Method for a neural network for representing imaging functions

• US 6,243,489 B1
• Filed: 05/14/1998
• Issued: 06/05/2001
• Est. Priority Date: 05/15/1997
• Status: Expired due to Term

• Current Assignee: Siemens AG
• Original Assignee: Siemens AG
• Inventors: Delong, Wolf-Ruediger
• Primary Examiner(s): Patel, Jay
• Application Number:
  US09/079,132
• Time in Patent Office: 1,118 Days
• Field of Search: 382/100, 382/128, 382/130, 382/131, 382/132, 382/155, 382/156, 382/157, 382/158-160, 382/312, 382/318, 706/15, 706/16, 706/25, 706/26, 706/45, 706/924
• US Class Current: 382/157
• CPC Class Codes: G06F 18/2137 based on criteria of topology preservation, e.g. multidimensional scaling or self-organising maps
• CPC Class Codes: G06N 3/082 modifying the architecture, e.g. adding, deleting or silencing nodes or connections

Abstract

In a method for a self-organizing neural network for representing multidimensional, nonlinear imaging functions onto simpler imaging functions use divider-membranes are employed for achieving an error free representation of the imaging function via the learning sample, allowing for a high level of generalization. Kohonen cell borders coincide with a required imaging function. The neural network can independently determine a number of neurons necessary for an error-free solution of a problem. A readout of the neural network can occur through the calculation of the minimum of the squares of the distances.

Claims

1. A method for creating and analyzing an image, comprising:

providing a device for creating an image;

providing a digital imaging system containing a neural network computer having a neural network therein, and also including a calculating unit for use in a learning phase for the neural network;

providing an image display device for receiving an analyzed image output;

by use of the digital imaging system in a learning phase, representing a multi-dimensional non-linear imaging function of said image in a simpler imaging function by use of at least one divider-membrane for obtaining an error-free representation of said non-linear imaging function by use of a learning sample employed with said neural network thus allowing for a high grade of generalization; and

using said simpler imaging function, outputting an analysis of the image to said image display device.

2. The method according to claim 1 wherein for the representation of the imaging functions, Kohonen cell borders coincide with a required imaging function.

3. The method according to claim 1 wherein the neural network independently determines a number of neurons necessary for an error-free solution of a problem.

4. The method according to claim 1 wherein a readout of the neural network occurs through the calculation of a minimum of the squares of distances.

5. The method according to claim 1 wherein for error-free representation of high-dimensional non-linear imaging functions whose results are allocated to certain regions, performing the further steps of:

determining a number of required regions;

splitting a universal set of all learning data for each region into two respective sets, wherein a first set with internal elements lies inside the required regions, and a second set with external elements lies outside the required region;

determining a minimal distance from each external element to a set of all internal elements, wherein said external element is added to the set of all internal elements as a virtual internal element if a shortest distance is actually to an internal element;

determining a minimal distance from each internal element to a set of all external elements, wherein said internal element is added to the set of all external elements as a virtual external element if a shortest distance is actually to an external element;

deleting all original real internal elements from the set of the internal elements;

deleting all original real external elements from the set of external elements;

inverting designations inside and outside; and

unifying the two sets.

6. The method according to claim 5, including the step of quantizing analog, continuous quantities.

7. The method according to claim 5 including the step of explicitly designating each element as an internal element or an external element when splitting the universal set.

8. A system for creating and analyzing an image, comprising:

an image creation device which outputs an image;

a digital imaging system containing a neural network computer having a neural network therein, and also including a calculating unit for use in a learning phase for the neural network;

an image display device for receiving an analyzed image output;

the digital imaging system, in a learning phase, representing a multi-dimensional non-linear imaging function of said image in a simpler imaging function by use of at least one divider-membrane for obtaining an error-free representation of said non-linear imaging function by use of a learning sample employed with said neural network, thus allowing for a high grade of generalization; and

the digital imaging system, by use of said simpler imaging function, outputting an analysis of the image to said image display device.

9. The system according to claim 8 wherein said digital imaging system segments medical images with a neural network.

10. The system according to claim 8, wherein the neural network is constructed such that it independently determines a number of neurons necessary for an error-free solution of a problem.

11. The system according to claim 8 wherein a circuit arrangement is connected to the neural network which calculates a minimum of squares of distances in a readout from the neural network.

12. The system according to claim 8 wherein said image creation device comprises an x-ray detector for analyzing a patient.