| 摘要: |
| 目的 探索并评估一种改良的Transformer模型自主识别乳腺结节并生成文本报告的价值。
方法 构建一种改良的Transformer模型，对832例乳腺结节患者（共计1284个结节）的超声图像进行智能分析，并生成相应文本报告。同时我们引入LGK数据集，将本文中的方法与目前最优秀的的几种隐写术方法进行比较。采用BLEU评分来评估模型性能。
结果 LGK数据集中，本文的模型BLEU-1, BLEU-2, BLEU-3及 BLEU-4评分分别为0.579、0.391、0.288、0.152；在乳腺结节数据集中，本文的模型BLEU-1, BLEU-2, BLEU-3及 BLEU-4评分分别为0.547、0.474、0.352、0.282，均高于其他几种模型。
结论 本文提出的改良Transformer模型能够快速识别乳腺结节并自主生成标准报告，与目前几种优秀的方法相比，获得了良好的BLEU得分，同时该模型在LGK数据集中也有不错的表现，说明本模型具有较高的文本泛化性能。 |
| 关键词: 深度学习，Transformer，乳腺结节，报告生成 |
| DOI： |
| 投稿时间：2023-05-16修订日期：2023-08-06 |
| 基金项目:国家自然科学基金(82071926) |
|
| A transformer model for intelligent generation of ultrasound report of breast nodules |
| wangyi,zhouxinyi,dengdan,xuliming,ranhaitao |
| (the Second Hospital Affiliated of Chongqing Medical University) |
| Abstract: |
| Objective: This study aims to evaluate the effectiveness of a modified transformer model for autonomous recognition of breast nodules and generation of text reports.
Methods: We constructed a transformer model to analyze ultrasound images of breast nodules of 832 patients (1284 nodules in total) and generate corresponding text reports. Additionally, we introduced the LGK dataset and compared our method with several state-of-the-art steganography methods. We evaluated the performance of the model using BLEU score.
Results: In the LGK dataset, the BLEU-1, BLEU-2, BLEU-3, and BLEU-4 scores were 0.579, 0.391, 0.288 and 0.152, respectively. In breast nodule dataset, our model achieved BLEU-1, BLEU-2, BLEU-3, and BLEU-4 scores of 0.547, 0.474, 0.352, and 0.282, respectively, which were higher than several other models.
Conclusion: Our results indicate that the proposed transformer model can quickly identify breast nodules and generate standard reports independently. Compared with the existing state of the art methods, it achieved a higher BLEU score. The model also demonstrated excellent performance in the LGK dataset, indicating high text generalization capability. |
| Key words: deep learning, Transformer, breast nodule, report generation |
