[go: up one dir, main page]
More Web Proxy on the site http://driver.im/

JP5660436B2 - Periodic information extraction method - Google Patents

Periodic information extraction method Download PDF

Info

Publication number
JP5660436B2
JP5660436B2 JP2010195648A JP2010195648A JP5660436B2 JP 5660436 B2 JP5660436 B2 JP 5660436B2 JP 2010195648 A JP2010195648 A JP 2010195648A JP 2010195648 A JP2010195648 A JP 2010195648A JP 5660436 B2 JP5660436 B2 JP 5660436B2
Authority
JP
Japan
Prior art keywords
amplitude
phase
present
wave
value
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Expired - Fee Related
Application number
JP2010195648A
Other languages
Japanese (ja)
Other versions
JP2012050685A (en
Inventor
康 成瀬
康 成瀬
村田 勉
勉 村田
真人 岡田
真人 岡田
健 瀧山
健 瀧山
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
National Institute of Information and Communications Technology
Original Assignee
National Institute of Information and Communications Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by National Institute of Information and Communications Technology filed Critical National Institute of Information and Communications Technology
Priority to JP2010195648A priority Critical patent/JP5660436B2/en
Publication of JP2012050685A publication Critical patent/JP2012050685A/en
Application granted granted Critical
Publication of JP5660436B2 publication Critical patent/JP5660436B2/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Landscapes

  • Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)

Description

本発明は、振幅・周波数が揺らぎノイズを含む系において、高精度で周期的情報を抽出する方法に関する。   The present invention relates to a method for extracting periodic information with high accuracy in a system in which amplitude and frequency include fluctuation noise.

神経活動など、振幅・周波数が揺らぎノイズを含む系において、高精度で位相及び振幅を推定したい場合がある。
例えば、α波は、8-13Hz の後頭の視覚野から発生する周期的活動であり、後頭由来の脳波の主要な成分である。近年、このα波の振幅が、working memory task 等のcognitive task のパフォーマンスに影響を与えること(非特許文献1〜2)や、刺激提示時のα波の位相が知覚に影響を与えること(非特許文献3〜5)が報告され、α波の位相及び振幅に注目が集まっている。
There are cases where it is desired to estimate the phase and amplitude with high accuracy in a system including amplitude and frequency fluctuation noise such as neural activity.
For example, alpha waves are periodic activities that occur from the occipital visual cortex at 8-13 Hz, and are the main component of the occipital brain waves. In recent years, the amplitude of this alpha wave affects the performance of cognitive tasks such as working memory tasks (Non-Patent Documents 1 and 2), and the phase of the alpha wave at the time of stimulus presentation affects perception (non- Patent Documents 3 to 5) have been reported, and attention has been focused on the phase and amplitude of the α wave.

従来の脳波を用いた研究では、α波の瞬時における位相及び振幅を推定するために、主にヒルベルト変換(非特許文献6〜8)、ウェーブレット変換(非特許文献9〜10) が用いられてきた。   In research using conventional electroencephalograms, Hilbert transform (Non-Patent Documents 6 to 8) and wavelet transform (Non-Patent Documents 9 to 10) have been mainly used to estimate the instantaneous phase and amplitude of α waves. It was.

しかし、脳波には多くのノイズが含まれていることから、単一試行からこれらの手法で推定したα波の瞬時における位相及び振幅の推定精度が悪いため、試行間平均をとることでノイズを軽減して議論してきた。
このように試行間平均をとることは、試行間で同期した活動しか抽出できないという問題点がある。脳の高次の活動は試行間で同期していない可能性が高く(非特許文献11)、このような活動は単一試行で議論する必要がある。そして、そのためには瞬時における位相及び振幅の推定精度を向上させる必要がある。
However, since the brain waves contain a lot of noise, the estimation accuracy of the instantaneous phase and amplitude of the α wave estimated by these methods from a single trial is poor. I've been discussing it alleviating.
Taking the average between trials in this way has a problem that only activities synchronized between trials can be extracted. Higher brain activity is likely not synchronized between trials (Non-Patent Document 11), and such activities need to be discussed in a single trial. For that purpose, it is necessary to improve the estimation accuracy of the instantaneous phase and amplitude.

従来の技術では、α波の位相や振幅を推定するには、その精度に限界があった。   In the prior art, there is a limit to the accuracy in estimating the phase and amplitude of the α wave.

