使用心律變異特徵預測院外心跳停止病人癲癇發作深度學習模型

Abstract

院外心跳停止的病患在恢復心跳自主循環後，大多數會出現心跳後停止症候群的症狀，其中以腦損傷引起的癲癇發作最為嚴重，常導致不良的治療結果。為了提高治療效果，需要預測癲癇的發生，以便及時施打抗癲癇藥物。本研究旨在讓病患在加護病房中能夠準確預測癲癇，使用心律變異特徵（Heart Rate Variability, HRV），並開發基於雙向長短期記憶網絡（Bi-LSTM）和領域對抗式學習神經網絡（DANN）的深度學習模型來進行預測。

本研究採用個體獨立型的模型訓練方式，使模型能夠快速預測從未見過的病患，並解決HRV特徵中的個體差異問題。研究結果顯示，模型在測試資料集中的最高精確度可達60%，能夠在癲癇發作前12.5到32.5分鐘進行預測，為醫護人員提供了充足的準備時間。

Patients who experience out-of-hospital cardiac arrest (OHCA) and subsequently regain spontaneous circulation often develop post-cardiac arrest syndrome. Among the various complications, brain injury leading to seizures can result in poor treatment outcomes. To improve these outcomes, it is essential to predict seizures promptly so that antiepileptic medications can be administered in a timely manner. This study aims to accurately predict seizures for patients in intensive care units (ICU) using heart rate variability (HRV) features and developing a deep learning model based on Bi-directional Long Short-Term Memory (Bi-LSTM) and Domain-Adversarial Neural Network (DANN).

This study employs an individual-independent model training approach, allowing the model to quickly predict seizures in previously unseen patients while addressing individual differences in HRV features. The results show that the model achieves a maximum accuracy of 60% on the test dataset and can predict seizures 12.5 to 32.5 minutes before onset, providing sufficient preparation time for healthcare professionals.