An Interpretable Clinical Transformer Framework (ICTF) for Disease Prediction using EHR

Electronic health records (EHRs) contain large amounts of unstructured clinical texts that require analysis beyond traditional approaches. While transformer models like BioBERT (bidirectional encoder representations from transformers for biomedical text mining) have greatly improved prediction accuracy, their black-box nature reduces trust in clinical applications. This paper provides an overview of recent research papers on clinical NLP tasks and reveals the lack of interpretability in conjunction with data standardization. This study presents ICTF (interpretable clinical transformer framework), which uses a dual pipeline approach to compare machine learning methods with BioBERT. This method also incorporates SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model agnostic Explanations) to provide explanation post-prediction, assisting clinicians in interpreting predictions. The ICTF framework will predict disease labels while providing visualization maps using the MTSamples dataset available on Kaggle.