W. Klimesch: Int. J.Psychophysiol. 24 (1996) 61W. Klimesch: Int. J. Psychophysiol. 24 (1996) 61 H. Van Dijk, J.M. Schoffelen, R. Oostenveld, and O. Jensen:J. Neurosci. 28 (2008) 1816H. Van Dijk, J.M.Schoffelen, R. Oostenveld, and O. Jensen: J. Neurosci. 28 (2008) 1816 N.A. Busch, J. Dubois, and R. VanRullen: J. Neurosci. 29 (2009) 7869N.A.Busch, J. Dubois, and R. VanRullen: J. Neurosci. 29 (2009) 7869 K.E. Mathewson, G. Gratton, M. Fabiani, D.M. Beck, and T. Ro: J.Neurosci. 29 (2009) 2725K.E.Mathewson, G. Gratton, M. Fabiani, D.M.Beck, and T. Ro: J. Neurosci. 29 (2009) 2725 F.J. Varela, A. Toro, E.R. John, and E.L. Schwartz: Neuropsychologia 19 (1981) 675F.J.Varela, A. Toro, E.R.John, and E.L.Schwartz: Neuropsychologia 19 (1981) 675 A. Matani, Y. Naruse, Y. Terazono, T. Iwasaki, N. Fujimaki,and T. Murata:to be published in IEEE Trans.Biomed. EngA. Matani, Y. Naruse, Y. Terazono, T. Iwasaki, N. Fujimaki, and T. Murata: to be published in IEEE Trans. Biomed. Eng V.V. Nikulin, and T. Brismar: Neuroscience 137 (2006)647V.V.Nikulin, and T. Brismar: Neuroscience 137 (2006) 647 V.V. Nikulin, K. Linkenkaer-Hansen, G. Nolte, S. Lemm,K.R. Muller,R.J.Ilmoniemi, and G. Curio: Eur. J. Neurosci.25 (2007) 3146Nikulin, K. Linkenkaer-Hansen, G. Nolte, S. Lemm, K.R.Muller, R.J.Ilmoniemi, and G. Curio: Eur. J. Neurosci. 25 (2007) 3146 J.M. Palva, S. Palva, and K. Kaila: J. Neurosci. 25 (2005)3962J.M.Palva, S. Palva, and K. Kaila: J. Neurosci. 25 (2005) 3962 Y. Naruse, A. Matani, Y. Miyawaki, and M. Okada: to be published in Hum.Brain. MappY. Naruse, A. Matani, Y. Miyawaki, and M. Okada: to be published in Hum.Brain. Mapp Arieli, A. Sterkin, A. Grinvald, and A. Aertsen: Science 273 (1996) 1868Arieli, A. Sterkin, A. Grinvald, and A. Aertsen: Science 273 (1996) 1868 K. Linkenkaer-Hansen, V.V. Nikouline, J.M. Palva, and R.J. Ilmoniemi:J.Neurosci. 21 (2001) 1370K. Linkenkaer-Hansen, V.V.Nikouline, J.M.Palva, and R.J.Illmoniemi: J. Neurosci. 21 (2001) 1370 J. Pearl: Probabilistic Reasoning in Intelligent Systems(Morgan Kaufmann,San Mateo, 1988)J. Pearl: Probabilistic Reasoning in Intelligent Systems (Morgan Kaufmann, San Mateo, 1988) K. Tanaka, H. Shouno, M. Okada, and D.M. Titterington:J. Phys. A Math.Gen. 37 (2004) 8675K. Tanaka, H. Shouno, M. Okada, and D.M.Titterington: J. Phys. A Math. Gen. 37 (2004) 8675 K. Takiyama, K. Katahira, and M. Okada: J. Phys. Soc.Jpn. 78 (2009) 064003K. Takiyama, K. Katahira, and M. Okada: J. Phys. Soc. Jpn. 78 (2009) 064003 K. Watanabe, H. Tanaka, K. Miura, and M. Okada: IEICE Trans. Inf. Syst. E92d (2009) 1362K. Watanabe, H. Tanaka, K. Miura, and M. Okada: IEICE Trans. Inf. Syst. E92d (2009) 1362 C.M. Bishop: Pattern Recognition and Machine Learning (Spinger-Verlag,Berlin, 2006)C.M.Bishop: Pattern Recognition and Machine Learning (Spinger-Verlag, Berlin, 2006) F.H. Lopes da Silva, J.P. Pijn, D. Velis, and P.C.G. Nijssen:Int.J.Psychophysiol. 26 (1997) 237F.H.Lopes da Silva, J.P. Pijn, D. Velis, and P.C.G.Nijssen: Int.J.Psychophysiol. 26 (1997) 237 C.J. Stam, J.P.M. Pijn, P. Suffczynski, and F.H. Lopes da Silva: Clin.Neurophysiol. 110 (1999) 1801C.J.Stam, J.P.M.Pijn, P. Suffczynski, and F.H.Lopes da Silva: Clin. Neurophysiol. 110 (1999) 1801 Y. Naruse, A. Matani, T. Hayakawa, and N. Fujimaki: Neuroimage 32 (2006) 1221Y. Naruse, A. Matani, T. Hayakawa, and N. Fujimaki: Neuroimage 32 (2006) 1221 S. Makeig, M. Westerfield, T.P. Jung, S. Enghoff, J.Townsend, E.Courchesne, and T.J. Sejnowski: Science 295 (2002) 690S. Makeig, M. Westerfield, T.P. Jung, S. Enghoff, J. Townsend, E. Couchesne, and T.J. Sejnowski: Science 295 (2002) 690 Mazaheri, and O. Jensen: Proc. Natl. Acad. Sci. U. S.A. 103 (2006) 2948Mazaheri, and O. Jensen: Proc. Natl. Acad. Sci. U. S.A. 103 (2006) 2948 M.L. Risner, C.J. Aura, J.E. Black, and T.J. Gawne: Neuroimage 45 (2009) 463M.L.Risner, C.J.Aura, J.E.Black, and T.J.Gawne: Neuroimage 45 (2009) 463 W. Klimesch, P. Sauseng, and W. Gruber: Neuroimage 47 (2009) 5W. Klimesch, P. Sauseng, and W. Gruber: Neuroimage 47 (2009) 5

