Presenting the DistilBert Language Model for Obtaining Appropriate Feature Space inReport Generation for Medical Images

increasing progress of science and the combination of two fields of computer vision and natural language processing,researchers have succeeded in producing medical reports, although they contained shortcomings such as the brevityof production reports; But it was a good foundation for helping doctors in the serious matter of treatment and it madethe process of diagnosing and treating diseases faster. Reporting for medical images is one of the emerging fields ofartificial intelligence, which despite extensive research in this field is still full of challenges and has become one ofthe active research fields. Our goal in this research is to take effective action to remove the limitations of medicalreporting so that we can have a more suitable feature space in the reporting process for Chest X-ray medical imagesand more complete and longer medical reports. We have used the DistilBert language model to obtain this appropriatefeature space in the production of medical reports. Then we evaluated our proposed model by using natural languagemeasurement criteria such as BLEU, ROUGE, and METEOR and we verified that our proposed model has workedwell and succeeded in producing more detailed medical reports.

کلیدواژه ها:

attention mechanism ، convolutional neural network ، encoder-decoder framework ، long short termmemory.