Patent No. 5785653 System and method for predicting internal condition of live body

 

Patent No. 5785653

System and method for predicting internal condition of live body (Kiyuna, et al., Jul 28, 1998)

Abstract

A live body internal condition predicting and expressing system is capable of predicting the internal condition of a live body on the basis of an electromagnetic field distribution at significantly reduced period. The system comprises an electromagnetic field distribution measuring portion, a training data generating portion generating a plurality of training data on the basis of an electromagnetic field distribution derived from a head model data designating a head model and a predetermined dipole parameter and an internal condition category data describing a relationship between the active region of a brain and the internal condition of the live body, an inference portion deriving a numeric value representative of the active region of the brain from the electromagnetic field distribution in the training data with employing a neural network having predetermined coupling coefficients representative of coupling condition between each unit forming each layer and transforming portion transforming the numeric value representative of the active region in the brain output from said inferencing means into an expression indicative of the internal condition of the live body.

Notes:

BACKGROUND OF THE INVENTION

1. Field of the Invention

The present invention relates generally to a system and method for predicting and expressing an internal condition of a live body on the basis of a distribution of electromagnetic fields on a scalp of the live body.

2. Description of the Related Art

Conventionally, a system for predicting an internal condition of a live body as illustrated in FIG. 16 has been employed for predicting a word to be spoken by a testee. Such system has been disclosed in Japanese Unexamined Patent Publication (Kokai) No. Heisei 2-232783, for example.

In FIG. 16, 91 denotes a plurality of electrodes, 92 denotes an electroencephalograph, 93 denotes a brain wave topographic pattern generator device, 94 denotes a neural network, 95 denotes a syllable data teaching portion, 96 denotes a syllable presenting portion, 97 denotes a control portion controlling respective components, 98 denotes a voice detector devices and 99 a pre-input processing devise.

In such a live body internal condition predicting system, at first, a certain syllable is spoken by the testee. At this time, a potential distribution of brain wave, i.e. the brain wave topographic pattern, is measured by the topographic pattern generator device 93 via the electrodes 91 and the electroencephalograph 92. By repeating the foregoing measurement sequence for a plurality of times, a plurality of training data is generated.

Employing the training data thus generated, a neural network, i.e. neural network 94 in the shown example, learns a relationship between the syllable thought in the brain of the testee, which is input by the syllable data teaching portion 95, and the potential distribution on the scalp. After completion of learning, the measured potential distribution of the brain wave is input to the neural network 94. Thus, it becomes possible to recognize the intended syllable of the testee without requiring the testee to speak. In the conventional method, the waveform of the brain wave is used as the input data. However, it is inherent to detect noise together with syllable pattern due to background brain wave, thermal noise of the measuring equipment, and so forth. Therefore, a long period has heretofore been required in preparatory processing for removing such noise. As a means for removing the noise, it is typical to perform filtering by fast Fourier transformation (FFT) for the measured brain wave data and thereby to remove frequencies other than the signal to be a subject for analysis.

However, in the conventional method, since the brain wave measured on the scalp of the testee is used as the training data, a huge amount of time is required in generation of data when a large amount of training data is to be prepared, to cause substantial load on the testee. On the other hand, when the amount of training data is reduced, reducing load on the testee, the accuracy of learning of the neural network becomes low, increasing a possibility of failure in recognition of the syllable.

In addition, it is required to maintain data of the entire period, in which data measurement is performed, in a frequency analyzing method employing the FFT. Therefore, to store the data of the entire measurement period, a large amount of memory storage capacity is required. In the filtering employing the FFT, a single common frequency is employed for all of the obtained data throughout the entire period. This causes a problem in that, when a low frequency brain wave component is important at one certain timing, and a high frequency brain wave component is important at another timing, one of the two frequency components will be excluded through the filtering.

Also, when the internal condition is to be predicted on the basis of the obtained measured data, it becomes necessary to perform retrieval on a dictionary. The retrieval process requires an additional period.

