Final Report Summary - GOCARB (Type 1 Diabetes Self-Management and Carbohydrate Counting: A Computer Vision based Approach)
GoCARB is a four year transatlantic research project funded by the EC. The consortium members are working in close collaboration towards enhanced and individualized diabetes treatment. The primary scope of GoCARB is to support diabetes patients in counting carbohydrates and estimating the appropriate size of their prandial insulin dose. Although studies have shown that planning meals and counting carbohydrates is of great importance for diabetic patients, even well trained diabetic patients find it difficult to calculate precisely the carbohydrates of their meals.
GoCARB aims to design, develop and validate a system which permits the automatic, near real-time recognition of the different types of foods on a plate and the estimation of their carbohydrate content. This information is used to optimize the calculation of the prandial insulin dose. In a typical scenario, the user places a reference card next to the meal and acquires two images using a smart phone’s camera. The first image is acquired above the plate at a distance of 30-40 cm, while the second is at about 15 degrees from the vertical axis crossing the centre of the dish. A graphical user interface guides the user to choose the optimal angles for image acquisition, using the smartphone’s built-in sensors (accelerometer and gravity sensor). Then, the different food items on the plate are segmented and recognized while their 3D shape is reconstructed. Based on the shape, the segmentation results and the reference object, the volume of each item is estimated. The carbohydrate content is calculated by combining the food type with its volume, and using nutritional databases. Finally, this information is fed to the bolus advisor and the optimal prandial insulin dose is estimated. (Figure 1 - GoCARB pipeline.png)
The prototype is designed to work for elliptical plates, single-dish images and fully visible food items. In order to validate the integrated prototype a three-step procedure was followed involving technical, preclinical and clinical validation. In the technical validation, the performance of the system was assessed in terms of absolute and absolute percentage error, under laboratory setup and after satisfying all the system’s assumptions. The results were very promising but additional validation was needed under more realistic conditions. The preclinical was the first evaluation study that involved type 1 diabetic (T1D) patients and was conducted in the Bern University Hospital, “Inselspital”. The nineteen (19) volunteers who participated were asked to estimate the carbohydrate content of 6 meals, first by themselves and then by using the GoCARB app. This way, the prototype was evaluated on a less constrained environment and simultaneously compared to the estimation of the patients. According to the results, the average error of the GoCARB system was less than half of the average participants’ error. Finally, a pilot clinical study was designed in which adult volunteers with T1D on sensor-augmented insulin pump therapy, were able to take GoCARB home and use it in their daily life for one month each. The study is on-going and it is expected to be completed before the end of 2015.
In summary, GoCARB addresses the need of people with diabetes for a more effective, automated and precise way to estimate the grams of carbohydrates in food, as well as improving their management of their disease, and enhancing their quality of life. In order to meet S&T objectives of GoCARB a number of secondments and recruitments took place, involving a total of four early stage researchers and seven experienced researchers (two ER<10 years and five ER>10 years).
Website: www.gocarb.eu
Contact: Stavroula Mougiakakou (stavroula.mougiakakou@artorg.unibe.ch) - project coordinator
GoCARB aims to design, develop and validate a system which permits the automatic, near real-time recognition of the different types of foods on a plate and the estimation of their carbohydrate content. This information is used to optimize the calculation of the prandial insulin dose. In a typical scenario, the user places a reference card next to the meal and acquires two images using a smart phone’s camera. The first image is acquired above the plate at a distance of 30-40 cm, while the second is at about 15 degrees from the vertical axis crossing the centre of the dish. A graphical user interface guides the user to choose the optimal angles for image acquisition, using the smartphone’s built-in sensors (accelerometer and gravity sensor). Then, the different food items on the plate are segmented and recognized while their 3D shape is reconstructed. Based on the shape, the segmentation results and the reference object, the volume of each item is estimated. The carbohydrate content is calculated by combining the food type with its volume, and using nutritional databases. Finally, this information is fed to the bolus advisor and the optimal prandial insulin dose is estimated. (Figure 1 - GoCARB pipeline.png)
The prototype is designed to work for elliptical plates, single-dish images and fully visible food items. In order to validate the integrated prototype a three-step procedure was followed involving technical, preclinical and clinical validation. In the technical validation, the performance of the system was assessed in terms of absolute and absolute percentage error, under laboratory setup and after satisfying all the system’s assumptions. The results were very promising but additional validation was needed under more realistic conditions. The preclinical was the first evaluation study that involved type 1 diabetic (T1D) patients and was conducted in the Bern University Hospital, “Inselspital”. The nineteen (19) volunteers who participated were asked to estimate the carbohydrate content of 6 meals, first by themselves and then by using the GoCARB app. This way, the prototype was evaluated on a less constrained environment and simultaneously compared to the estimation of the patients. According to the results, the average error of the GoCARB system was less than half of the average participants’ error. Finally, a pilot clinical study was designed in which adult volunteers with T1D on sensor-augmented insulin pump therapy, were able to take GoCARB home and use it in their daily life for one month each. The study is on-going and it is expected to be completed before the end of 2015.
In summary, GoCARB addresses the need of people with diabetes for a more effective, automated and precise way to estimate the grams of carbohydrates in food, as well as improving their management of their disease, and enhancing their quality of life. In order to meet S&T objectives of GoCARB a number of secondments and recruitments took place, involving a total of four early stage researchers and seven experienced researchers (two ER<10 years and five ER>10 years).
Website: www.gocarb.eu
Contact: Stavroula Mougiakakou (stavroula.mougiakakou@artorg.unibe.ch) - project coordinator