In this article, we present a simple and effective approach for extracting bilingual lexicon from comparable corpora enhanced with parallel corpora. We make use of structural characteristics of the documents comprising the comparable corpus to extract parallel sentences with a high degree of quality. We then use state-of-the-art techniques to build a specialized bilingual lexicon from these sentences and evaluate the contribution of this lexicon when added to the comparable corpus-based alignment technique. Finally, the value of this approach is demonstrated by the improvement of translation accuracy for medical words.