BERT-BiGRU-CRF Application in Insulin Dosage Monitoring System for Control of Blood Glucose Levels in Type 2 Diabetic Patients

The number of people diagnosed with diabetes every year has skyrocketed in recent years. To avoid potentially catastrophic outcomes, it is essential to follow recommended insulin dose guidelines. In this pilot trial, patients with type 2 diabetes (T2D) were given a novel adherence detection method. An artificial Continuous Glucose Monitor (CGM) was used to model real-world glucose readings. To simulate the vast range of CGM signals experienced by T2D patients, a T2D variant of the Medtronic Virtual Patient (MVP) model for T1D was used. These data were used in a thorough grid search to compare several categorization methods employing these signals. Preprocessing, feature selection, and model training are the three main components of the proposed method. Normalization, CFS feature selection, and model training are all part of the preprocessing phase. The model is trained with BERT-BiGRU-CRF. There are primarily three levels to the model. The BERT layer is at the top of the stack. The BERT language model used for pretraining reduces the dimensionality of each word in a sentence. BiGRU is used as the second layer to automatically extract contextual information, such as semantic and temporal details. The third layer, called the CRF layer, works to resolve the interdependence of the output tags in order to determine the global optimum annotation sequence of the text. When compared against GRU and BiGRU, two established alternatives, the innovative method fares quite well.