SUMMARY OF THE INVENTION

The present invention has been worked out for solving the problems set forth above. It is an object of the present invention to provide a system for predicting the internal condition of a live body which can significantly reduce load on a testee, and instantly detects a content of category which the testee wishes to speak.

According to the first aspect of the invention, a system for expressing internal condition of a live body which predicts internal condition of the live body on the basis of an electromagnetic field distribution measured on a scalp of the live body, comprises:

electromagnetic field distribution measuring means for measuring electromagnetic field distribution caused on the scalp of the live body;

training data generating means for generating a plurality of training data on the basis of an electromagnetic field distribution derived from a head model data designating a head model and a predetermined dipole parameter and an internal condition category data describing a relationship between the active region of a brain and the internal condition of the live body;

inferencing means for deriving a numeric value representative of the active region of the brain from the electromagnetic field distribution in the training data with employing a neural network consisted of an input layer, an output layer and at least one hidden layer and having predetermined coupling coefficients representative of coupling condition between each unit forming each layer;

transforming means for transforming the numeric value representative of the active region in the brain output from the inferencing means into an expression indicative of the internal condition of the live body; and

the inferencing means having a coupling coefficient modifying means for modifying the coupling coefficient of the neural network for reducing an error between a correct numeric value representative of the correct active region of the brain described in the training data and the numeric value derived by the neural network and indicative of the active region of the brain so that a numeric value indicative of the active region in the brain is derived by the neural network with taking the electromagnetic field distribution measured by the electromagnetic field distribution measuring means when the error becomes smaller than a predetermined reference value.

The live body internal condition expressing system may further comprise noise data adding means for generating a noise and adding the generated noise to the training data generated by the training data generating means.

In the preferred construction, the inferencing means may employ a re-current type neural network.

According to the second aspect of the invention, a method for expressing the live body, in which the internal condition of the live body is predicted on the basis of an electromagnetic field distribution measured on a scalp of the live body, comprises the steps of:

generating a plurality of training data on the basis of an electromagnetic field distribution derived from a head model data designating a head model and a predetermined dipole parameter and an internal condition category data describing a relationship between the active region of a brain and the internal condition of the live body;

deriving a numeric value representative of the active region of the brain from the electromagnetic field distribution in the training data with employing a neural network consisted of an input layer, an output layer and at least one hidden layer and having predetermined coupling coefficients representative of coupling condition between each unit forming each layer;

modifying the coupling coefficient of the neural network for reducing an error between a correct numeric value representative of the correct active region of the brain described in the training data and the numeric value derived by the neural network and indicative of the active region of the brain; and

deriving a numeric value indicative of the active region in the brain by the neural network with taking the electromagnetic field distribution measured by the electromagnetic field distribution measuring means when the error becomes smaller than a predetermined reference value.

According to the third aspect of the invention, a system for predicting an internal condition of a live body for predicting the internal condition of the live body on the basis of an electromagnetic field distribution measured on a scalp of the live body, comprises:

electromagnetic field distribution measuring means for measuring an electromagnetic field distribution on the scalp of the live body;

wavelet transforming means for performing wavelet transformation for the electromagnetic field distribution measured by the electromagnetic field distribution measuring means for eliminating a noise component therefrom;

dipole parameter generating means for generating a dipole parameter employing a random number;

training data generating means for generating a plurality of training data on the basis of an electromagnetic field distribution derived from a head model data designating a head model and a predetermined dipole parameter and an internal condition category data describing a relationship between the active region of a brain and the internal condition of the live body;

inferencing means for deriving a numeric value representative of the active region of the brain from the electromagnetic field distribution in the training data with employing a neural network consisted of an input layer, an output layer and at least one hidden layer and having predetermined coupling coefficients representative of coupling condition between each unit forming each layer;

transforming means for transforming the numeric value representative of the active region in the brain output from the inferencing means into an expression indicative of the internal condition of the live body.