そこで、本発明は、振幅・周波数が揺らぎノイズを含む系において、高精度で周期的情報を抽出する方法を提供することを課題とする。   Therefore, an object of the present invention is to provide a method for extracting periodic information with high accuracy in a system in which amplitude / frequency includes fluctuation noise.

上記課題を達成するため、本発明の周期的情報抽出方法は、振幅・周波数が揺らぎ、ノイズを含む系において、高精度で位相及び振幅を推定する方法であって、マルコフ確率場(MRF)のもとで確率伝播法による確率的推論手法を用い、瞬時の位相及び振幅を含むパラメータとその周辺尤度を算出し、その周辺尤度を最大化する第二種最尤推定法に基づいて、位相及び振幅のなめらかさを含むハイパーパラメータを算出し、得られたハイパーパラメータのもとで、瞬時の位相及び振幅を含むパラメータを推定することを特徴とする。   In order to achieve the above object, a periodic information extraction method according to the present invention is a method for estimating a phase and amplitude with high accuracy in a system including fluctuations in amplitude and frequency and including noise, and includes a Markov random field (MRF). Based on the second type maximum likelihood estimation method that calculates the parameter including the instantaneous phase and amplitude and its marginal likelihood, and maximizes the marginal likelihood using the stochastic reasoning method based on the probability propagation method. A hyper parameter including smoothness of phase and amplitude is calculated, and a parameter including instantaneous phase and amplitude is estimated based on the obtained hyper parameter.

ここで、ハイパーパラメータに、中心周波数も含めてもよい。   Here, the center frequency may be included in the hyper parameter.

ハイパーパラメータに、観測ノイズの強度も含も含めてもよい。   The hyper parameter may include the intensity of observation noise.

対象となる周期的情報としては、神経活動情報に好適に適用できる。   The target periodic information can be suitably applied to neural activity information.

神経活動情報としては例えば、脳において観測されるα波が挙げられる。   Examples of the neural activity information include α waves observed in the brain.

本発明によると、脳活動のα波など、ゆらぎとノイズのある系において、位相及び振幅などの周期的情報を高精度で求められる。   According to the present invention, periodic information such as phase and amplitude can be obtained with high accuracy in a system with fluctuation and noise, such as α wave of brain activity.

脳波の実測データ(EEG) と観測データ(Observed data) のFourier spectrumの比較を示すグラフGraph showing comparison of Fourier spectrum between EEG measured data (EEG) and observed data (Observed data) 位相、振幅に対する確率の典型例を示すグラフGraph showing typical examples of probability for phase and amplitude 観測値(Observed data)と本発明による推定値(MRF) と真値との間の絶対誤差の典型例を示すグラフGraph showing typical examples of absolute error between observed data (Observed data) and estimated value (MRF) according to the present invention and true value 観測値(Observed data)と本発明による推定値(MRF)と真値との間の平均絶対誤差の試行間平均値を示すグラフA graph showing the average value between trials of the average absolute error between the observed data (Observed data) and the estimated value (MRF) according to the present invention and the true value 本発明(MRF)及びヒルベルト変換(HT)による位相推定値と真の位相値との間の絶対誤差の典型例を示すグラフGraph showing a typical example of absolute error between phase estimate and true phase value by the present invention (MRF) and Hilbert transform (HT) 本発明(MRF)及びヒルベルト変換(HT)による振幅推定値と真の振幅値との間の絶対誤差の典型例を示すグラフGraph showing typical example of absolute error between amplitude estimate and true amplitude value by the present invention (MRF) and Hilbert transform (HT) 本発明(MRF)及びヒルベルト変換(HT) よる位相推定値と真の位相値との間の平均絶対誤差の試行間平均値を示すグラフThe graph which shows the average value between trials of the average absolute error between the phase estimate by the present invention (MRF) and the Hilbert transform (HT) and the true phase value 本発明(MRF)及びヒルベルト変換(HT) よる振幅推定値と真の振幅値との間の平均絶対誤差の試行間平均値を示すグラフThe graph which shows the average value between trials of the average absolute error between the amplitude estimate by the present invention (MRF) and the Hilbert transform (HT) and the true amplitude value

以下に、本発明の実施形態を図面に基づいて説明する。実施形態は、前記特許文献など従来公知の技術を援用して適宜設計変更可能であり、また、計算の方法及び装置についても従来公知の技術を適宜適用可能である。
以下の実施例では、対象をα波とした例を挙げるが、本発明は、振幅・周波数が揺らぎノイズを含む系一般に適用可能である。
Embodiments of the present invention will be described below with reference to the drawings. The embodiment can be appropriately changed in design by using a conventionally known technique such as the above-mentioned patent document, and a conventionally known technique can be appropriately applied to a calculation method and apparatus.
In the following embodiment, an example in which the target is an α wave will be described. However, the present invention is applicable to a general system including fluctuation noise in amplitude and frequency.

