Measuring semantic similarity among words, terminologies, and concepts is a fundamental research problem with significant applications in the field of natural language processing (NLP). To date, accurately gauging semantic similarity among terms within a specialized domain remains a substantial challenge. The issue of terminology similarity is critically important in domains such as healthcare research and medical data. In the context of Vietnamese NLP, this problem currently receives limited attention and study. This study introduces a novel technique for assessing semantic similarity among medical terms by integrating word embedding models with an optimization algorithm applied to a weighted graph. Experimental results demonstrate that the proposed technique achieves superior performance compared to previous methods based on WordNet and word embeddings.