Vision Transformer and Language Model Based Radiology Report Generation

Recent advancements in transformers exploited computer vision problems which results state-of-the-art models. Transformer-based models various sequence prediction tasks such as language translation, sentiment classification, and caption generation have shown remarkable performance. Auto report scenarios medical imaging through is one of the applied for strong social impact. In these models, convolution neural networks been used encoder to gain spatial information recurrent are decoder generate or report. However, using transformer architecture writing task still unexplored. this research, we explored effect losing biasness by pre-trained vanilla image combine it with different decoder. order evaluate proposed methodology, Indiana University Chest X-Rays dataset where ablation study also conducted respect evaluations. The comparative analysis shows that methodology has represented performance when compared existing techniques terms parameters.