本発明では、マルコフ確率場(MRF) モデルを用い、従来の脳波解析で使われてきた手法よりも精度よく瞬時の位相及び振幅を推定する方法を提供する。観測されたα波は、特定の周波数の位相及び振幅があり、そこに観測ノイズが加わっているとモデル化することができる。   The present invention provides a method for estimating an instantaneous phase and amplitude with higher accuracy than a method used in a conventional electroencephalogram analysis using a Markov random field (MRF) model. The observed α wave has a phase and amplitude of a specific frequency, and can be modeled when observation noise is added thereto.

α波の位相は、各人により異なるα波中心周波数の周辺で時間とともになめらかに変化しながら進行し、その振幅も時間とともになめらかに変化している(非特許文献12)。そこで、本発明では、特定のサンプリングポイントの瞬時の位相及び振幅は、その前後のサンプリングポイントからなめらかに変化しているという位相及び振幅に関するMRFモデルによりα波を記述する。   The phase of the α wave progresses while changing smoothly with time around the α wave center frequency that varies from person to person, and its amplitude also changes smoothly with time (Non-Patent Document 12). Therefore, in the present invention, the α wave is described by the MRF model relating to the phase and amplitude in which the instantaneous phase and amplitude of a specific sampling point are smoothly changed from the sampling points before and after the specific sampling point.

しかし、位相及び振幅の変化のなめらかさやα波の中心周波数は、被験者ごとに異なり、さらに、観測ノイズは実験ごとに異なっているため、定かではない。そこで、各サンプリングポイントにおける瞬時の位相及び振幅といったパラメータのみならず、位相及び振幅の変化のなめらかさ、α波の中心周波数、観測ノイズの強度といったハイパーパラメータも同時に推定する必要がある。   However, the smoothness of changes in phase and amplitude and the center frequency of the α wave are different from subject to subject, and the observation noise is different from experiment to experiment. Therefore, it is necessary to simultaneously estimate not only parameters such as instantaneous phase and amplitude at each sampling point, but also hyper parameters such as smoothness of phase and amplitude changes, center frequency of α wave, and intensity of observation noise.

MRF の元では、確率的推論手法の一つである確率伝播法(非特許文献13)を用いることで瞬時の位相及び振幅といったパラメータの推定及び周辺尤度を計算することが可能である(非特許文献14〜16)。そこで、本発明では、この周辺尤度を最大化する第二種最尤推定(非特許文献17)に基づいて、位相及び振幅のなめらかさ、α波の中心周波数、観測ノイズの強度といったハイパーパラメータを推定する。そして、この推定したハイパーパラメータの元で瞬時の位相及び振幅といったパラメータを推定するアルゴリズムを構築する。   Under the MRF, it is possible to estimate parameters such as instantaneous phase and amplitude and calculate marginal likelihood by using the probability propagation method (Non-patent Document 13), which is one of the probabilistic inference methods (non-patent document 13). Patent Documents 14 to 16). Therefore, in the present invention, hyperparameters such as the smoothness of the phase and amplitude, the center frequency of the α wave, and the intensity of the observation noise based on the second type maximum likelihood estimation (Non-Patent Document 17) that maximizes the marginal likelihood. Is estimated. Then, an algorithm for estimating parameters such as instantaneous phase and amplitude based on the estimated hyperparameters is constructed.

以下の実施例では、まず、脳波を模した人工データを用いて、本発明を用いて推定した時に観測ノイズを軽減できるかについて検証する。続いて、本発明を用いて推定した時の瞬時の位相及び振幅が、脳波データ解析で従来一般的に使われるヒルベルト変換よりも推定精度がよいことを示す。   In the following examples, first, it will be verified whether observation noise can be reduced when estimated using the present invention, using artificial data simulating brain waves. Next, it will be shown that the instantaneous phase and amplitude when estimated using the present invention have better estimation accuracy than the Hilbert transform that is conventionally used in brain wave data analysis.

脳波によって計測されるα波の時系列における各サンプリングポイントk の観測データyk は、サンプリングポイントk におけるα波の瞬時の位相及び振幅をそれぞれxk、ak とすると、そこに各時刻間で独立なノイズσ がのっているという次式で表すことができる。   The observation data yk at each sampling point k in the time series of the α wave measured by the brain waves is independent noise between each time, assuming that the instantaneous phase and amplitude of the α wave at the sampling point k are xk and ak, respectively. It can be expressed by the following equation that σ is on.

