부산대학교 도서관

The application of clustering techniques to glucose time series data has the potential to enhance early detection and effective management of diabetes and other metabolic disorders. This, in turn, can contribute to improved clinical outcomes and patient-reported outcomes. In this study, we aim to address the issue of personalized clustering of glucose time series data, specifically those recorded during the nocturnal sleep period, via Self Organized Maps. The clustering model was constructed using a dataset comprising 22 patients diagnosed with type 1 diabetes who were closely monitored for a maximum duration of 4 weeks as part of the GlucoseML study. Based on the silhouette coefficient score, two distinct clusters have been identified, indicating the presence of two separate glucose distributions observed during nocturnal sleep. Within Cluster 1, it has been observed that 90% of glucose values fall within the desired euglycemic range of 70-180 mg/dL. Consequently, there is a corresponding increase in the likelihood of hypoglycemia, with a rate of 2.4% occurring below the threshold of 70 mg/dL. Within Cluster 2, patients show a mainly hyperglycemic profile, characterized by 64.4% of values exceeding 180 mg/dL, while no instances of hypoglycemia were observed.