文献速递妇产影像技术与安全US

基于深度学习的三维超声先天性子宫畸形智能诊断框架 CUA-Net

Deep Learning-based Intelligent Diagnosis of Congenital Uterine Anomalies in 3-D Ultrasound
RSS 订阅 · Ultrasound in Medicine and Biology · 2026-09-12

本研究开发并验证了用于三维超声先天性子宫畸形自动分类的深度学习框架 CUA-Net,无需冠状面重建。CUA-Net 在内部测试集准确率 93.88%、外部测试集准确率 91.52%,表现优于初级超声医师,多数指标与高年资超声医师相当。

要点

  • CUA-Net 可直接对三维超声容积分类 CUA,无需冠状面重建,有望简化流程。
  • 模型在外部测试集保持较高准确率,提示一定泛化能力。
  • 在多数评估指标上达到与高年资超声医师相当水平,优于初级医师。

核心结论

  • CUA-Net 在常见先天性子宫畸形分类中显示良好准确性与泛化能力。
  • 该模型对少见畸形类型的识别具有初步潜力。
  • 该框架可能有助于优化临床工作流程并支持更标准化的诊断。

研究设计与样本

  • 模型基于 3-D ResNet-18 构建。
  • 采用动态数据重采样策略以缓解数据不平衡。
  • 引入困难样本挖掘技术通过损失调整充分学习难诊断病例。
  • 提出自监督重建策略以全面探索三维容积信息,并使用在线数据增强优化错误预测并提升泛化。
  • 使用内部和外部测试集比较 CUA-Net 与其他深度学习方法及初/高年资超声医师表现。

诊断标准与关键数值

  • 内部测试集:准确率 93.88%,精确率 87.01%,召回率 95.92%,F1-score 88.09%,micro-AUC 0.9982,macro-AUC 0.9997。
  • 外部测试集:准确率 91.52%,精确率 83.27%,召回率 88.63%,F1-score 81.49%,micro-AUC 0.9945,macro-AUC 0.9990。
  • CUA-Net 在所有性能指标上优于初级超声医师,在多数指标上达到与高年资超声医师相当的水平。

临床应用与阅片要点

  • CUA-Net 不需要冠状面重建即可对三维超声容积进行分类,有助减少操作依赖。
  • 可作为 CUA 筛查或初诊辅助工具,尤其在初级医师经验不足时提供参考。

局限性

  • 原文提及 CUA-Net 对少见畸形类型的识别能力仍为初步潜力,提示仍需进一步验证。
英文原文(摘要)· 点击展开

OBJECTIVE: This study aimed to develop an intelligent framework, termed CUA-Net, for the automated classification of congenital uterine anomalies (CUAs) without the requirement of coronal plane reconstruction, and to evaluate its clinical applicability.

METHODS: CUA-Net was built on 3-D ResNet-18, equipped with a dynamic data re-sampling strategy to mitigate the data imbalance issue and a hard sample mining technique to fully learn from difficult cases through loss adjustment. We further proposed self-supervised reconstruction to comprehensively explore volumes and online data augmentation to refine incorrect predictions and enhance the model's generalization. We used internal and external test sets to compare CUA-Net with different deep learning methods as well as junior/senior sonographers. The evaluation metrics included accuracy, precision, recall, F1-score, micro-area under the curve (AUC) and macro-AUC.

RESULTS: CUA-Net exhibited satisfactory performance in both the internal and external test sets. In the internal cohort, the model achieved an accuracy of 93.88%, precision of 87.01%, recall of 95.92%, F1-score of 88.09%, micro-AUC of 0.9982 and macro-AUC of 0.9997. In the external set, it maintained good performance with an accuracy of 91.52%, precision of 83.27%, recall of 88.63%, F1-score of 81.49%, micro-AUC of 0.9945 and macro-AUC of 0.9990. CUA-Net outperformed junior sonographers across all performance indicators and achieved performance comparable to that of senior sonographers across most metrics.

CONCLUSION: CUA-Net demonstrates favorable accuracy and generalizability in classifying common CUA categories while showing preliminary potential for recognizing less prevalent anomalies. These capabilities may help to optimize clinical workflows and support more standardized diagnosis.

来源:Ultrasound in Medicine and Biology,查看原文。版权归原作者所有。

引用

  1. Ultrasound in Medicine and Biology:Deep Learning-based Intelligent Diagnosis of Congenital Uterine Anomalies in 3-D Ultrasound
  2. Xu, Huang, Deng, et al. Deep Learning-based Intelligent Diagnosis of Congenital Uterine Anomalies in 3-D Ultrasound. Ultrasound Med Biol. 2026. PMID: 42731934. doi:10.1016/j.ultrasmedbio.2026.08.015

供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。