多尺度融合注意力机制的胆囊癌显微高光谱图像分类  

A micro-hyperspectral image classification method of gallbladder cancer based on multi-scale fusion attention mechanism

作　　者：高红民[1,3] 朱敏 曹雪莹[1,3] 李臣明[1] 刘芹 许佩佩[2,3] Gao Hongmin;Zhu Min;Cao Xueying;Li Chenming;Liu Qin;Xu Peipei(College of Computer and Information Engineering,Department of Information,Hohai University,Nanjing 211100,China;Department of Hematology,Drum Tower Hospital,School of Medicine,Nanjing University,Nanjing 211108,China;Shanghai Key Laboratory of Multidimensional Information Processing,East China Normal University,Shanghai 200241,China;Department of Oncology,Drum Tower Hospital,School of Medicine,Nanjing University,Nanjing 211108,China)

机构地区：[1]河海大学信息学部计算机与信息学院,南京211100 [2]南京大学医学院附属鼓楼医院血液科,南京211108 [3]华东师范大学上海市多维度信息处理重点实验室,上海200241 [4]南京大学医学院附属鼓楼医院肿瘤科,南京211108

出　　处：《中国图象图形学报》2023年第4期1173-1185,共13页Journal of Image and Graphics

基　　金：南京市卫生科技发展专项资金项目(YKK22087);上海市多维信息处理重点实验室开放课题基金项目。

摘　　要：目的胆囊癌作为胆道系统中一种恶性程度极高的肿瘤,早期诊断困难、预后极差,因此准确鉴别胆囊病变对早期发现胆囊癌具有重要意义。目前胆囊癌的诊断主要依赖于超声、CT(computed tomography)等传统影像学方法,但准确性较低。显微高光谱能够在获取生物组织图像信息的同时从生化角度对生物组织进行分析,从而实现对胆囊癌的早期诊断,相比于传统医学图像更具优势。因此,本文基于胆囊癌显微高光谱图像设计了一种基于多尺度融合注意力机制的网络模型,以提高分类准确率。方法提出多尺度融合注意力模块(multiscale squeeze-andexcitation-residual,MSE-Res)。MSE-Res模块引入改进的多尺度特征提取模块实现通道维上特征的融合,用一个最大池化层和一个上采样层代替1×1的卷积层来提取图像的显著特征。为了弥补池化层丢失的局部信息,在跳跃连接中加入一个1×1的卷积层。在多尺度特征提取模块后,引入注意力机制来学习不同通道间特征的相关性,实现通道间特征的融合,并通过残差连接使网络在提取图像深层特征的同时避免出现过拟合现象。结果在胆囊癌高光谱数据集上进行实验,本文模型的总体分类精度、平均分类精度和Kappa系数分别为99.599%、99.546%和0.990,性能优于SE-ResNet(squeeze-and-excitation-residualnetwork)和Inception-SE-ResNet(inception-squeeze-andexcitation-residual network)。结论本文提出的MSE-ResNet能够有效利用高光谱图像的空间信息和光谱信息,提高胆囊癌分类准确率,在对胆囊癌的医学诊断方面具有一定的研究价值和现实意义。Objective Gallbladder carcinoma is recognized as one of the most malignant tumors in relevant to biliary sys⁃tem.Its prognosis is extremely poor,and only 6 months of overall average.It is challenged for missed diagnose because of the lack of typical clinical manifestations in early stage of gallbladder cancer.To clarify gallbladder lesions for early detec⁃tion of gallbladder carcinoma accurately,current gallbladder cancer-related diagnosis is mainly focused on the interpreta⁃tion of digital pathological section images(such as b-ultrasound,computed tomography(CT),magnetic resonance imaging(MRI),etc.)in terms of the computer-aided diagnosis(CAD).However,the accuracy is quite lower because the molecu⁃lar level information of diseased organs cannot be obtained.Micro-hyperspectral technology can be incorporated the fea⁃tures of spectral analysis and optical imaging,and it can obtain the chemical composition and physical features for biologi⁃cal tissue samples at the same time.The changes of physical attributes of cancerous tissue may not be clear in the early stage,but the changes of chemical factors like its composition,structure and content can be reflected by spectral informa⁃tion.Therefore,micro hyperspectral imaging has its potentials to achieve the early diagnosis of cancer more accurately.Micro-hyperspectral technology,as a special optical diagnosis technology,can provide an effective auxiliary diagnosis method for clinical research.However,it can provide richer spectral information but large amount of data and information redundancy are increased.To develop an improved accuracy detection method and use the rich spatial and hyperspectral information effectively,we design a multi-scale fusion attention mechanism-relevant network model for gallbladder canceroriented classification accuracy optimization.Method The multiscale squeeze-and-excitation-residual(MSE-Res)can be used to realize the fusion of multiscale features between channel dimensions.First,an improved multi-scale feature extrac⁃tion mo

关 键 词：胆囊癌高光谱图像 多尺度特征融合 残差网络 图像分类 SE模块 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]