《计算机应用》唯一官方网站 ›› 2023, Vol. 43 ›› Issue (S2): 41-46.DOI: 10.11772/j.issn.1001-9081.2023020252

• 人工智能 • 上一篇    下一篇

基于多模态融合注意力的肝细胞癌疗效预测方法

文含1,2, 付忠良1,2, 赵莹3, 姚宇1,2, 刘爱连3,4()   

  1. 1.中国科学院 成都计算机应用研究所,成都 610213
    2.中国科学院大学,北京 100049
    3.大连医科大学 附属第一医院,辽宁 大连 116011
    4.大连市医学影像人工智能工程技术研究中心,辽宁 大连 116011
  • 收稿日期:2023-03-13 修回日期:2023-03-24 接受日期:2023-03-29 发布日期:2024-01-09 出版日期:2023-12-31
  • 通讯作者: 刘爱连
  • 作者简介:文含(1993—),男,重庆合川人,博士研究生,主要研究方向:机器学习、医学图像分析与应用
    付忠良(1967—),男,重庆合川人,研究员,硕士,主要研究方向:机器学习
    赵莹(1991—),女,辽宁辽阳人,主治医师,博士,主要研究方向:肝癌多模态MRI的人工智能
    姚宇(1980—),男,四川宜宾人,研究员,博士,主要研究方向:机器学习、模式识别
    刘爱连(1963—),女,辽宁大连人,教授,博士,主要研究方向:双能量CT及MRI新技术的临床应用、医学影像人工智能。
  • 基金资助:
    国家自然科学基金资助项目(61971091);四川省科技计划项目(2022YFS0384);大连市青年科技之星项目(2022RQ074);大连市医学科学研究计划项目(2212011)

Prediction method of hepatocellular carcinoma efficacy based on multimodal fusion attention

Han WEN1,2, Zhongliang FU1,2, Ying ZHAO3, Yu YAO1,2, Ailian LIU3,4()   

  1. 1.Chengdu Institute of Computer Applications,Chinese Academy of Sciences,Chengdu Sichuan 610213,China
    2.University of Chinese Academy of Sciences,Beijing 100049,China
    3.The First Affiliated Hospital of Dalian Medical University,Dalian Liaoning 116011,China
    4.Dalian Medical Imaging Artificial Intelligence Engineering Technology Research Center,Dalian Liaoning 116011,China
  • Received:2023-03-13 Revised:2023-03-24 Accepted:2023-03-29 Online:2024-01-09 Published:2023-12-31
  • Contact: Ailian LIU

摘要:

针对传统方法预测肝细胞癌(HCC)疗效通常只采用图像信息、临床信息或基因信息等单一模态信息的问题,提出一种基于多模态融合注意力的HCC疗效预测方法。首先,使用残差网络(ResNet)提取图像特征和多层感知机提取临床特征和常规放射学特征;其次,构建一个多模态融合注意力模块,通过计算不同模态特征之间的相关性有效地融合图像特征、临床特征和常规放射学特征;最后,通过一个分类网络实现对HCC患者疗效的准确分类预测。实验结果表明,与单一模态预测疗效的方法相比,所提方法的准确率在实验数据集上提升了5.95个百分点,验证了所提方法能显著改善HCC患者疗效预测的结果。

关键词: 多模态融合, 注意力机制, 肝细胞癌, 疗效预测, 深度学习

Abstract:

A multimodal fusion attention-based HepatoCellular Carcinoma (HCC) efficacy prediction method was proposed to address the problem that traditional methods for predicting the efficacy of HCC usually use single modal information such as image information, clinical information, or genetic information. Firstly, Residual Network (ResNet) was used to extract image features, and a multi-layer perceptron was employed to extract clinical and conventional radiological features. Then, a multimodal fusion attention module was constructed to efficiently fuse image features, clinical features, and conventional radiological features by calculating the correlation between different modal features. Finally, a classification network was used to accurately predict and classify HCC patient outcomes. The experimental results show that compared to the single-modality method for predicting efficacy, the accuracy of the proposed method increases by 5.95 percentage points on experimental dataset, confirming that the proposed method can significantly improve the prediction results of HCC patient efficacy.

Key words: multimodal fusion, attention mechanism, HepatoCellular Carcinoma (HCC), efficacy prediction, deep learning

中图分类号: