Contribution of each predictor to each class in Linear Discriminant Analysis

Hello,

I'm interested in knowing the contribution of each predictor to each class in Linear Discriminant Analysis. Matlab's 'fitcdiscr' function affords the boundaries between clasess ('Coeffs' property), which can be used to compute observations' scores in the case of two classes, and, therefore, it is indicative of each variable's contribution in classifying observations to any of those classes. However, for more than 2 classes this assimilation is nolonger valid. ¿Can anyone provide some insight on this question?.

Thanks,

Jorge