Gradual Fine-Tuning for accurate Blood Glucose Level Prediction

Fontanellaz, Matthias Andreas; Jankovic, Marko; Mougiakakou, Stavroula (March 2021). Gradual Fine-Tuning for accurate Blood Glucose Level Prediction (Unpublished). In: 14th International Conference on Advanced Technologies & Treatments for Diabetes.

Official URL:

Background and Aims
For individuals with Type 1 diabetes (T1D) is of eminence importance to avoid hypo- and hyperglycemic events. The availability of long glucose time-series along with powerful AI methods allowed the development of glucose prediction algorithms. Nonetheless open issues remain such as prediction time-delays, amount of history needed, and how heterogeneous and sparse diabetes information affect the performance.
In this study, we utilized data from 100 individuals with T1D provided by the Juvenile Diabetes Research Foundation. The dataset provides pump settings, sensor outputs (e.g. insulin-rates, continuous glucose monitoring- CGM) and conceptual information such as age, years of diabetes. To mitigate the adverse impact of large inter-patient variability, we propose a training scheme based on gradual fine-tuning. Initially, the novel AI-model is trained on all data and subsequently fine-tuned over groups with shared characteristics to individual patient-level. The individuals with T1D are assigned to groups based on similarity measures defined using glucose variability indices. For each individual, an ensemble of five dedicated sequence-to-sequence LSTM networks is used. The ensemble uses CGM data, bolus dose and meal intake as input and outputs blood glucose predictions 30 min ahead in time.
As shown in Table 1 the root-mean-square-error (RMSE), mean-average-error (MAE), and time-lag used as performance measures for the various training schemes.

Item Type:

Conference or Workshop Item (Poster)


10 Strategic Research Centers > ARTORG Center for Biomedical Engineering Research > ARTORG Center - AI in Health and Nutrition
10 Strategic Research Centers > ARTORG Center for Biomedical Engineering Research
04 Faculty of Medicine > Department of Intensive Care, Emergency Medicine and Anaesthesiology (DINA) > University Emergency Center

Graduate School:

Graduate School for Cellular and Biomedical Sciences (GCB)

UniBE Contributor:

Fontanellaz, Matthias Andreas; Jankovic, Marko and Mougiakakou, Stavroula


600 Technology > 610 Medicine & health
600 Technology > 620 Engineering




Stavroula Mougiakakou

Date Deposited:

20 Apr 2021 12:08

Last Modified:

20 Apr 2021 12:08


Actions (login required)

Edit item Edit item
Provide Feedback