Figure 0005660436
Figure 0005660436

ここで、ノイズをガウスノイズであると仮定すると、尤度関数は次式で表すことができる。   Here, assuming that the noise is Gaussian noise, the likelihood function can be expressed by the following equation.

Figure 0005660436
Figure 0005660436

ここで、α はノイズ強度を表すハイパーパラメータである。
α波の周波数及び振幅は、時間とともになめらかに変化することが知られているので(非特許文献11)、位相xk はα波中心周波数の周辺で時間とともになめらかに変化しながら進行し、また振幅ak も時間とともになめらかに変化する場合を考える。
このときの観測値y = {y1, y2, . . . , yN} から、振幅a = {a1, a2, . . . , aN}と、位相x = {x1, x2.
. . . , xN} を、確率伝播法により同時に推定する。
ここで、瞬時位相及び振幅の生成モデルは、次式とする。
Here, α is a hyper parameter representing noise intensity.
Since it is known that the frequency and amplitude of the α wave change smoothly with time (Non-Patent Document 11), the phase xk proceeds while changing smoothly with time around the α wave center frequency, and the amplitude Consider the case where ak also changes smoothly with time.
From the observed value y = {y1, y2,..., YN}, the amplitude a = {a1, a2,..., AN} and the phase x = {x1, x2.
, xN} are estimated simultaneously by the probability propagation method.
Here, the instantaneous phase and amplitude generation model is as follows.

Figure 0005660436
Figure 0005660436

Figure 0005660436
Figure 0005660436

ただし、I0(β) は、modified Bessel function of order zero であり、次式で表すことができる。   However, I0 (β) is a modified Bessel function of order zero and can be expressed by the following equation.

Figure 0005660436
Figure 0005660436

振幅ak は、0 以上であるため、規格化定数Zampは次式となる。   Since the amplitude ak is 0 or more, the normalization constant Zamp is expressed by the following equation.

Figure 0005660436
Figure 0005660436

ここで、β、 γ、ω はそれぞれ、位相変化のなめらかさ、振幅変化のなめらかさ、α波中心周波数を表すハイパーパラメータである。
振幅と位相の事前分布を独立と仮定すると、次式となる。
Here, β, γ, and ω are hyperparameters representing the smoothness of the phase change, the smoothness of the amplitude change, and the α-wave center frequency, respectively.
Assuming that the prior distribution of amplitude and phase is independent, the following equation is obtained.

Figure 0005660436
Figure 0005660436

マルコフ確率場を仮定していることから、観測データ数をN とすると、次式で表わされる。   Assuming a Markov random field, if the number of observation data is N, it is expressed by the following equation.

Figure 0005660436
Figure 0005660436

また、周辺分布は次式で表わされる。   The peripheral distribution is expressed by the following equation.

Figure 0005660436
Figure 0005660436

ここで、次式より、Z = p(y) とおくと、下式が得られる。   Here, from the following equation, if Z = p (y), the following equation is obtained.

Figure 0005660436
Figure 0005660436

Figure 0005660436
Figure 0005660436

この式における重積分は一般に計算困難である。しかし、確率伝播法により以下で定義されるmessage ML、MR に関する漸化式を構築することで、数値積分をする上での振幅の分割数をna、位相の分割数をnp とすると2(N −1) 重積分の計算量をO(na(N-1)×np(N-1))からO((N−1)na2×(N−1)np2)へ削減することができる。
ここでML、MR を、次式とすると下式が得られる。
The multiple integral in this equation is generally difficult to calculate. However, by constructing a recurrence formula for message ML and MR defined below by the probability propagation method, the number of amplitude divisions for numerical integration is na and the number of phase divisions is np. -1) The amount of calculation of multiple integral can be reduced from O (na (N-1) × np (N-1) ) to O ((N−1) na 2 × (N−1) np 2 ) .
If ML and MR are the following equations, the following equations are obtained.

Figure 0005660436
Figure 0005660436

Figure 0005660436
Figure 0005660436

Figure 0005660436
Figure 0005660436

Figure 0005660436
Figure 0005660436

Figure 0005660436
Figure 0005660436

次に、ML(ak, xk)、MR(ak, xk) の漸化式を構築する。   Next, a recurrence formula of ML (ak, xk) and MR (ak, xk) is constructed.

Figure 0005660436
Figure 0005660436

Figure 0005660436
Figure 0005660436

これらを用いることで、任意のk におけるML(ak, xk)、 MR(ak, xk)を計算できる。
また、ak、 xk を周辺化することで、次式として、規格化定数Z を求めることができる。
By using these, ML (ak, xk) and MR (ak, xk) at an arbitrary k can be calculated.
Further, by normalizing ak and xk, the normalization constant Z can be obtained as the following equation.

