A BERT-Based Interpretable Deep Learning Model for Medical Diagnosis Assisted by Natural Language Processing

Abstract

The combination of deep learning and natural language processing holds great potential in the medical field. This study aims to explore and develop a deep learning-based natural language processing model to assist in medical diagnosis. We used a large, de-identified dataset of electronic health records, which includes patient complaints, medical histories, and final diagnoses. First, we preprocessed these unstructured text data, for example, by tokenization and removing stop words. Then, we built a deep learning model based on a pre-trained language model like BERT. This model can automatically extract key features from clinical texts and learn the complex relationships between these features and specific diseases. Experimental results show that our proposed model achieves significantly higher accuracy and recall than traditional baseline models on several disease prediction tasks. More importantly, we also introduced an attention mechanism, which gives the model a certain level of interpretability: it can highlight the keywords or phrases that most influence the diagnostic decision. We conclude that deep learning and natural language processing can not only improve the accuracy of disease prediction but also provide valuable references for clinicians, thereby enhancing the overall quality and efficiency of medical services.