Introduction: MiniMed 770G is a new insulin pump equipped with Hybrid Closed Loop (HCL). Its clinical impact on the actual clinical user in Japanese adolescents is still unclear.

Material and methods: We compared glucose profiles, including HbA1c, Time in Range (TIR), and insulin dose in patients with established type 1 diabetes mellitus who had already used Continuous Glucose Monitoring before the HCL introduction.

Results: Results: Thirty-seven participants were included in the analysis. The median age was 15 years (Median values are used below). Before and three months after HCL introduction, HbA1c significantly decreased from 7.51% to 7.08%, and TIR significantly increased from 60.4% to 69.6%, respectively. Basal insulin as a percentage of total daily insulin dose (Basal%) showed an increasing trend from 32.3% to 37.0%. The stratified analysis revealed that the group with increased Basal% had a significantly lower Basal% before HCL introduction than the group with decreased Basal%. Younger age group showed a significant insulin sensitivity discrepancy of more than 100 between the calculated and the default setting before HCL introduction.

Conclusion: HbA1c and TIR improved after HCL introduction, and the basal% showed an increasing trend. Insulin sensitivity of some younger age group deviated after the HCL introduction, suggesting a risk of hypoglycemia due to HCL.

Aim: Knowing the nutrients contained in various foods is important for patients who need medical nutrition therapy such as those with diabetes mellitus. We conducted a pilot study of the meals provided by hospital food services to assess the accuracy of a prototype “AI registered dietician” app by building a database of meals with known nutritional contents and analyzing meals at hospitals.

Material and methods: The app was developed using Python. The images of the meals were captured by smart phone by taking a movie. A cube of 1 cm3 was used as the size marker. The images of the meals of one cycle (28 days) were captured at Hospital A, and multiple still images were extracted from the movie and to be used as teacher images for the AI. The amount of rice was calculated using Open CV technology. The accuracy of object detection and the quantification was tested by images captured at hospitals B and C.

Results: The precision and recall of rice were 100% and 98.3%, respectively. The precision and recall of entire meals were 42.0% and 26.9%, respectively. Items that were not included in teacher images were detected as other items or were not detected. The amount of a 200 g serving of rice was estimated as 752 g (95% confidence interval: 482-1022 g, P < 0.001) at Hospital B and as 130 g (95% confidence interval: 96-143 g, P = 0.982) at Hospital C.

Conclusion: The object detection of rice demonstrated high precision and recall; however, further improvement of accuracy in the estimation of the quantity is necessary.