Figure 0005660436
Figure 0005660436

Figure 0005660436
Figure 0005660436

Figure 0005660436
Figure 0005660436

Figure 0005660436
Figure 0005660436

Figure 0005660436
Figure 0005660436

Figure 0005660436
Figure 0005660436

Figure 0005660436
Figure 0005660436

Figure 0005660436
Figure 0005660436

Figure 0005660436
Figure 0005660436

図4は、観測値(Observed data) 及び、本発明による推定値(MRF)と真値との間の平均絶対誤差の試行間平均値を示すグラフである。
真値からの平均絶対誤差を観測データ及び本発明の推定値とで比較した結果、平均絶対誤差はすべての試行において観測データよりも推定値の方が小さかった(Wilcoxon signed rank test、 p < 0.0001)。このことから、本発明を用いて推定することで、観測ノイズを軽減できていることが示された。
FIG. 4 is a graph showing the observed value (Observed data) and the average value between trials of the average absolute error between the estimated value (MRF) and the true value according to the present invention.
As a result of comparing the average absolute error from the true value with the observed data and the estimated value of the present invention, the average absolute error was smaller in the estimated value than the observed data in all trials (Wilcoxon signed rank test, p <0.0001). ). From this, it was shown that the observation noise can be reduced by estimating using the present invention.

次に、本発明による瞬時位相及び振幅の推定精度を調べた。
図5は、本発明(MRF)及びヒルベルト変換(HT)による位相推定値と真の位相値との間の絶対誤差の典型例を示すグラフであり、図6は、本発明(MRF)及びヒルベルト変換(HT)による振幅推定値と真の振幅値との間の絶対誤差の典型例を示すグラフである。
図7は、本発明(MRF)及びヒルベルト変換(HT) よる位相推定値と真の位相値との間の平均絶対誤差の試行間平均値を示すグラフであり、図8は、本発明(MRF)及びヒルベルト変換(HT) よる振幅推定値と真の振幅値との間の平均絶対誤差の試行間平均値を示すグラフである。
Next, the estimation accuracy of the instantaneous phase and amplitude according to the present invention was examined.
FIG. 5 is a graph showing a typical example of the absolute error between the phase estimate by the present invention (MRF) and the Hilbert transform (HT) and the true phase value, and FIG. 6 shows the present invention (MRF) and the Hilbert. It is a graph which shows the typical example of the absolute error between the amplitude estimated value by conversion (HT), and a true amplitude value.
FIG. 7 is a graph showing the average value between trials of the average absolute error between the phase estimate by the present invention (MRF) and the Hilbert transform (HT) and the true phase value, and FIG. ) And Hilbert transform (HT) are graphs showing the average value between trials of the average absolute error between the amplitude estimation value and the true amplitude value.

Figure 0005660436
Figure 0005660436

Figure 0005660436
Figure 0005660436

位相、振幅の平均絶対誤差はともにすべての試行において、本発明にて推定した方が小さかった(Wilcoxon signed rank test, p < 0.0001)。   The average absolute error of both phase and amplitude was smaller in all trials as estimated by the present invention (Wilcoxon signed rank test, p <0.0001).

以上のように、本発明では、α波を特定のサンプリングポイントの瞬時の位相及び振幅は、前後のサンプリングポイントからなめらかに変化するというMRF モデルにより記述し、これを確率伝播法を用いて瞬時位相及び振幅を推定し、周辺尤度を計算する手法を構築した。そして、周辺尤度を最大化することでハイパーパラメータを推定した。
尤度関数(数2)は、位相、振幅の非線形項を含む。しばしば非線形項を含む事後分布から隠れ変数を推定する際にはラプラス近似が用いられる(非特許文献17)。しかし、ラプラス近似は単峰のガウス分布で事後分布を近似してしまうため、図2に示した多峰性の事後分布へ適用は、精度の低い推定値を与える可能性が高く、本発明にて構築したMRF モデルには適さない。
As described above, in the present invention, the α wave is described by the MRF model in which the instantaneous phase and amplitude of a specific sampling point change smoothly from the preceding and succeeding sampling points, and this is expressed by using the stochastic propagation method. And the method to estimate the amplitude and calculate the marginal likelihood was constructed. The hyperparameters were estimated by maximizing the marginal likelihood.
The likelihood function (Equation 2) includes phase and amplitude nonlinear terms. Laplace approximation is often used when estimating hidden variables from posterior distributions containing nonlinear terms (Non-patent Document 17). However, since the Laplace approximation approximates the posterior distribution with a unimodal Gaussian distribution, application to the multimodal posterior distribution shown in FIG. It is not suitable for the MRF model constructed by

