Medical Named Entity Recognition (NER) plays a crucial role in attaining precise patient portraits as well as providing support for intelligent diagnosis and treatment decisions. Federated Learning (FL) enables collaborative modeling and training across multiple endpoints without exposing the original data. However, the statistical heterogeneity exhibited by clinical medical text records poses a challenge for FL methods to support the training of NER models in such scenarios. We propose a Federated Contrast Enhancement (FedCE) method for NER to address the challenges faced by non-large-scale pre-trained models in FL for label-heterogeneous. The method leverages a multi-view encoder structure to capture both global and local semantic information, and employs contrastive learning to enhance the interoperability of global knowledge and local context. We evaluate the performance of the FedCE method on three real-world clinical record datasets. We investigate the impact of factors, such as pooling methods, maximum input text length, and training rounds on FedCE. Additionally, we assess how well FedCE adapts to the base NER models and evaluate its generalization performance. The experimental results show that the FedCE method has obvious advantages and can be effectively applied to various basic models, which is of great theoretical and practical significance for advancing FL in healthcare settings.

Clinical diagnosis for complex disease conditions is a complicated decision process involving systematic inference and differentiation. Artificial Intelligence (AI) models have been a widely established approach to help improve the efficiency of various kinds of clinical decision tasks (e.g., diagnosis, treatment, and prognosis). However, due to the critical requirement of time efficiency, lack of sufficient information, and high probability of comorbid diseases in Outpatient and Emergency Settings (OES), it is still challenging to build clinically feasible AI models using the free text clinical records in OES for complex disease conditions, such as neurosurgery. Here we propose an AI diagnosis model, named LLM4DEU, for neurosurgery disease differentiations by fine-tuning a large language model (i.e., ChatGLM) using the Department of Neurosurgery, the Beijing Tiantan Hospital OES electronic health records. LLM4DEU obtained state-of-the-art performance on clinical diagnosis with a F1 score of 78.53%, which is superior to five well-known baselines (including deep learning models). In addition, we evaluated the actual performance of the model by case studies on the diagnosis of specific neurosurgical diseases (e.g., subdural hematoma, cerebral hemorrhage, and cerebral infarction). The experimental results show that the LLM4DEU model has significant advantages in diagnosing low-incidence disease conditions, and comparative analyses with clinical experts confirm the predictive power of the model in neurosurgical diagnosis.