Preferably, the inferencing means has coupling coefficient modifying means for modifying the coupling coefficient of the neural network for reducing an error between a correct numeric value representative of the correct active region of the brain described in the training data and the numeric value derived by the neural network and indicative of the active region of the brain so that a numeric value indicative of the active region in the brain is derived by the neural network with taking the electromagnetic field distribution eliminated the noise when the error becomes smaller than a predetermined reference value.

The live body internal condition predicting system may further comprise parameter number modifying portion for varying number of the coupling coefficients and number of the units forming the hidden layer to minimize a description length of the neural network, which description length is derived by dividing a product of a logarithm of number of the training data and number of parameters to be employed in the neural network by two, and multiplying the quotient with a value derived by subtracting one from a maximum logarithm likelihood. In the alternative, the live body internal condition predicting system may further comprise parameter number modifying portion for varying number of the coupling coefficients and number of the units forming the hidden layer to minimize an Akaike's Information Criterion of the neural network, which Akaike's Information Criterion is derived by summing a value derived by multiplying the number of parameters to be employed in the neural network by 2 and a value derived by multiplying the maximum logarithm likelihood by -2.

According to the fourth aspect of the invention, a system for expressing internal condition of a live body which predicts internal condition of the live body on the basis of an electromagnetic field distribution measured on a scalp of the live body, comprises:

electromagnetic field distribution measuring means for measuring electromagnetic field distribution caused on the scalp of the live body;

training data generating means for generating a plurality of training data on the basis of an electromagnetic field distribution derived from a head model data designating a head model and a predetermined dipole parameter and an internal condition category data describing a relationship between the active region of a brain and the internal condition of the live body;

inferencing means for deriving a numeric value representative of the active region of the brain from the electromagnetic field distribution in the training data with employing a neural network consisted of an input layer, an output layer and at least one hidden layer and having predetermined coupling coefficients representative of coupling condition between each unit forming each layer, the inference means deriving a numeric value indicative of the active region in the brain by the neural network with taking the electromagnetic field distribution measured by the electromagnetic field distribution measuring means; and

transforming means for transforming the numeric value representative of the active region in the brain output from the inferencing means into an expression indicative of the internal condition of the live body.

Preferably, the neural network in the inferencing means has the coupling coefficients at which an error between a correct numeric value representative of the correct active region of the brain described in the training data and the numeric value derived by the neural network and indicative of the active region of the brain becomes minimum. The inferencing means may have a coupling coefficient modifying means for modifying the coupling coefficient of the neural network for reducing an error between a correct numeric value representative of the correct active region of the brain described in the training data and the numeric value derived by the neural network and indicative of the active region of the brain to be minimum so that the inferencing means derives a numeric value indicative of the active region in the brain by the neural network with taking the electromagnetic field distribution measured by the electromagnetic field distribution measuring means when the error becomes smaller than a predetermined reference value.

The live body internal condition expressing system may further comprise normalizing means for normalizing the training data and the measured value of the electromagnetic field distribution measured by the electromagnetic field distribution measuring means.

The live body internal condition expressing system may further comprise noise eliminating means for eliminating noise component from the measured value of the electromagnetic field distribution measured by the electromagnetic field distribution measuring means. In the preferred construction, the noise eliminating means comprises wavelet transforming means for performing wavelet transformation for removing noise from the measured value of the electromagnetic field distribution measured by the electromagnetic field distribution measuring means.

----------------------------------------------

The present invention is applicable in detection of will of the testee who cannot express the will due to handicap or any other reason. Also, it is possible to vary the situation setting in interactive matter with detecting the internal condition of the player in the television game or so forth to make game playing more exciting. Furthermore, the present invention may be use to detect the internal condition of surveillance in criminal investigation.

Although the invention has been illustrated and described with respect to exemplary embodiment thereof, it should be understood by those skilled in the art that the foregoing and various other changes, omissions and additions may be made therein and thereto, without departing from the spirit and scope of the present invention. Therefore, the present invention should not be understood as limited to the specific embodiment set out above but to include all possible embodiments which can be embodies within a scope encompassed and equivalents thereof with respect to the feature set out in the appended claims.