S108 - Transforming Radiology Workflows: Pretraining for Automated Chest X-ray Report Generation
Yuhang Jiang, Shashank Gupta, Abdullah Al Zubaer Imran
Automated chest X-ray report generation using machine learning has emerged as a promising technology for improving the accuracy and efficiency of chest X-ray interpretation. In this paper, we present a novel approach for automated report generation that combines the power of vision transformers for image information encoding and PubMedBERT for text decoding. Our model extracts image features using a vision transformer and text features using PubMedBERT. The encoded features are then fed into a text decoder to generate standardized reports. We trained our model on a dataset of chest X-rays and corresponding report findings (IU dataset) and evaluated its performance on a small subset of the MIMIC-CXR dataset.
Schedule: Monday, July 10: Posters — 11:00–12:00 & 15:00–16:00
Poster location: M49