Scholarlink

<p>Hypoglycemia is a serious and life-threatening complication for individuals with Type 1 Diabetes (T1D), occurring when blood glucose (BG) levels drop below 70&#xa0;mg/dL. If untreated, it can result in cognitive dysfunction, seizures, unconsciousness, or death. The aim of this study is to present an innovative framework for hypoglycemia prediction utilizing deep learning (DL) techniques. The proposed work integrates data from Continuous Glucose Monitoring (CGM) systems, insulin records, carbohydrate intake, and additional physiological parameters collected via a smartwatch to develop predictive models. Three neural network architectures—Long Short-Term Memory (LSTM), Bidirectional LSTM (BiLSTM), and Convolutional LSTM (ConvLSTM)—were implemented and evaluated. The Ohio dataset was employed for model training and validation, while real-time validation incorporated CGM data from the FreeStyle Libre (FSL) system and vital signs captured by the TicWatch Pro 3 Ultra smartwatch. The ConvLSTM model demonstrated the best performance at a 15-minute prediction horizon, achieving sensitivity, specificity, and accuracy of 98.06%, 94.17%, and 96.12%, respectively. The optimized ConvLSTM model was integrated into a mobile application that offers real-time BG monitoring and alerts. The application tracks glucose levels, detects critical thresholds, and notifies users of hypoglycemia events, enabling timely interventions. By combining robust predictive capabilities with real-time application, this framework significantly reduces hypoglycemia risks, enhancing safety and quality of life for individuals with T1D.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Mobile-based smart wearable intelligent Real-time hypoglycemia prediction for type 1 diabetes

  • 期刊论文

  • 发表日期：

    2026年3月14日

查看全文
  • Nevien A. Mahdy,
  • Mai S. Mabrouk,
  • Wael A. Mohamed,
  • Ahmed F. Elnokrashy

摘要

Hypoglycemia is a serious and life-threatening complication for individuals with Type 1 Diabetes (T1D), occurring when blood glucose (BG) levels drop below 70 mg/dL. If untreated, it can result in cognitive dysfunction, seizures, unconsciousness, or death. The aim of this study is to present an innovative framework for hypoglycemia prediction utilizing deep learning (DL) techniques. The proposed work integrates data from Continuous Glucose Monitoring (CGM) systems, insulin records, carbohydrate intake, and additional physiological parameters collected via a smartwatch to develop predictive models. Three neural network architectures—Long Short-Term Memory (LSTM), Bidirectional LSTM (BiLSTM), and Convolutional LSTM (ConvLSTM)—were implemented and evaluated. The Ohio dataset was employed for model training and validation, while real-time validation incorporated CGM data from the FreeStyle Libre (FSL) system and vital signs captured by the TicWatch Pro 3 Ultra smartwatch. The ConvLSTM model demonstrated the best performance at a 15-minute prediction horizon, achieving sensitivity, specificity, and accuracy of 98.06%, 94.17%, and 96.12%, respectively. The optimized ConvLSTM model was integrated into a mobile application that offers real-time BG monitoring and alerts. The application tracks glucose levels, detects critical thresholds, and notifies users of hypoglycemia events, enabling timely interventions. By combining robust predictive capabilities with real-time application, this framework significantly reduces hypoglycemia risks, enhancing safety and quality of life for individuals with T1D.