In this paper a new similarity index for neurograms is proposed. This index is inspired by the Needleman-Wunsch algorithm which determines the minimum number of operations to transform a vector into another in terms of insertions, deletions and substitutions. The Needleman-Wunsch algorithm can be extended to the two dimensional case and the number of transformations required to change a matrix into another is used to define a measure of similarity. This similarity measure is applied to neurograms and optimized to perform prediction of speech intelligibility in noise. Word recognition scores for for speech samples in noise are evaluated using the proposed similarity index, showing a clear improvement in speech intelligibility estimation with respect to other neurogram similarity metrics in the literature. The proposed similarity index is not restricted to a certain time resolution and could serve to evaluate neurogram similarity with respect to temporal fine structure in future.