摘要
针对肛管直肠瘘的术前分型临床需求,基于深度学习的自动分类研究存在瘘管形态异质性显著、组织边界模糊等挑战,提出一种融合自适应图通道注意力机制(AGCA)的改进型VGG网络模型(AGCA-VGG),旨在提升神经网络对磁共振影像中肛瘘特征的学习能力。实验采用回顾性研究方法,纳入189例经病理确诊肛瘘患者的T2WI序列磁共振影像数据,依据临床分型标准划分为单纯型(n=93)与复杂型(n=96)。通过随机分层抽样建立150例训练集及39例外部测试集,采用五折交叉验证策略进行模型优化。实验结果表明,AGCA-VGG在外部测试集上取得最优分类性能,其准确率、召回率与精确率分别达到84.62%、85.03%和84.47%,性能优于其它先进的分类方法,能够为临床准确分型提供有效的帮助。
In response to the clinical demands for preoperative classification of anorectal fistulas,the research on automatic classification based on deep learning faces challenges such as significant heterogeneity in fistula morphology and blurred tissue boundaries.An improved VGG network model(AGCA-VGG)integrated with adaptive graph channel attention(AGCA)mechanism is designed to enhance the ability of neural networks to learn the characteristics of anorectal fistula from magnetic resonance images.A retrospective study is conducted on T2WI sequence magnetic resonance imaging data of 189 patients with pathologically confirmed anorectal fistula.According to clinical classification criteria,the patients are categorized into simple type(n=93)and complex type(n=96).A training set containing 150 cases and an external test set consisting of 39 cases are established through stratified random sampling,and a 5-fold cross-validation strategy is adopted for model optimization.Experimental results show that AGCA-VGG achieves the best classification performance on the external test set,with an accuracy of 84.62%,a recall rate of 85.03%,and an accuracy of 84.47%,outperforming other advanced classification methods,and provides effective support for accurate clinical classification.
作者
柏敦徽
耿辰
倪政欣
姚冰
王佳
杨传红
付志辉
戴亚康
BAI Dunhui;GENG Chen;NI Zhengxin;YAO Bing;WANG Jia;YANG Chuanhong;FU Zhihui;DAI Yakang(School of Medical Imaging,Xuzhou Medical University,Xuzhou 221004,China;Suzhou Institute of Biomedical Engineering and Technology,Chinese Academy of Sciences,Suzhou 215163,China;Suzhou TCM Hospital Affiliated to Nanjing University of Chinese Medicine,Suzhou 213000,China)
出处
《中国医学物理学杂志》
2025年第12期1593-1600,共8页
Chinese Journal of Medical Physics
基金
国家自然科学基金(U23A20483,62441114)
苏州市科技计划项目(SKY2023070,SKJYD2021121,SYSD2013123)
苏州市重点实验室(SZS2024007)。
关键词
肛管直肠瘘
深度学习
磁共振
图像分类
anorectal fistula
deep learning
magnetic resonance
image classification
