基于肿瘤形状特征与点云方法的PET-CT多模态图像神经母细胞瘤分割  

Segmentation of Neuroblastoma with PET-CT Multimodal Images Based on Tumor Shape Characteristics and Point Clouds Method

作　　者：周维钦 王朝立[1] 孙占全 陈素芸[2] 李超[2] 傅宏亮[2] 刘晓虹[3] ZHOU Weiqin;WANG Chaoli;SUN Zhanquan;CHEN Suyun;LI Chao;FU Hongliang;LIU Xiaohong(College of Optoelectronic Information and Computer Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China;Department of Nuclear Medicine,Xinhua Hospital Affiliated to Shanghai Jiaotong University School of Medicine,Shanghai 200092,China;Departmentof Radiology,Shanghai Eighth People's Hospital,Shanghai 200235,China)

机构地区：[1]上海理工大学光电信息与计算机工程学院,上海200093 [2]上海交通大学医学院附属新华医院核医学科,上海200092 [3]上海市第八人民医院放射科,上海200235

出　　处：《软件导刊》2024年第3期128-133,共6页Software Guide

基　　金：国家自然科学基金项目(6217323);国防科工局基础研究项目(JCKY2019413D001)。

摘　　要：通过智能学习方法对PET-CT图像进行肿瘤自动分割,是辅助医生制定诊疗计划的重要研究领域。PET-CT图像兼具PET和CT两种模态优点,传统方法大多只简单的将两种模态的图像进行配准和融合后提取特征,忽略了神经母细胞瘤具有肿瘤边界轮廓不规则的特点。为此,提出一种两阶段的自动分割框架结构模型。首先,利用3D卷积神经网络定位肿瘤位置;然后在分割出的肿瘤区域附近生成多模态点云数据,并提取肿瘤的形状轮廓特征;最后,将两个网络提取的特征进行融合用来预测最终分割结果。在自有数据集和公共数据集上,将所提模型与其他多模态方法进行比较实验,实验结果验证了所提模型的优越性与有效性。以期为研究神经母细胞瘤分割的研究人员提供参考与借鉴。The automatic tumor segmentation of PET-CT images through intelligent learning methods is an important research field to assist doctors in formulating diagnosis and treatment plans.PET-CT images have the advantages of both PET and CT modalities.Traditional methods mostly simply register and fuse the images of the two modalities to extract features,ignoring the irregular tumor boundary contour of neuroblas⁃toma.To this end,a two-stage automatic segmentation framework structure model is proposed.Firstly,use 3D convolutional neural networks to locate the tumor location;Then generate multimodal point cloud data near the segmented tumor area and extract the shape contour features of the tumor;Finally,the features extracted by the two networks are fused to predict the final segmentation result.The proposed model was compared with other multimodal methods on both proprietary and public datasets,and the experimental results verified the superiority and ef⁃fectiveness of the proposed model,which can provide reference and inspiration for researchers studying the segmentation of neuroblastoma.

关 键 词：深度学习 点云 多模态 医学图像分割 PET-CT 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]