IEEE J Biomed Health Inform. 2020 May;24(5):1237-1245. doi: 10.1109/JBHI.2019.2931782. Epub 2019 Jul 29.
Hypoglycemia or low blood glucose is the most feared complication of insulin treatment of diabetes. For people with diabetes, the mismatch between the insulin therapy and the body's physiology could increase the risk of hypoglycemia. Nocturnal hypoglycemia is particularly dangerous for type-1 diabetes patients because its symptoms may obscure during sleep. The early onset detection of hypoglycemia at night time is necessary because it can result in unconsciousness and even death. This paper presents new electroencephalogram spectral features for nocturnal hypoglycemia detection. The system uses high-order spectral moments for feature extraction and Bayesian neural network for classification. From a clinical study of hypoglycemia of eight patients with type-1 diabetes at night, we find that these spectral moments of theta band and alpha band changed significantly. During hypoglycemia episodes, the theta moments increased significantly (P < 0.001) while the features of alpha band reduced significantly (P < 0.001). Using the optimal Bayesian neural network, the classification results were 85% and 52% in sensitivity and specificity, respectively. The significant correlation (P < 0.001) with real blood glucose profiles shows the effectiveness of the proposed features for the detection of nocturnal hypoglycemia.
低血糖或低血糖是糖尿病胰岛素治疗最可怕的并发症。对于糖尿病患者来说，胰岛素治疗与身体生理之间的不匹配可能会增加低血糖的风险。夜间低血糖对 1 型糖尿病患者尤其危险，因为其症状在睡眠中可能会变得模糊。夜间低血糖的早期检测是必要的，因为它可能导致无意识甚至死亡。本文提出了用于夜间低血糖检测的新脑电图频谱特征。该系统使用高阶谱矩进行特征提取和贝叶斯神经网络进行分类。通过对 8 名 1 型糖尿病患者夜间低血糖的临床研究，我们发现θ带和α带的这些频谱矩发生了显著变化。在低血糖发作期间，θ矩显著增加（P<0.001），而α带的特征显著降低（P<0.001）。使用最优贝叶斯神经网络，灵敏度和特异性的分类结果分别为 85%和 52%。与实际血糖谱的显著相关性（P<0.001）表明了所提出的特征对夜间低血糖检测的有效性。
