Named entitiy recognation from radiology reports using natural language processing techniques

Within the scope of medical text mining, it is of great importance to make sense of complex statements, especially in radiology reports. These reports pose challenges for both automated analysis systems and clinical decision processes due to the various medical terms and long contextual relationships they contain. In this study, a hybrid model that can automatically extract entity names from radiology reports is developed. In the model, pre-trained language models such as BERT, BioBERT and ClinicalBERT are integrated with BiLSTM and CRF layers, and hyperparameter optimization is performed with GA and PSO. Within the scope of the method, different model configurations were compared and the combination that provided the highest success was determined. The training and validation processes were performed on the RadGraph dataset. In the evaluations performed on MIMIC-CXR and CheXpert test sets, the GA-optimized ClinicalBERT+BiLSTM+CRF model achieved F1 scores of 98,90% and 97,92%, respectively. The results show that the proposed approach works with high accuracy on clinical texts and can contribute to decision support systems.