Automatic Chest X-ray Report Generation Using Machine Learning

Abstract

In this study, a deep neural network is suggested for the automatic creation of precise radiologist reports from chest X-ray pictures. The proposed network responds to the need for medical image captioning by learning to extract key features from the image and creating tag embeddings for each patient's X-ray images. Medical image captioning demands coherence and high accuracy in identifying abnormalities and extracting information. For a finer representation, the network encodes picture and tag features with self-attention using transformers for self and cross-attention. The generated findings are obtained by applying cross-attention between the image and tag features and the input sequence, while the impressions are generated by applying cross-attention between the generated findings and the input sequence. Evaluation using a publicly available dataset shows that the proposed network can generate a readable radiology report with a higher BLEU score compared to the state-of-the-art approach. This study demonstrates the potential of deep neural networks in automating radiology report generation and improving healthcare facilities.