癲癇發作預測之演算法和系統設計

Algorithm and System Design for Epileptic Seizure Prediction

張乃夫(Nai-Fu Chang)

指導教授 ： 陳良基

國立臺灣大學/電機資訊學院/電子工程學研究所/碩士(2012年)

https://doi.org/10.6342/NTU.2012.00530

摘要

癲癇是一種慣性神經失調，影響世界約莫五千萬人。對於那些病情無法用藥物控制的病人，癲癇的發作是一種突發、無法預見的形式且被稱為是讓人最無力的疾病之一。除了有嚴重傷害的危險，一種極度的無助感時常地影響病人每天的生活，如果有方法可以預測癲癇發作將大大的改善治療的可能性並且病患的生活品質。然而，預測問題尚未被解決的就是變動問題，由其是腦波會隨著被許多複合因子像是藥物、病人狀況、癲癇成因等影響。

此論文中，我們提出一個持續訓練方法，利用最近觀察的數據，逐步擴大訓練資料，並持續訓練找出最佳的模型。我們所使用的資料庫為德國弗萊堡顱內腦電信號資料庫、CHB - MIT顱外腦電信號資料庫和台大醫院病人的顱外腦電信號。我們提出的方法在顱內及顱外的資料庫上分別可達到69.7％和54.5％的預測率，和傳統離線訓練方式相較分別提高了37.9％和22.7％。藉由頻道選擇方法可以進一步提高預測率到78.8%和77.3%並且減少電極的數目到三或四而非幾十個。分析顯示高gamma和高beta頻帶在癲癇預測問題上最為有效。

考量可行性、便利性和彈性，我們提出了一個癲癇系統，包含三個擷取顱外腦電信號的乾電極，無線連接和癲癇預測器去警告使用者癲癇的發生。內建的癲癇偵測模組免去使用者標示的困擾。此外，系統提供極大的彈性，電極的位置、癲癇預測的時間、和模型更新的時間都可以隨著病人的狀態而改變。

在附錄中，我們介紹了另一個關於支援希爾伯特黃轉換的擬態模組分解生醫微處理器。希爾伯特黃轉換的晶片解決方案將會對於非線性、非穩態信號分析在可攜式或植入式感應器有著很大的影響。而擬態模組分解則是希爾伯特黃轉換中最重要的一環。傳統上，擬態模組分解通常在取得一大段信號後執行而可能不符合實際上的應用。因此我們提出了擬態模組分解的硬體解決方案架構，藉由資料再利用的在線插補方法以及元件和重複的分解來達到低延遲和低硬體成本需求。全世界第一顆擬態模組分解晶片使用聯電90奈米低漏電製程並且消耗57.3微瓦。

關鍵字

癲癇 ； 癲癇預測

並列摘要

Epilepsy is a chronic neurological disorder that affects around 50 million people worldwide. For patients with medically intractable epilepsy, it is the sudden, unforeseen way in which seizures occur that represents one of the most disabling aspects of the disease. Apart from the risk of serious injury, there is often an intense feeling of helplessness that has a strong impact on the everyday life of a patient. A method capable of predicting the occurrence of seizures could significantly improve the therapeutic possibilities and thereby the quality of life for epilepsy patients. However, a fundamental issue in prediction problem that has not been properly resolved is variation issue, especially in light of the many confounding variables such as medications, fluctuating patient state, seizure heterogeneity, and the inherently stochastic nature of these events.

In the thesis, we proposed an online-retraining scheme to deal with variation issue. The performance of the method is evaluated on Electrocorticogram(ECoG) recording from Freiburg database as well as long-term scalp EEG recording from CHB-MIT EEG Database and National Taiwan University Hospital. The algorithm shows 69.7% sensitivity and 54.5% sensitivity in ECoG and scalp EEG databases, while improving the sensitivity of on-line training method by 37.9% and 22.7% in ECoG database and EEG database respectively. Fixed channel selection method and adaptive channel selection method are proposed to boost the sensitivity up to 78.8% and
77.3% in ECoG database as well as EEG database, and reduce the number of channels required to only three or four rather than several or tens. Band selection shows that the patterns of high gamma band and high beta band are more effective on the seizure prediction problem to classify inter-ictal and pre-ictal states.

Considering feasibility, convenience and flexibility, we proposed a seizure prediction system, consisting of 3-channel EEG dry sensors, wireless connection and seizure prediction engine, to warn the users the upcoming of seizure onset. A seizure detection module is added into the system to support automatic seizure onset marking. Besides, the system retains large flexibility for users. The location of key channels, the supposed length of prediction horizon and the suitable retraining period for online retraining scheme are all could be changed easily in accordance to patients condition.

In the appendix, we introduced another work about design and implementation of on-line Empirical Mode Decomposition (EMD) biomedical microprocessor for Hilbert-Huang transform (HHT). On-chip implementation of HHT has great impact to analyze the non-linear and non-stationary biomedical signals on wearable or implantable sensors for the real-time applications. EMD is the key component for the HHT processor. In tradition, EMD is usually performed after the collection of a large window of signals, and the long latency may not be feasible for the real-time applications. In this work, the architecture of on-line EMD for biomedical signals is proposed. The on-line interpolation method with data reuse as well as component and iteration loop decomposition is applied to obtain low latency and low hardware cost. The first chip of EMD processor is fabricated in UMC 90nm LL process and consumes 57.3 μW.

並列關鍵字

Epilepsy ； Seizure Prediction

參考文獻

[5] P. Mirowski, D. Madhavan, Y. Lecun, and R. Kuzniecky, “Classification of patterns of eeg synchronization for seizure prediction.,” Clinical Neurophysiology, vol. 120, no. 11, pp. 1927–1940, 2009.

[7] “Guidelines for epidemiologic studies on epilepsy. Commission on Epidemiology and Prognosis, International League Against Epilepsy,” Epilepsia, vol. 34, no. 4, pp. 592–6, 1993.

[8] R. S. Fisher, W. Boas, W. Blume, C. Elger, P. Genton, P. Lee, and J. Engel, “Epileptic Seizures and Epilepsy: Definitions Proposed by the International League Against Epilepsy (ILAE) and the International Bureau for Epilepsy (IBE),” Epilepsia, vol. 46, pp. 470–472, Apr. 2005.

[10] R. S. Fisher, B. G. Vickrey, P. Gibson, B. Hermann, P. Penovich, A. Scherer, and S. Walker, “The impact of epilepsy from the patient’s perspective I. Descriptions and subjective perceptions,” Epilepsy Res., vol. 41, pp. 39–51, Aug 2000.

[11] Seizure Prediction in Epilepsy: From Basic Mechanisms to Clinical Applications. Wiley-VCH; 1 edition (October 21, 2008).