Association Handi'Chiens, 13 Rue de l'Abbé Groult, Paris, France.
Univ Rennes, Normandie Univ, CNRS, EthoS (Éthologie animale et humaine), UMR 6552, 35000, Rennes, France.
Sci Rep. 2020 Oct 27;10(1):18365. doi: 10.1038/s41598-020-75478-8.
Although epilepsy is considered a public health issue, the burden imposed by the unpredictability of seizures is mainly borne by the patients. Predicting seizures based on electroencephalography has had mixed success, and the idiosyncratic character of epilepsy makes a single method of detection or prediction for all patients almost impossible. To address this problem, we demonstrate herein that epileptic seizures can not only be detected by global chemometric analysis of data from selected ion flow tube mass spectrometry but also that a simple mathematical model makes it possible to predict these seizures (by up to 4 h 37 min in advance with 92% and 75% of samples correctly classified in training and leave-one-out-cross-validation, respectively). These findings should stimulate the development of non-invasive applications (e.g., electronic nose) for different types of epilepsy and thereby decrease of the unpredictability of epileptic seizures.
虽然癫痫被认为是一个公共卫生问题，但发作的不可预测性带来的负担主要由患者承担。基于脑电图预测癫痫发作的效果喜忧参半，而且癫痫的个体特征使得为所有患者采用单一的检测或预测方法几乎不可能。为了解决这个问题，我们在此证明，癫痫发作不仅可以通过对选定的离子流管质谱数据进行全局化学计量分析来检测，而且还可以通过一个简单的数学模型来预测这些癫痫发作（在训练中提前 4 小时 37 分钟，分别有 92%和 75%的样本正确分类，在留一法交叉验证中）。这些发现应该会刺激针对不同类型癫痫的非侵入性应用（例如电子鼻）的发展，从而降低癫痫发作的不可预测性。
