Arya News - Using an AI-trained camera feature in the app, a patient would snap a picture of his food. The app would then identify meal components in the picture and use data from the patient’s wearable continuous glucose monitor (CGM) to perform the calculations.
SINGAPORE – Tan Tock Seng Hospital (TTSH) and biomedical technology company Buzud are working on an AI or artificial intelligence-enabled tool in an app to help diabetes patients accurately calculate their correct insulin dosage before a meal.
Using an AI-trained camera feature in the app, a patient would snap a picture of his food. The app would then identify meal components in the picture and use data from the patient’s wearable continuous glucose monitor (CGM) to perform the calculations.
Most diabetic patients today estimate the amount of insulin they should jab before their meals, said senior nurse clinician Lian Xia from TTSH’s Department of Endocrinology.
“This has been the way for a long time because there is no simpler way to do it,” said Ms Lian, who is spearheading the project on the hospital’s end.
She shared that patients of the endocrine clinic often experience anxiety over miscalculation of insulin doses, especially among those newly diagnosed or adjusting to their need for multiple daily injections.
Patients with Type 1 diabetes, advanced Type 2 diabetes, gestational and pancreatogenic diabetes are required to either perform calculations before their meals to know the precise amount of insulin to self-administer before eating, or make a guess based on meals they previously had.
Insulin dose calculation involves a measurement of all the carbohydrates in the meal before performing a finger-prick test to determine the patient’s excess blood glucose levels.
Patients would then need to determine how many additional units of insulin are needed to counter any excess glucose and sum up the total amount of insulin required.
This test has to be done multiple times a day.
In 2024, a team of 20 dieticians, nurses, endocrinologists and medical social workers from TTSH started working with the developers of the app, which is also called Buzud, to refine the accuracy of the app’s image recognition.
The calculation tool will be launched only when it achieves 99.9 per cent accuracy.
Over the next two years, the meal recognition camera will undergo refinement and clinical trials as the two partners further train the accuracy of the AI to attain that level of accuracy.
Trained on Western and a large number of Asian dishes, it currently has an accuracy of about 90 per cent, according to Mr Frankie Fan, chief executive of Buzud.
Its current limitations include the inability to detect some sauces and condiments, as well as ingredients that are small in size or quantity.
“The camera could fail to pick up small amounts of noodles, for example, or mistake it for something else, because of its small quantity, or the angle at which the photo was taken. Factors like lighting affect the outcome, too,” Mr Fan said.
TTSH nurses will continue feeding Buzud information on the nutritional needs of diabetic patients and up to 600,000 more images of foods captured through clinical trials, he added.
Ms Lian said: “By contributing to the development of an intuitive, AI-enhanced tool, we hope to bridge the gap between clinical care and real-time decision-making at mealtimes.”
No changes will be made to the glucose monitor, which can be self-fitted by the patient and has a lifespan of 15 days.
The monitor – a device that is attached to the inside of the forearm – contains a fine plastic filament that penetrates the skin in order to come into contact with the bloodstream. This filament continuously takes blood glucose levels which are relayed live, every minute, via Bluetooth to the Buzud app on the patient’s mobile phone. Patients are also alerted to any abnormal glucose events.
The health management app was launched in 2022, and currently has about 520,000 users across Singapore, Malaysia and Indonesia.
While the monitor and a basic version of the AI-enabled food recognition are currently available, the insulin calculator is expected to be launched in 2027.
Ms Lian said: “We hope this digital solution will empower the patients to reduce the risk of hypoglycaemia (low blood sugar) and hyperglycaemia (high blood sugar). It might potentially reduce emergency visits and hospital admissions, too.”
Mr Eddie Tan, 47, is a Type 1 diabetes patient of 15 years. He said that his active lifestyle as a sailing coach requires him to prick his fingers 10 times or more a day – before and after training and meals, and before bed.
Buzud app and CGM-user Eddie Tan says the introduction of the insulin calculator would enable him to “get more adventurous” with food. PHOTO: THE STRAITS TIMES
Despite having quite a healthy appetite for food, Mr Tan said he has avoided trying new foods, opting instead for familiar food choices to avoid having to go through the ordeal of pin pricks and calculations.
The Buzud app and CGM-user of one year said that the introduction of the insulin calculator would enable him to “get more adventurous” with food, as he could widen his repertoire of food without having to worry about calculating the correct amount of insulin before eating.
“It is nice to know that with everything consolidated on my phone, and (updates) provided in real time, I can rectify things immediately and accurately in times of episodes.
“In the long term, this will be extremely helpful in reducing episodes of hyperglycaemia and hypoglycaemia,” Mr Tan said.