Scholarlink

Diabetes is a chronic condition that affects nearly every household, presenting significant difficulties for both developing and developed nations. This study introduces a system that calculates blood glucose levels (BGL) through a non-invasive approach utilizing Photoplethysmography (PPG) signals. Prior research has demonstrated the measurement of blood glucose levels (BGL) through optical sensors, albeit with reduced precision. A data acquisition system utilizing optical sensors is constructed to record the PPG signal of the participants. The primary focus of this study lies in investigating different characteristics of a PPG signal through the Single Pulse Analysis (SPA) method to accurately estimate fasting blood glucose levels. Data from 151 participants was collected over a period of 3 min each to assess blood glucose levels using SPA. The required signal processing and signal conditioning techniques are done for feature extraction. Deep learning (DL) models trained on the proposed feature set achieved 92.06% accuracy in BGL prediction. The proposed features in SPA resulted in significant improvements in accuracy and Clarke Error grid analysis, with 86.37% of data samples in class A and 13.63% in class B. The SPA technique with the proposed feature set is a suitable option for implementing a non-invasive glucose measurement system.

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

Photoplethysmography Based Non-invasive Continuous Blood Glucose Monitoring with Improved Feature Selection and Deep Learning Techniques

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Elsa Sharu Johnson,
  • G. Uma,
  • Khaviya Venkatraj,
  • Midhuram Manoj,
  • H. Sandhya

摘要

Diabetes is a chronic condition that affects nearly every household, presenting significant difficulties for both developing and developed nations. This study introduces a system that calculates blood glucose levels (BGL) through a non-invasive approach utilizing Photoplethysmography (PPG) signals. Prior research has demonstrated the measurement of blood glucose levels (BGL) through optical sensors, albeit with reduced precision. A data acquisition system utilizing optical sensors is constructed to record the PPG signal of the participants. The primary focus of this study lies in investigating different characteristics of a PPG signal through the Single Pulse Analysis (SPA) method to accurately estimate fasting blood glucose levels. Data from 151 participants was collected over a period of 3 min each to assess blood glucose levels using SPA. The required signal processing and signal conditioning techniques are done for feature extraction. Deep learning (DL) models trained on the proposed feature set achieved 92.06% accuracy in BGL prediction. The proposed features in SPA resulted in significant improvements in accuracy and Clarke Error grid analysis, with 86.37% of data samples in class A and 13.63% in class B. The SPA technique with the proposed feature set is a suitable option for implementing a non-invasive glucose measurement system.