نوع مقاله : مقاله کامپیوتر
نویسندگان
1 کارشناسی ارشد، گروه مهندسی پزشکی، دانشکده مهندسی پزشکی و مکانیک، دانشگاه صنعتی همدان، همدان، ایران
2 استادیار گروه مهندسی پزشکی- دانشگاه صنعتی همدان
3 دانشیار، گروه مهندسی پزشکی، دانشکده مهندسی پزشکی و مکانیک، دانشگاه صنعتی همدان، همدان، ایران
چکیده
کلیدواژه‌ها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
Medical images are extensively used in medical science for diagnosis and treatment protocol design. Writing medical reports in text form can be error-prone for inexperienced physicians due to the deep understanding of the disease and its analysis. It is also time-consuming and laborious for experts due to the large number of patients they see in a day. Also, the existence of template reports for physicians can significantly increase their accuracy in diagnosing diseases and reduce errors caused by inattention to details. This research presents a deep learning-based model for the automatic generation of radiology reports. This model is based on a combination of a convolutional recurrent structure and an attention-based architecture called Res-LSTM-Attn. In this model, features are first extracted from medical images using a convolutional residual network, and based on a multi-label word model, a report is predicted. Then, using the LSTM recurrent neural network and multi-head attention layers, the final report is generated. The performance of the proposed models was evaluated based on the BLEU 1-4, ROUGE-L, and CIDEr-D criteria. The results showed that the proposed model outperformed previous studies in generating long reports in terms of CIDEr-D and ROUGE-L metrics, with improvements of 7.2% and 3.2%, respectively.
کلیدواژه‌ها [English]
)