In this paper we present the results obtained during BioASQ Task 11B Phase A on document ranking. We proposed a two-stage retrieval approach composed of a retriever and a reranker. The former is based on BM25 and developed with Pyserini. The latter is an implementation of a BERT cross-encoder named CEDR. It tackles the issue of input length limitation and takes advantage of the word embeddings in output of the model to compute a relevance score. We investigate a strategy to integrate biomedical thesaurus (MeSH) knowledge into this model.