Secure IoMT smartwatch-based blood glucose monitoring using multimodal activity and nutrition data with transfer learning.

In the modern era, healthcare faces critical challenges as individuals often consume unbalanced diets and neglect physical activity. A primary concern is elevated blood glucose levels, which commonly result from high carbohydrate intake and a sedentary lifestyle. To address this, the paper proposes a novel system: Enhanced Body Sugar Monitoring-Secure Smartwatches Leveraging IoMT for Activity and Nutrition Execution Based on Transfer Learning. The system collects multimodal data, including subject, nutrition, and activity records, to predict and display blood sugar levels under varying dietary and activity conditions using open-world multimodal datasets. The presented smartwatch-enabled framework is equipped with various Internet of Medical Things (IoMT) sensors, including heart rate, blood pressure, oxygen saturation, and more. These sensors are the inputs to different tasks that have collected data from them and offloaded execution to remote services. At the same time, the TL-DCNNOS algorithm processes the entire workflow through separate pipelines, such as data collection and encryption, and offloads task data to nearby edge nodes for secure execution. For real-time learning and training on sensor data while executing tasks across different nodes, we employ transfer learning and DCNNs to learn patterns of behavior such as eating, sitting, walking, and more to identify normal and abnormal behavior. We used the open-world IoMT dataset to train the initial model, and then trained and classified at runtime during real-time experiments on the testbed with different subjects. Simulation results show that we minimized time consumption and security risk and improved sugar prediction accuracy to 99% with various runtime activities, compared with existing studies.