Excessive sodium intake is widely recognized for increasing the risk of cardiovascular diseases such as hypertension and myocardial infarction. As well as triggering chronic diseases across the body, such as kidney failure, stomach cancer, and osteoporosis.
South Korea’s average daily sodium intake is about 1.6 times the recommended level. However, individuals often find it challenging to monitor their daily salt consumption accurately.
On the 18th, professors Ryu Ji Won, Kim Hye Won, and Kim Se Jung from the Department of Nephrology at Seoul National University Bundang Hospital unveiled their research. Their study focused on the effectiveness of artificial intelligence (AI) in estimating salt intake from pictures of food.
Traditionally, the 24-hour urinary sodium test is considered the most accurate method for measuring sodium intake, especially for patients with conditions like kidney disease that require sodium restriction. However, this method is cumbersome, involving multiple daily urine samples. Therefore, a simpler and more applicable method in daily life is needed. The research team turned their attention to rapidly advancing AI and conducted a study to verify the usefulness of technology that estimates sodium intake from food photos alone.
The AI used in the research is a model that uses the YOLO (You Only Look Once) v4 architecture to detect food areas, MST++, ResNet-101 artificial neural network models to classify food types, and hyperspectral imaging technology to measure food volume. By comparing before and after photos of meals, the AI calculates the difference in salt content.
The research team at Seoul National University Bundang Hospital, led by professors from the Department of Nephrology, utilized AI to calculate sodium intake based on patients’ before and after meal photos. They compared these AI-generated estimates with the results from 24-hour urinary sodium tests. The findings confirmed that when considering variables such as gender, age, kidney function, and use of diuretics, the AI’s analysis could closely match the sodium test results. Furthermore, the team successfully developed a formula that predicts actual 24-hour urinary sodium test results using only the estimated glomerular filtration rate (eGFR) and the AI-measured sodium intake.
This study highlights the potential of a more convenient AI-based method to measure sodium intake for hospital inpatients, with promising applications in clinical settings and daily life through future advancements.
Ryu explained, “Taking before-and-after-meal photos with a smartphone application is much more convenient than self-assessment records or surveys. The estimated glomerular filtration rate can predict 24-hour urinary sodium levels, indicating high potential for inpatient use.”
Kim said, “High salt intake can increase blood pressure, damaging the kidneys’ glomeruli and surrounding blood vessels, potentially worsening hypertension and creating a vicious cycle. Therefore, management in daily life is important, and AI sodium measurement technology can be a good tool.”
 in estimating salt intake from pictures of food.){kind=link}