Multi Label Clinical Text Classification Using Attention Mechanism

Clinical text comprises of structured and unstructured text. Most clinical text is unstructured and rich in information that can be leveraged. Consequently, text classification, a prevalent application in natural language processing, is employed, involving the use of deep learning models such as convolutional neural networks and bidirectional long short-term memory. However, these models may still fall short in discerning and prioritizing critical terms for accurate classification. This shortcoming can compromise their performance in processing diverse biomedical texts found in literature and clinical notes. Thus, the innovation of this study lies in integrating attention mechanisms to augment deep learning models by emphasizing relevant information. Consequently, the model becomes adept at dynamically focusing on significant biomedical terms, thereby refining its comprehension and processing of input data. This project will focus on the clinical notes found in the MIMIC-III database. The implementation of attention mechanism showed improvement of f1-score in 7 out of 10 ICD-9 labels. Beyond academia, this breakthrough holds promise for healthcare, enhancing medical research and clinical practices by facilitating superior knowledge extraction from biomedical data, potentially elevating patient care and fostering advancements in medical science.