Electronic medical record (EMR) data play a pivotal role in clinical research. However, the utilization of EMR data is severely hindered by the non-standardized expression of medical terminology. This study introduces a clinical terminology standardization model that addresses these challenges. Initially, the model concurrently trains on two tasks: recall and implication number prediction, employing a novel approach that integrates self-attention with augmented features. This integration enhances local features and assimilates global semantic information, specifically for implication number prediction. Subsequently, the model's performance is further improved through cross-attention mechanisms, which amplify feature interactions and facilitate the ranking of candidate terms. Comparative experiments demonstrate that the proposed method significantly outperforms existing state-of-the-art approaches, achieving more accurate and comprehensive clinical terminology standardization.