With the development of clinical informatization technology, a large amount of data resources has been accumulated in the medical field, among which Athletes’ electronic medical records (EMRs) are one of the important data sources of clinical informatization and contain rich medical knowledge, how to obtain valuable biometric features from these data has become the basis of medical intelligence research. Therefore, this paper takes structured text as the entry point, and firstly, the research status of information extraction is elaborated. Moreover, in the work of e-medical biometrics recognition, medical text is added to the language model BERT for pre-training and fine-tuning, and a multi-head self-attention mechanism is introduced to incorporate a bidirectional LSTM model for feature extraction, and biometric features are extracted by using a stochastic conditional field as a classification constraint. This design bypasses the character-level image segmentation step for text line images, thus avoiding the overall accuracy degradation caused by the backward accumulation of errors in character segmentation. Finally, the experimental results show that the model can effectively accomplish the related biological feature extraction tasks.