Tan Tock Seng Hospital (TTSH) and Buzud, a biomedical technology company, have partnered to refine an artificial intelligence (AI)-powered tool aimed at helping diabetics calculate the correct insulin dosage before meals.
The innovative feature, integrated into the Buzud app, uses AI-driven image recognition. Users take a photo of their meal, which the app analyzes to identify its components. This data is then combined with readings from the patient’s continuous glucose monitor (CGM) to calculate the precise insulin dosage needed.
Currently, diabetic patients must manually estimate their insulin dosage, often relying on past meals or guesswork, particularly those newly diagnosed or adjusting to the regimen of multiple daily injections. According to Lian Xia, senior nurse clinician at TTSH’s Department of Endocrinology, this process can cause significant anxiety due to the risk of miscalculating insulin.
Patients with Type 1 diabetes, advanced Type 2 diabetes, gestational, and pancreatogenic diabetes, need to perform detailed carbohydrate counting or guess based on previous meals. To determine the correct insulin dose, they must measure the meal’s carbohydrate content, followed by a finger-prick test to assess their blood glucose levels. The amount of insulin needed is then adjusted to counter any excess glucose, a process repeated multiple times a day.
A team of 20 healthcare professionals, including dietitians, nurses, endocrinologists, and medical social workers from TTSH, began collaborating with Buzud developers in 2024 to improve the app’s accuracy. While the AI is currently trained to recognize a wide range of Western and Asian dishes, achieving an accuracy rate of 90%, the goal is to refine the tool further until it reaches 99.9% accuracy. Clinical trials and further enhancements to the app’s image recognition capabilities are expected to take place over the next two years.
However, there are some limitations to the app’s current functionality. The tool struggles with detecting small ingredients, such as tiny amounts of sauces or noodles, especially if the photo angle or lighting is unfavorable. Despite this, Buzud’s CEO, Frankie Fan, remains optimistic, stating that the company is working closely with TTSH to collect additional data, including up to 600,000 more food images, to improve accuracy.
The app, which has approximately 520,000 users across Singapore, Malaysia, and Indonesia, was launched in 2022. It allows real-time glucose monitoring via a wearable device placed on the forearm, which sends continuous glucose readings to the Buzud app. This feature alerts patients to abnormal blood sugar events, helping them manage their condition proactively.
While the glucose monitor and basic AI-powered food recognition are already available, the insulin dosage calculator is still in development, with plans for its release in 2027. TTSH’s Lian hopes that the app will empower patients to avoid dangerous fluctuations in blood sugar levels, such as hypoglycemia and hyperglycemia, potentially reducing hospital visits and emergency interventions.
For users like Eddie Tan, a 47-year-old Type 1 diabetes patient and sailing coach, the app’s insulin calculator promises to bring significant relief. Tan, who has lived with diabetes for 15 years, currently needs to check his blood sugar multiple times a day before meals and training sessions. He sees the introduction of the AI tool as a way to simplify food management and avoid the challenges of calculating insulin doses for unfamiliar meals.
“With everything consolidated on my phone, I can address any issues immediately,” Tan shared. “In the long term, this will help reduce episodes of hyperglycemia and hypoglycemia, allowing me to enjoy a wider variety of food without the constant worry of calculating insulin doses.”
The collaboration between TTSH and Buzud represents a promising step forward in diabetes care, using AI to bridge the gap between clinical treatment and everyday decision-making for patients.
Related topics:
Diabetes: A Hidden Catalyst for Antibiotic-Resistant Superbugs
Shorter Glucose Test May Revolutionize Postpartum Diabetes Screening
Breakthrough Diabetes Treatment Increases Insulin-Producing Cells by 700%