Abstract: In order to address the limitations of the current state-of-the-art clinical named entity recognition（CNER） models, which fail to fully exploit the global information and semantic features in text and address issues like character substitutions, we had improved the traditional word embedding model and proposed a novel approach that combines deep convolutional neural networks with bidirectional short-term memory conditional random field（DCNN-BiLSTM-CRF） for clinical text named entity recognition. The enhanced word embedding model integrated the meanings of word roots, phonetics, and characters themselves. It utilized bidirectional encoder representations from Transformers, enabling the word embedding vectors to capture the characteristics of both Chinese characters and clinical text. By introducing DCNN in the task of clinical named entity recognition, we addressed the issue of losing information that cannot be retrieved during CNN prediction. Through the utilization of DCNN, our approach was capable of capturing global information more effectively, capturing weight relationships between characters, and extracting multi-level semantic feature information, thereby improving the accuracy of clinical named entity recognition. We conducted experiments on the CCKS2017 and CCKS2018 datasets. The experimental results show that F1 score of our model improves 0. 48%, 0. 68%, 0. 6%, 0. 58%, 0. 04% and 1. 43%, 2. 36%, 3. 31%, 1. 11%, 0. 17% respectively when compared to the baseline model. Furthermore, to further validate our model, we performed two ablation experiments. Compared to variant model M1, our model achieved F1 score improvements of 0. 79% and 0. 84% on the CCKS2017 and CCKS2018 datasets respectively. Compared to variant model M2, our model achieved F1 score improvements of 0. 53% and 0. 64% on the same datasets. These experimental results confirm the feasibility of the proposed algorithm in this study.