Applying Multichannel Scalp Map to Auxilary Epilepsy Decision

瀏覽統計 Email 通知 RSS Feed

簡歷

基本資料

Project title

Code/計畫編號

NSC101-2221-E019-068

Translated Name/計畫中文名

應用多通道腦像圖於輔助癲癇診斷

Project Coordinator/計畫主持人

Jung-Hua Wang

Funding Organization/主管機關

National Science and Technology Council

Department/Unit

Department of Electrical Engineering

Website

https://www.grb.gov.tw/search/planDetail?id=2635948

Year

2012

Start date/計畫起

01-08-2012

Expected Completion/計畫迄

31-07-2013

Bugetid/研究經費

676千元

ResearchField/研究領域

資訊科學--軟體
醫學工程

本計畫擬透過對單一通道之癲癇波驗證，其後透過發掘多通道關聯性的資訊，期設計一套輔助醫師於臨床判讀之工具。從執行前期國科會計畫的經驗，吾人觀察到EEG訊號其實有很比例並非屬於高度明顯性典型放電樣波，例如: 醫師觀察五位病患腦電波時，僅能辨認出其中一人的腦電圖棘波訊號具有典型放電樣波，其餘四位病患並非屬於典型放電樣波，醫師無法以肉眼清楚辨識，這顯示出即便是專業醫師也會面臨典型放電樣波資料不足的問題，故如何運用資訊工程技術以發掘非典型放電樣波的資訊，是一項值得研究的課題，況且只憑單通道並未能確認癲癇腦波。吾人嘗試藉由一具處理時序型(spatiotemporal)資料能力的遞迴式類神經網路(Simple Recurrent Neural Network, SRNN)，搭配一基於頻譜資訊之訓練資料的選取策略，用來訓練RNN學習癲癇腦波變化趨勢，冀望透過其對資料概泛化(generalization)的能力，快速找出異常之腦波，並客觀地標記出正確棘波的時間點位置，完成單通道偵測。最後，延伸單通道技術至多通道，目標建立一圖像化腦像圖，試圖標定出發生癲癇樣放電波之區塊，局部區域以顏色深淺表式癲癇波發生之機率，其優點有三：(1) 提供發生棘波最明顯之區域，若只記錄在單一通道或可歸類為為干擾波。(2) 定位癲癇樣放電約略位置，作為醫師未來手術前之參考依據。(3) 此技術若能成功辨識已確定之癲癇病患，能客觀地協助醫師判讀其餘四位病患，解決典型放電樣波不足的問題。特別感謝財團法人長庚紀念醫院基隆院區-神經內科主任彭宗義醫師於前期中提供專業意見與病患腦波資料，藉此吾人得以將神經網路技術與醫療應用做跨領域的結合。This research proposal aims to, based on verification of single-channel EEG, design a reliable tool for assisting physicians in clinical identification of epilepsy by exploiting correlation information from multiple EEG channels. From the results of previous NSC-granted projects, we find that there is high percentage of atypical epileptiform discharges, for example, in five patients’ sampling data, only one of them can be confidently identified by physician, this means that even an experienced physician is likely to encounter the problem of lacking typical epileptiform discharges, let alone that a single-channel EEG often failed to reach an affirmative decision on identifying epileptiform discharges. Therefore, there exists an urgent need of developing techniques for mining atypical epileptiform discharges, and single-channel EEG has failed to fully assist the interpretation of epilepsy.In this proposal, a method based on a spatiotemporal type neural network (simple recurrent neural network, SRNN) is developed for automatically screening epileptic EEG, i.e. typical and atypical epileptiform discharges can be discriminated by SRNN. Spectrum feature data are used to train SRNN. Specifically, time-frequency features derived from HHT will be selected using a screening strategy for preparing training data. Subsequently, result of this single-channel EEG is extended by exploiting correlations between multiple channels so that a real-tile scalp map is established, and from which physicians can locate the lesion on the brain by pin point the epileptiform discharges. The beauty of this approach is twofold: (1) it can effectively remove interference waves (false discharges), as a single EEG channel identified in the scalp map indicates it is not a true discharge (2) can solve the aforementioned problem of insufficient sampling data available to physicians.

Keyword(s)

多通道關聯性
腦像圖
棘波偵測
EEG
遞迴式神經網路
希爾伯特-黃轉換
Multichannel Correlation
Spike Detection
EEG
Recurrent Neural Network
Hilbert-Huang Transformation (HHT)
Scalp map

DSpace CRIS

Applying Multichannel Scalp Map to Auxilary Epilepsy Decision

基本資料

Description