また、本発明では瞬時の位相及び振幅の推定にMPM 推定を用いたが、他にpoterior mean 推定(PM 推定) を用いることもできる。しかし、各瞬時の位相値及び振幅値に対する確率は多峰性を示すので、このような場合のPM 推定結果は、確率が高い部分の間の確率が低い部分を推定してしまう可能性があり、適切ではない。そのため、本発明ではMPM 推定を用いることが適切である。
従来、周波数が揺らいでいるときの脳波における瞬時の位相及び振幅の推定にヒルベルト変換が広く使われてきた。脳波には観測ノイズがのっているにもかかわらず、ヒルベルト変換ではそのノイズも信号と見なして瞬時の位相及び振幅を推定してしまっていたため、精度が悪かった。しかし、本発明では瞬時の位相及び振幅の推定と同時に観測ノイズも軽減していることから、ヒルベルト変換よりも精度よく瞬時位相及び振幅が推定できていると考えられる。
本発明を脳波の実データに適応することにより、より精度よく瞬時位相及び振幅の推定が可能となる。
In the present invention, MPM estimation is used for instantaneous phase and amplitude estimation, but poterior mean estimation (PM estimation) can also be used. However, since the probability for each instantaneous phase value and amplitude value shows multimodality, the PM estimation result in such a case may estimate the low probability portion between the high probability portions. ,Not appropriate. Therefore, it is appropriate to use MPM estimation in the present invention.
Conventionally, the Hilbert transform has been widely used for estimation of instantaneous phase and amplitude in an electroencephalogram when the frequency fluctuates. Despite the presence of observation noise in the brain waves, the Hilbert transform assumed that the noise was also a signal and estimated the instantaneous phase and amplitude, so the accuracy was poor. However, in the present invention, since the observation noise is reduced simultaneously with the estimation of the instantaneous phase and amplitude, it is considered that the instantaneous phase and amplitude can be estimated more accurately than the Hilbert transform.
By applying the present invention to actual brain wave data, the instantaneous phase and amplitude can be estimated more accurately.

近年、α波の振幅や位相がcognitive task のパフォーマンスや知覚に影響を与えることが報告されているが(非特許文献1〜5)、試行間平均をとった結果でしか議論できていないため、試行間で同期していない活動を検出できていない。さらに、α波は非線形振動子と見なすことが出来る可能性が示唆されているが(非特許文献10、18〜19)、α波の位相が視覚刺激などの外部刺激により影響を受けるか否かについては結論が出ていない(非特許文献20〜24)。この原因も、単一試行における推定精度が悪いことに依存する。本発明によると、高い精度でα波の位相や振幅を推定することで、これらの問題を解決することができる可能性があり、脳波を用いた神経科学的研究の発展に寄与する。   In recent years, it has been reported that the amplitude and phase of alpha waves affect the performance and perception of cognitive tasks (Non-Patent Documents 1 to 5), but since only the results of taking the average between trials can be discussed, Unable to detect activities that are not synchronized between attempts. Furthermore, although it has been suggested that the α wave can be regarded as a non-linear oscillator (Non-Patent Documents 10, 18 to 19), whether or not the phase of the α wave is affected by an external stimulus such as a visual stimulus. There is no conclusion on (Non-patent Documents 20 to 24). This cause also depends on poor estimation accuracy in a single trial. According to the present invention, there is a possibility that these problems can be solved by estimating the phase and amplitude of the α wave with high accuracy, which contributes to the development of neuroscientific research using brain waves.

本発明によると、脳活動のα波など、ゆらぎとノイズのある系において、位相及び振幅などの周期的情報を高精度でを求められるので、脳波を用いた神経科学的研究や工学的応用としてブレインマシンインターフェースなどに寄与するので、産業上利用価値が高い。   According to the present invention, it is possible to obtain periodic information such as phase and amplitude with high accuracy in a system with fluctuation and noise, such as alpha wave of brain activity, so that it can be used for neuroscience research and engineering applications using brain waves. As it contributes to the brain machine interface, it is highly industrially useful.

Claims (1)

振幅・周波数が揺らぎ、ノイズを含む系において、高精度で位相及び振幅を推定する方法であって、
マルコフ確率場(MRF)のもとで確率伝播法による確率的推論手法を用い、瞬時の位相及び振幅を含むパラメータ周辺尤度を算出し、
その周辺尤度を最大化する第二種最尤推定法に基づいて、位相及び振幅のなめらかさを含むハイパーパラメータを算出し、
得られたハイパーパラメータのもとで、位相及び振幅を推定する
ことを特徴とする周期的情報抽出方法。
A method for estimating phase and amplitude with high accuracy in a system including fluctuations in amplitude and frequency and noise.
Calculate the marginal likelihood of parameters including instantaneous phase and amplitude using a stochastic inference method based on belief propagation under a Markov random field (MRF),
Based on the second type maximum likelihood estimation method that maximizes the marginal likelihood, calculate the hyperparameters including the smoothness of the phase and amplitude,
A periodic information extraction method characterized by estimating the phase and amplitude based on the obtained hyperparameters.
JP2010195648A 2010-09-01 2010-09-01 Periodic information extraction method Expired - Fee Related JP5660436B2 (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
JP2010195648A JP5660436B2 (en) 2010-09-01 2010-09-01 Periodic information extraction method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
JP2010195648A JP5660436B2 (en) 2010-09-01 2010-09-01 Periodic information extraction method

Publications (2)

Publication Number Publication Date
JP2012050685A JP2012050685A (en) 2012-03-15
JP5660436B2 true JP5660436B2 (en) 2015-01-28

Family

ID=45904758

Family Applications (1)

