JPH03114435A - Diagnostic system - Google Patents

Abstract

PURPOSE:To achieve high accuracy of reasoning by providing a plurality of neural network circuits with different characteristics on reasoning process, providing weighting on the outputs of the plurality of neural network circuits and putting outputs together to obtain an output information to a plurality of input information elements. CONSTITUTION:A data input part 10 capable of taking a plurality of input information elements in gives a plurality of a first neural network circuit 20 and a second neural network circuit 30. The first and the second neural network circuits 20 and 30 have each different sensitivity characteristics to input information which are characteristics of its reasoning process. A synthetic judgement part 40 gives a synthetic judgement with a proper weighting with helps of a knowledge data base 50 and a reasoning engine 60 to two reasoning results output from the first and the second neural network circuits 20 and 30 and the reasoning result is output in a form of an image or a hardcopy through an interface. On a synthetic process of reasoning results, an experience information, etc., can be reflected with helps of an image data base 50 and a reasoning engine 60 and therefore it is possible to achieve to make reasoning with high accuracy.

Description

DETAILED DESCRIPTION OF THE INVENTION [Object of the Invention] (Industrial Application Field) The present invention relates to a diagnostic system using a neural network circuit that obtains one output information from a plurality of input information through an inference process.

(Prior Art) Neural network circuits, which have been attracting attention recently, are information processing systems based on the human brain nerve network, and are composed of a combination of many processing units made of hardware and software. There is. Here, the processing unit or neural network circuit operates to obtain one piece of output information from a plurality of pieces of input information through an inference process.

Conventionally, a diagnostic system using this type of neural network circuit has been constructed by incorporating one neural network circuit into the diagnostic system.

(Problems to be Solved by the Invention) Here, the characteristics of the inference process in the neural network circuit are fixed in terms of hardware or software.

Therefore, for example, even if multiple pieces of information are obtained in an environment where the phenomenon inferred as an output result can be confirmed empirically, the characteristics of the inference process of the neural network circuit are fixed. Therefore, there is a problem in that the experience information is not equally utilized and inferences are made based on fixed system characteristics. This is nothing but a decrease in the accuracy of inference.

SUMMARY OF THE INVENTION Therefore, an object of the present invention is to provide a diagnostic system using a neural network circuit that can achieve high inference accuracy.

[Structure of the Invention (Means for Solving the Problems) The present invention has a structure in which the following means are taken to solve the above problems and achieve the objects.

That is, the present invention provides a diagnosis system using a neural network circuit that obtains one piece of output information from a plurality of input information through an inference process, which has a plurality of neural network circuits with different inference process characteristics, - It is characterized by weighting the outputs of the network circuits and combining the respective outputs to obtain one output information for a plurality of input information.

(Operation) According to such a configuration, the characteristics of the inference process of each of the plurality of neural network circuits are different, and due to the different conditions of these characteristics, it is difficult to reflect empirical information etc. in the inference process. Therefore, higher accuracy of inference can be achieved than in the conventional case of a single characteristic.

(Embodiment) An embodiment of the diagnostic system according to the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram of a diagnostic system according to this embodiment, and FIG. 2 is a diagram showing a processing flow when diagnosing a disease using the system shown in FIG.

As shown in FIG. 1, the data input unit 10 that can take in a plurality of pieces of input information includes a first neural network circuit 20 and a second neural network circuit 30.
Give multiple input information to each.

The first neural network circuit 20 and the second neural network circuit 30 have different sensitivity characteristics to input information, which are characteristics of their inference process.

The comprehensive judgment unit 40 determines the first. Appropriate weighting is applied to the two inference results output from the second neural network circuit 20.30 with the support of the knowledge database 50 and the inference engine 60, and a comprehensive judgment is made. The inference results are output in the form of images or hard copies.

Here, the knowledge database 50 stores, for example, disease information, patient information, and the like. In addition, the inference engine 60 selects information from the knowledge database 50 to support processing in the two inference processes output from the first and second neural network circuits 20 and 30 in the comprehensive judgment unit 40. It has become a thing.

Next, the operation of the inference system of this embodiment configured as described above will be explained in the case of disease inference. As a specific example, an example in which alcoholic hepatitis is diagnosed based on test data collected from a patient will be described with reference to FIG.

First, test data from the patient including γ-GTP value and GO
Collect the T value, GPT value, etc., and input it to the data input section 10 in FIG.
is input to the first neural network circuit 20 and the second neural network circuit 30 as step S1.

Here, it is assumed that the sensitivity characteristic of the first neural network circuit 20 is ω (0≦ω≦1), and the sensitivity characteristic of the second neural network circuit 30 is (1−ω).

Then, in step S2, the first neural network circuit 20 determines that it is "alcoholic hepatitis" based on the test data (γ-GTP value, GOT value, GPT value), that is, the sensitivity characteristic ω Quantitatively determined. Further, in step S3, the second neural network circuit 30 outputs the test data (γ-GTP value).

"It is not alcoholic hepatitis", that is, negativity, is quantitatively determined based on the sensitivity characteristic (1-ω) based on the GOT value and GPT value.

Next, in step S4, the general judgment unit 4o uses the knowledge database 50 for the above-mentioned affirmative judgment and negative judgment.
The weighting process is performed with the support of the inference engine 60, and the inference result is outputted through the interface 70 in step S5.

As described above, in this embodiment, the characteristics of the inference processes of the two neural network circuits 20 and 30 are different. As a result, it is possible to take into account positivity and negativity in the process of inference based on conditions with different characteristics, and in addition, in the process of synthesizing each inference result, experience with the support of the knowledge database 50 and the inference engine 60 is used. Since information etc. can be reflected, higher accuracy of inference can be achieved than in the conventional case of a single characteristic.

Although the above embodiment describes the case where two neural network circuits are used, it is also possible to use three or more neural network circuits with different characteristics.

[Effects of the Invention] As described above, the present invention has a plurality of neural network circuits with different inference process characteristics, weights the outputs of the plurality of neural network circuits, and integrates the respective outputs to generate a plurality of neural network circuits. By obtaining one output information for each input information, it is possible to reflect experience information, etc. in the inference process based on the different conditions of this characteristic. High precision can be achieved.

Therefore, according to the present invention, it is possible to provide a diagnostic system using a neural network circuit that can achieve high inference accuracy.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the configuration of an embodiment of a diagnostic system according to the present invention, and FIG. 2 is a flowchart showing the operation of the embodiment. 10...Data input section, 20.30...Neural network circuit, 40...Comprehensive judgment section, 50...
・Knowledge database, 60...Inference engine, 70・
··interface.

Claims (1)

[Claims] A diagnostic system using a neural network circuit that obtains one output information from multiple input information through an inference process has multiple neural network circuits with different inference process characteristics, and these multiple neural network circuits have different inference process characteristics.
A diagnostic system characterized by weighting the outputs of network circuits and integrating the respective outputs to obtain one output information for a plurality of input information.