Children who eat more quickly during meals or snacks are at a higher risk of becoming obese, as found by researchers in the Penn State Department of Nutritional Sciences. However, studies examining this link are typically restricted to small-scale experiments in controlled settings, mainly due to the challenge of measuring a child's eating speed; it involves reviewing video footage of a child eating and manually noting each bite.
Subscribe to our newsletterfor the most recent science and technology news updates.
To enable the measurement of bite rate in larger studies and various settings, scientists from the Penn State Departments of Nutritional Sciences and Human Development and Family Studies worked together to create an artificial intelligence (AI) model that tracks eating frequency.
A pilot study—recently published in Frontiers in Nutrition—showed that the system is currently approximately 70% as effective as human bite counters. Although it needs further refinement, the researchers noted that the AI model has potential to assist researchers—and ultimately parents and healthcare professionals—in recognizing when children need to slow down or modify their eating habits.
Eating at a rapid pace and the likelihood of obesity
When we eat rapidly, our digestive system doesn't have enough time to register the calories," explained Kathleen Keller, a professor and Helen A. Guthrie Chair of nutritional sciences at Penn State, who co-authored this study. "The quicker you eat, the faster the food moves through your stomach, and the body isn't able to release hormones in time to signal that you're full. Later, you might feel as though you've eaten too much, but if this pattern continues, people who eat quickly are more likely to face a higher risk of becoming obese.
A quicker chewing rate, particularly when paired with bigger bites, is linked to increased obesity levels in children, according toprevious research from Keller's laboratory group. Other studieshave shown that consuming larger bites can also increase the likelihood of choking.
Often, the goal of interventions designed to slow down eating is to influence bite rate," noted Alaina Pearce, a research data management librarian at Penn State and co-author of this study. "This is due to the fact that bite rate represents a consistent aspect of children's eating habits, which can be addressed to decrease their eating speed, food consumption, and ultimately, their risk of obesity.
Assessing the frequency of bites is a time-consuming and physically demanding task, making it costly and often resulting in limited data being used in studies on bite rates, as noted by Keller, a faculty member affiliated with the Penn State Social Science Research Institute.
Using technology to maintain children's well-being
To tackle this issue, Yashaswini Bhat, a doctoral student in nutritional sciences and the study's primary author, aimed to create the first AI-based food intake counter for use in research on children's eating habits.
"I am interested in AI and data science, but I had never created a system such as this," Bhat stated.
She worked with Timothy Brick, an associate professor of human development and family studies at Penn State and a co-author of the study, to develop a system capable of recognizing children's faces in videos featuring multiple individuals and subsequently identifying each child's individual bites while they were eating.
"A seasoned and well-informed partner such as Dr. Brick proved to be essential for this project," Bhat stated.
The scientists utilized 1,440 minutes of footage from Keller'sFood and Brain StudyA research investigation into the neural processes that could contribute to overeating among children. The video recordings featured 94 children aged 7 to 9 years old eating four different meals on separate days with varying quantities of the same food items.
The scientists detected bites in 242 videos by reviewing the footage and recording every instance of a bite. They then utilized this data to teach the AI model. After the model could recognize events that seemed like bites, the researchers had it analyze 51 additional videos from the same dataset. The researchers then checked the bites found by the model against those noted by research assistants.
A successful first step
Our system proved highly effective in recognizing the children's faces," Bhat stated. "It also performed exceptionally well in identifying bites when it had a clear, unimpeded view of a child's face.
However, the system is not yet prepared for broad application, as stated by Bhat. The findings showed that the model was approximately 97% as effective as a human in recognizing a child's face in the video, but only about 70% as effective as a human in identifying each bite.
The system was less precise when a child's face wasn't fully visible to the camera or when a child chewed on their spoon or played with their food, which frequently occurs near the end of a meal," Bhat stated. "As one might expect, this kind of behavior is far more prevalent in children than in adults. Chewing on an eating tool occasionally looked like a bite, making the task more challenging for the AI model.
Although further effort is required, the researchers mentioned that this study serves as a promising initial test. With additional training, they stated that the system—known as ByteTrack—will better recognize bites and become capable of disregarding other activities, such as drinking a beverage.
The ultimate aim is to create a reliable system that operates effectively in real-world conditions," Bhat stated. "In the future, we could potentially provide a smartphone application that alerts children when they should slow down their eating, helping them establish healthy habits that endure throughout their lives.
More information:Yashaswini Rajendra Bhat and others, ByteTrack: a machine learning method for determining the number of bites and bite rate using meal videos in children,Frontiers in Nutrition (2025). DOI: 10.3389/fnut.2025.1610363
Supplied by Pennsylvania State University
This narrative was first released onMedical Xpress.