NLP-Driven Image Label Extraction from Radiology Reports: A Review

Abstract

Development of high-performance machine learning models in medical imaging depends on automated extraction of image labels from radiography reports. Medical image annotations need time and sources since they name for knowledge of experienced certified doctors. By using natural Language Processing (NLP) methods, image labels from radiology reports can be automatically generated, therefore saving much manual effort. With chest X-ray datasets and matching radiologist reports, this work investigates algorithms for image and text data. Models comprising VGG16, Xception, ResNet50, NASNetMobile, and an ensemble of Xception and NASNetMobile were examined for picture categorisation; the ensemble model showed exceptional accuracy. Deep learning models including BERT, large language models (LLM), LSTM, GRU, and a hybrid LSTM-GRU model were evaluated in text-based label extraction; the LSTM-GRU combination had the best performance. Integrating deep learning methods for multimodal learning seems to improve label extraction efficiency, therefore facilitating the production of annotated datasets for clinical imaging uses. The results guide automated radiology annotation systems, therefore allowing scalable and accurate training facts production for models of analysis artificial intelligence.

Ähnliche Arbeiten