A new tool has become available for people wanting to improve their nutrition in general and avoiding diabetes specifically. By scanning their meals with an AI-based smartphone app named CarboAI, users will receive a fast estimate of the content of carbohydrates and other relevant nutrients.
The app was developed in a research project led by Dr. Alice Ojwang, Nutrition and Dietetic Faculty at the Technical University of Kenya (TUK). KENET, the national research and education network (NREN) of Kenya, has contributed with the necessary computing resources.
As indicated by the title “AI in Nutrition: Artificial Intelligence Photographic Application to support carbohydrate management in persons with Type 2 diabetes”, the app was developed with diabetes prevention as the primary goal. Still, general nutritional purposes may also be served.
Taking control of personal nutrition
For millions of people managing medical conditions or striving for healthier lifestyles, decoding the right amount of nutrients in everyday foods has long been a challenge, often relying on guesswork, labels, or expert knowledge that most people simply don’t have at their fingertips.
CarboAI bridges the gap between science and daily living, empowering individuals to take control of their nutrition, one scan at a time.
“People cook and eat, but they eat the wrong portions at the wrong time and that affects their health in the long run. We want to support people to understand their foods, the portions they should take, and how they need to eat to have better health. Most importantly is to improve diabetes outcomes across Africa. We know that within our populations, because of poorly controlled diabetes, people end up getting complications when they are very young,” says Alice Ojwang.
Translated into sugar teaspoon equivalents
Dr. Ojwang and her team started by collecting data, which involved cooking, measuring and weighing up to 6,000 Kenyan foods. They then used food composition tables to calculate the carbohydrate content in each one of them. Images of the food were subsequently stored in the KENET cloud, annotated and used to train the AI model.
In addition to estimating carbohydrate content and the corresponding sugar levels, the app provides personalized recommendations on how individuals can adjust their diet, considering their weight and any existing medical conditions.
“The carbohydrate content has been classified and estimated as teaspoons of sugar so that the person using understands how different amounts of carbohydrates impact their blood sugar. So, for example, when you scan your food with the application, it tells you you’re taking 100 grammes of carbohydrates equivalent to 10 teaspoons of sugar,” Dr. Ojwang explains.
High-performance computing
The project was made possible through a $60,000 Amazon Web Services (AWS) Health Equity Grant, awarded in cloud credits after KENET reviewed and submitted the project proposal. Since the project required a server with GPU, the team benefited from KENET’s GPU Cluster for high-performance computing, enabling faster data processing, efficient model training, and improved accuracy in recognizing foods and analyzing their nutritional content.
Looking ahead, Dr. Ojwang is determined to expand the app’s food database to include a wider variety of foods from around the world, along with integrated modules for nutrition education. She also envisions the app playing a key role in shaping policies and guiding public health interventions on diabetes awareness and management.
To enable global adoption, she plans to have the app translated from English into widely spoken languages, beginning with East African languages, followed by West African languages, and gradually expanding to other regions.
Source: the article “How AI-Based App is Set to Revolutionize Healthcare in Africa and Beyond” at the KENET website.
Contributor
