Support vector machine-based method for analysis of spectral data

Abstract

Support vector machines are used to classify data contained within a structured dataset such as a plurality of signals generated by a spectral analyzer. The signals are pre-processed to ensure alignment of peaks across the spectra. Similarity measures are constructed to provide a basis for comparison of pairs of samples of the signal. A support vector machine is trained to discriminate between different classes of the samples. to identify the most predictive features within the spectra. In a preferred embodiment feature selection is performed to reduce the number of features that must be considered.

1 . A method for classifying samples into one or more classes based upon spectral characteristics of the samples, the method comprising:
 downloading raw spectra obtained from the samples into a storage device in data communication with a processor adapted for executing support vector machines,   aligning the raw spectra using the processor by constructing a similarity measure for comparing pairs of spectra with a baseline spectrum and, based upon the similarity measure, offsetting each example spectrum from the baseline to provide a set of aligned raw spectra;   training at least one support vector machine to discriminate between the plurality of different sample classes using the aligned raw spectra;   processing one or more live spectrum from a subject sample having an unknown characteristic using the trained at least one support vector machine to classify the subject sample as having one of the different characteristics; and   generating an output comprising an identification of the characteristic into which the subject sample is classified for display on a graphical display or for storage in the storage device.   
     
     
         2 . The method of  claim 1 , wherein the baseline is derived from the raw spectral data after applying a baseline subtraction algorithm. 
     
     
         3 . The method of  claim 2 , wherein the baseline subtraction algorithm finds a smooth line that follows the bottom of a curve within the baseline spectrum without rising into the peaks. 
     
     
         4 . The method of  claim 1 , wherein the similarity measure is determined according to the relationship 
       
         
           
             
               
                 
                   S 
                    
                   
                     ( 
                     
                       
                         x 
                         i 
                       
                       , 
                       
                         x 
                         0 
                       
                     
                     ) 
                   
                 
                 = 
                 
                   
                     ∑ 
                     i 
                     
                         
                     
                   
                    
                   
                       
                   
                    
                   
                     
                       
                         ( 
                         
                           
                             x 
                             i 
                           
                           - 
                           
                             x 
                             0 
                           
                         
                         ) 
                       
                       2 
                     
                     
                       
                         
                            
                           
                             x 
                             i 
                           
                            
                         
                         1 
                       
                        
                       
                         
                            
                           
                             x 
                             0 
                           
                            
                         
                         1 
                       
                     
                   
                 
               
               , 
             
           
         
       
       where x i  and x 0  are feature vectors corresponding to peaks of an i th  example spectrum and the baseline spectrum, respectively, and ∥x i ∥ is the l 1  norm, 
       
         
           
             
               
                 
                    
                   
                     x 
                     i 
                   
                    
                 
                 1 
               
               = 
               
                 
                   ∑ 
                   
                     j 
                     = 
                     1 
                   
                   d 
                 
                  
                 
                     
                 
                  
                 
                   
                      
                     
                       x 
                       ij 
                     
                      
                   
                   . 
                 
               
             
           
         
       
     
     
         5 . The method of  claim 1 , wherein the similarity measure is determined according to the relationship S(x i −x 0 )=∥x i −x 0 ∥ 2   2 , where x i  and x 0  are feature vectors corresponding to peaks of an i th  spectrum and the baseline spectrum, respectively. 
     
     
         6 . The method of  claim 1 , wherein the processor is further adapted for executing a feature selection algorithm for identifying peaks within the plurality of spectra to select a subset of spectral peaks that discriminate between the different classes, wherein the feature selection algorithm is selected from SVM-recursive feature elimination and l 0 -norm minimization.