information and computer sciences at UMass Amherst and one of the paper’s authors. “Through UMass Amherst’s Center for Data Science, we have been working with the Red Cross for years in helping them to build a computer vision tool that could accurately count buildings damaged during events like earthquakes or wars. At the same time, we were helping ornithologists at Colorado State University and the University of Oklahoma interested in using weather radar data to get accurate estimates of the size of bird flocks.”

Maji and his co-authors, lead author Gustavo Pérez, who completed this research as part of his doctoral training at UMass Amherst, and Dan Sheldon, associate professor of information and computer sciences at UMass Amherst, thought they could solve the damaged-buildings-and-bird-flock problems with computer vision, a type of AI that can scan enormous archives of images in search of something particular—a bird, a rubble pile—and count it.

But the team was running into the same roadblocks on each project: “the standard computer visions models were not accurate enough,” says Pérez. “We wanted to build automated tools that could be used by non-AI experts, but which could provide a higher degree of reliability.”

The answer, says Sheldon, was to fundamentally rethink the typical approaches to solving counting problems.

“Typically, you either have humans do time-intensive and accurate hand-counts of a very small data set, or you have computer vision run less-accurate automated counts of enormous data sets,” Sheldon says. “We thought: why not do both?”

DISCount is a framework that can work with any already existing AI computer vision model. It works by using the AI to analyze the very large data sets—say, all the images taken of a particular region in a decade—to determine which particular smaller set of data a human researcher should look at. This smaller set could, for example, be all the images from a few critical days that the computer vision model has determined best show the extent of building damage in that region. The human researcher could then hand-count the damaged buildings from the much smaller set of images and the algorithm will use them to extrapolate the number of buildings affected across the entire region. Finally, DISCount will estimate how accurate the human-derived estimate is.

“DISCount works significantly better than random sampling for the tasks we considered,” says Pérez. “And part of the beauty of our framework is that it is compatible with any computer-vision model, which lets the researcher select the best AI approach for their needs. Because it also gives a confidence interval, it gives researchers the ability to make informed judgments about how good their estimates are.”

“In retrospect, we had a relatively simple idea,” says Sheldon. “But that small mental shift—that we didn’t have to choose between human and artificial intelligence, has let us build a tool that is faster, more comprehensive, and more reliable than either approach alone.”

 

Contacts: Subhransu Maji, smaji@cs.umass.edu

                 Daegan Miller, drmiller@umass.edu

 

---

JOURNAL

Proceedings of the AAAI Conference on Artificial Intelligence

DOI

10.1609/aaai.v38i20.30235 

ARTICLE TITLE

DISCount: Counting in Large Image Collections with Detector-Based Importance Sampling