Clinical Reports Classification using Fine-Tuned DistilBERT Model

Due to the fast-growing nature of healthcare data, information extraction from clinical reports is becoming challenge. Because clinical reports are covering a major portion of patient care. These reports are unstructured and densely composed with specialized medical terminology which makes it quite difficult for efficient classification and retrieval. This study deploys a novel fine-tuned Transformer model, DistilBERT, to automatically classify clinical reports. Various preprocessing steps are conducted, to normalize the data and remove the redundant information. Various padding approaches are applied and hyperparameters are tuned to obtained better results. The proposed model is validated using the i2b2 dataset. The proposed model achieves the highest accuracy of 0.92%. The obtained results are better compared to state-of-the art works. This study provides a scalable and robust solution for classifying clinical reports, which contributes to the field of medical informatics by assisting in reducing the workload on health professionals due to manual triage based only on reading list table information.