Medical imaging report
Data
Indiana University Chest X-ray Acquisition Open-i (Open Access Biomedical Image Search Engine) a subset. The dataset contains 3955 radiology reports and 7470 chest x-rays. Of these, 2314 reports (58.51%) contained abnormalities, and each report consisted of three parts: indication, finding and impression. Finally, a selection of all frontal image data with impressions and findings, which were divided into a training set at a 70% proportion into a training set, 10% validation set and 20% test set.
Train
Set the Multi Attention Layer to 12, model parameters set to 512, image data size set to 224*224, the learning rate is set to 0.0001, the dropout_rate is set to 0.2, and the number of training rounds is set to 50.0. rate is set to 0.2, and the number of training rounds is set to 50.
Citation https://github.com/tengfeixue-victor/Medical-Report-Generation