Method and magnetic resonance apparatus for automated analysis of the raw data of a spectrum

1. A computerized method for automated analysis of raw data representing a spectrum acquired from an individual patient, said method comprising:

providing said raw data representing a spectrum acquired from an individual patient to a processor and, in said processor, applying at least one post-processing procedure to said raw data, thereby obtaining a processed spectrum of said individual patient;

in said processor, generating a feature vector from numerical values of the processed spectrum of said individual patient;

in said processor, automatically allocating said feature vector to one group of a plurality of groups in said knowledge base, where said knowledge bases comprises training feature vectors of tissues classified according to in vitro measurement results of said tissues, with each of training feature vector of each tissue is generated from a processed spectrum from a training patient comprising said tissue; and

in said processor, based on said one group to which said feature vector is allocated, identifying a material from which said spectrum of said individual patient originates and emitting an electronic signal at an output of said processor indicating said material.

2. A method as claimed in claim 1 comprising acquiring said raw data representing each spectrum as raw magnetic resonance data representing each spectrum.

3. A method as claimed in claim 2 comprising acquiring said raw magnetic resonance data as raw magnetic resonance data representing a proton spectrum.

4. A method as claimed in claim 1 comprising acquiring said raw data representing each spectrum by a data acquisition procedure selected from the group consisting of an SVS (Single Voxel Spectroscopy) measurement and a CSI (Chemical Shift Imaging) measurement.

5. A method as claimed in claim 1 comprising determining each feature vector by binning in said processor.

6. A method as claimed in claim 1 comprising determining each feature vector by executing a principal component analysis in said processor.

7. A method as claimed in claim 1 comprising determining each feature vector by executing a procedure in said processor selected from the group consisting of an independency analysis and a Sammon mapping.

8. A method as claimed in claim 1 comprising allocating said feature vector of said individual patient to said one of said groups using a linear classifier.

9. A method as claimed in claim 8 comprising, in said processor, using a support vector machine as said linear classifier.

10. A method as claimed in claim 1 comprising allocating said feature vector of said individual patient to said one of said groups using a non-linear classifier.

11. A method as claimed in claim 10 comprising using a support vector machine with an RBF (Radial Basis Function) kernel in said processor as said non-linear classifier.

12. A method as claimed in claim 1 comprising, for the raw data representing each spectrum implementing a water suppression using a Hankel singular value decomposition in said processor, before determining the feature vector for that respective spectrum.

13. A method as claimed in claim 1 , said method further comprising:

acquiring raw data representing a spectrum of a selected region of each of a plurality of patients in a patient population, and also acquiring an in vitro measurement from the selected region of each patient in said patient population;

providing said raw data, and an electronic designation of said in vitro measurement, for each patient in said patient population to a processor;

in said processor, generating a feature vector for each spectrum from numerical values of that respective spectrum represented by the raw data thereof, and storing said feature vectors and said in vitro measurements in a memory of said processor, and thereby forming a knowledge base in said memory; and

in a training phase for said knowledge base, classifying, in said processor, the respective feature vectors of the spectra in said knowledge base into a plurality of groups dependent on said biopsy samples in said knowledge base.

14. A method as claimed in claim 1 wherein said in vitro measurement result is a biopsy sample.

15. A magnetic resonance apparatus comprising:

a control unit configured to operate said magnetic resonance apparatus to acquire raw MR data representing a spectrum in an individual patient situated in the apparatus;

a computerized processor supplied with said raw MR data, said computerized processor being configured to apply at least one post-processing procedure to said raw MR data, thereby obtaining a processed spectrum of said individual patient;

said processor being configured to generate a feature vector from numerical values of the processed spectrum to a feature vector;

said computerized processor being configured to allocate said feature vector to one group among a plurality of groups in said knowledge base, where said knowledge bases comprises training feature vectors of tissues classified according to in vitro measurement results of said tissues, with each of training feature vector of each tissue is generated from a processed spectrum from a training patient comprising said tissue; and

said computerized processor being configured to identify said material, based on said one group to which said feature vector is allocated, and to emit an electronic signal at an output of said computerized processor indicating said material.