DESIGN AND VALIDATION OF THE VOICEDRIVEN EXPERT SYSTEM SUPPORTING PRANDIAL INSULIN DOSE CALCULATION IN PERSONS WITH DIABETES
P. Ladyzynski1, P. Foltynski1, E. Pankowska2, K. Mazurczak2, J. Krzymien3
1Nalecz Institute of Biocybernetics and Biomedical Engineering, Polish Academy of Sciences, Poland;
2The Institute of Diabetology in Warsaw, Poland;
3Department of Internal Diseases and Diabetology, Medical University of Warsaw, Warsaw, Poland
We developed the voice-driven client-server expert system (VoiceDiab) supporting prandial insulin dose calculation in persons with diabetes. The VoiceDiab consists of an Android controlled smartphone with the client application communicating with a set of three servers responsible for: (1) automatic speech recognition and transformation of the verbal description of meals into text, (2) analysis of the textual description of the meal to determine its composition, (3) calculation of the insulin dose compensating the meal.
The efficacy of the VoiceDiab system was validated on a group of 44 patients (30 female and 14 male) aged 17.3 ± 10.2 years with type 1 diabetes in the cross-over study. The study group was appropriately educated regarding the insulin therapy. Each patient estimated prandial insulin doses by herself / himself in one study period (Control), while in the other study period insulin doses were calculated by the VoiceDiab system (VoiceDiab). An order of the study periods (Control vs. VoiceDiab) was randomly assigned. Blood glucose was monitored continuously during both periods. The mean blood glucose (MBGtot), mean postprandial blood glucose (MBGpp), mean standard deviation of blood glucose (MSDBG) and mean time of glycemia in the range 70-180 mg/dl (%BGnorm) were analyzed.
Values of all the assessed parameters indicated slightly better metabolic control during the VoiceDiab period (Control vs. VoiceDiab): MBGtot – 164 ± 31 mg/ dl vs. 163 ± 30 mg/dl, MBGpp – 169 ± 35 mg/dl vs. 167 ± 30 mg/dl, MSDBG – 74 ± 15 mg/dl vs. 70 ± 15 mg/dl and %BGnorm – 54.9% ± 13.6% vs. 55.65 ± 13.8%.
The obtained results suggest that the outcome of the intensive insulin treatment in properly educated patients with diabetes is similar when prandial insulin doses are estimated by patients and when they are calculated by the VoiceDiab system based on the voice description of meals.