An Algorithm Based on Voice Description of Meal for Insulin Dose Calculation to Compensate Food Intake

Diabetes may lead to serious complications when blood glucose levels are not maintained in save ranges. Proper insulin dosing is crucial in good metabolic control, however it is not easy for all patients. There are some computer applications supporting insulin bolus calculation based on the meal composition, but they may be difficult to use for some patients. Our team decided to developed a voice-driven system helping patients in calculation of the proper insulin dose for food intake compensation. The system consists of several software applications (on smartphone and MS Windows server) working with developed language dictionaries and with nutrient database. The speech recognition software was trained with sets of words from meal descriptions. The algorithm for transition from the text description is based on language dictionaries with variety of food item descriptions related to one product in nutrient database. The system calculates the number of insulin units for patients using insulin pen or delivery time and number of insulin units for patients with insulin pump. The algorithm for insulin dosage works not only on the basis of the amount of carbohydrates, but also takes into consideration protein and fat content in the meal. The system was successfully tested on the group of 34 subjects achieving 92.3% of correctly recognized food descriptions in the first try. The overall percentage of fails in recognition was 1.5%.