Application Number Title Priority Date Filing Date
JP2010195648A Expired - Fee Related JP5660436B2 (en) 2010-09-01 2010-09-01 Periodic information extraction method

Country Status (1)

Country Link
JP (1) JP5660436B2 (en)

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
GB0410248D0 (en) * 2004-05-07 2004-06-09 Isis Innovation Signal analysis method
US7212665B2 (en) * 2004-11-05 2007-05-01 Honda Motor Co. Human pose estimation with data driven belief propagation
US7620227B2 (en) * 2005-12-29 2009-11-17 General Electric Co. Computer-aided detection system utilizing temporal analysis as a precursor to spatial analysis
JP5175515B2 (en) * 2007-10-02 2013-04-03 株式会社東芝 Model construction apparatus, model construction method and program

Also Published As

Publication number Publication date
JP2012050685A (en) 2012-03-15

Similar Documents

Publication Publication Date Title
Fu et al. Classification of seizure based on the time-frequency image of EEG signals using HHT and SVM
Lee et al. Classification of normal and epileptic seizure EEG signals using wavelet transform, phase-space reconstruction, and Euclidean distance
Siuly et al. Designing a robust feature extraction method based on optimum allocation and principal component analysis for epileptic EEG signal classification
Xie et al. Wavelet-based sparse functional linear model with applications to EEGs seizure detection and epilepsy diagnosis
Lima et al. Kernel machines for epilepsy diagnosis via EEG signal classification: A comparative study
Val-Calvo et al. Optimization of real-time EEG artifact removal and emotion estimation for human-robot interaction applications
Zhang et al. EEG-based expert system using complexity measures and probability density function control in alpha sub-band
Kaleem et al. Patient-specific seizure detection in long-term EEG using signal-derived empirical mode decomposition (EMD)-based dictionary approach
Parvez et al. Epileptic seizure detection by exploiting temporal correlation of electroencephalogram signals
Tian et al. Dual-encoder VAE-GAN with spatiotemporal features for emotional EEG data augmentation
Arnau-González et al. ES1D: A deep network for EEG-based subject identification
Ibrahim et al. Deep‐learning‐based seizure detection and prediction from electroencephalography signals
Li et al. Infrasound signal classification based on spectral entropy and support vector machine
Lee et al. Automated epileptic seizure waveform detection method based on the feature of the mean slope of wavelet coefficient counts using a hidden Markov model and EEG signals
Movahed et al. Automatic diagnosis of mild cognitive impairment based on spectral, functional connectivity, and nonlinear EEG‐Based features
Xie et al. Signal decomposition by multi-scale PCA and its applications to long-term EEG signal classification
Prabhakar et al. Comparison of fuzzy output optimization with expectation maximization algorithm and its modification for epilepsy classification
Fryz Conditional linear random process and random coefficient autoregressive model for EEG analysis
Yucelbas et al. Effect of EEG time domain features on the classification of sleep stages
Parvez et al. Features extraction and classification for Ictal and Interictal EEG signals using EMD and DCT
Ahmad et al. Robust epileptic seizure detection based on biomedical signals using an advanced multi-view deep feature learning approach
Hellar et al. Epileptic electroencephalography classification using embedded dynamic mode decomposition
JP5660436B2 (en) Periodic information extraction method
Huang et al. Automatic artifact removal in EEG using independent component analysis and one-class classification strategy
Jamil et al. Exploring the potential of pretrained CNNs and time-frequency methods for accurate epileptic EEG classification: a comparative study

Legal Events

Date Code Title Description
A621 Written request for application examination

Free format text: JAPANESE INTERMEDIATE CODE: A621

Effective date: 20130726

A977 Report on retrieval

Free format text: JAPANESE INTERMEDIATE CODE: A971007

Effective date: 20140121

A131 Notification of reasons for refusal

Free format text: JAPANESE INTERMEDIATE CODE: A131

Effective date: 20140304

A521 Request for written amendment filed

Free format text: JAPANESE INTERMEDIATE CODE: A523

Effective date: 20140407

TRDD Decision of grant or rejection written
A01 Written decision to grant a patent or to grant a registration (utility model)

Free format text: JAPANESE INTERMEDIATE CODE: A01

Effective date: 20141105

A61 First payment of annual fees (during grant procedure)

Free format text: JAPANESE INTERMEDIATE CODE: A61

Effective date: 20141120

R150 Certificate of patent or registration of utility model

Ref document number: 5660436

Country of ref document: JP

Free format text: JAPANESE INTERMEDIATE CODE: R150

S533 Written request for registration of change of name

Free format text: JAPANESE INTERMEDIATE CODE: R313533

R350 Written notification of registration of transfer

Free format text: JAPANESE INTERMEDIATE CODE: R350

R250 Receipt of annual fees

Free format text: JAPANESE INTERMEDIATE CODE: R250

R250 Receipt of annual fees

Free format text: JAPANESE INTERMEDIATE CODE: R250

R250 Receipt of annual fees

Free format text: JAPANESE INTERMEDIATE CODE: R250

LAPS Cancellation because of no payment of annual fees