English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

The image captioning task is increasingly prevalent in artificialintelligence applications for medicine. One important application is clinicalreport generation from chest radiographs. The clinical writing of unstructuredreports is time consuming and error-prone. An automated system would improvestandardization, error reduction, time consumption, and medical accessibility.In this paper we demonstrate the importance of domain specific pre-training andpropose a modified transformer architecture for the medical image captioningtask. To accomplish this, we train a series of modified transformers togenerate clinical reports from chest radiograph image input. These modifiedtransformers include: a meshed-memory augmented transformer architecture withvisual extractor using ImageNet pre-trained weights, a meshed-memory augmentedtransformer architecture with visual extractor using CheXpert pre-trainedweights, and a meshed-memory augmented transformer whose encoder is passed theconcatenated embeddings using both ImageNet pre-trained weights and CheXpertpre-trained weights. We use BLEU(1-4), ROUGE-L, CIDEr, and the clinicalCheXbert F1 scores to validate our models and demonstrate competitive scoreswith state of the art models. We provide evidence that ImageNet pre-training isill-suited for the medical image captioning task, especially for less frequentconditions (eg: enlarged cardiomediastinum, lung lesion, pneumothorax).Furthermore, we demonstrate that the double feature model improves performancefor specific medical conditions (edema, consolidation, pneumothorax, supportdevices) and overall CheXbert F1 score, and should be further developed infuture work. Such a double feature model, including both ImageNet pre-trainingas well as domain specific pre-training, could be used in a wide range of imagecaptioning models in medicine.

Understanding Transfer Learning for Chest Radiograph Clinical Report Generation with Modified Transformer Architectures