基于VGG的SPECT骨扫描图像关节炎分类  

VGG-Based Classification of Arthritis in Bone SPECT Images

作　　者：曹传贵 林强 满正行 曹永春[1,2] 邓涛[1,2] 李同同[1,2] 王海军[3] CAO Chuan-gui;LIN Qiang;MAN Zheng-xing;CAO Yong-chun;DENG Tao;LI Tong-tong;WANG Hai-jun(School of Mathematics and Computer Science,Northwest Minzu University,Lanzhou 730030,China;Key Laboratory of Dynamic Streaming Data Computing and Application,Northwest Minzu University,Lanzhou 730030,China;Department of Nuclear Medicine,Gansu Provincial Hospital,Lanzhou 730030,China)

机构地区：[1]西北民族大学数学与计算机科学学院,甘肃兰州730030 [2]西北民族大学动态流数据计算与应用实验室,甘肃兰州730030 [3]甘肃省人民医院核医学科,甘肃兰州730020

出　　处：《西北民族大学学报（自然科学版）》2021年第2期36-45,共10页Journal of Northwest Minzu University(Natural Science)

基　　金：国家自然科学基金项目(61562075);中央高校科研项目(31920180114);西北民族大学中央高校基本科研业务费专项资金资助研究生项目(Yxm2020100)。

摘　　要：核医学SPECT是主要的功能成像模态,在骨转移、关节退行性改变等疾病的诊治中发挥着重要作用.关节炎是常见且多发性生理疾病,临床上容易在骨转移特别是溶骨性转移之间产生误判.为了从SPECT图像中可靠识别关节炎病变,借助于深度学习的特征自动提取功能,研究并构建了面向关节炎自动诊断的SPECT图像分类器.首先,对SPECT骨显像数据进行归一化及扩展处理,适度扩充数据量并转化到模型要求的数据格式;其次,基于标准的VGG模型构建具有不同深度的关节炎分类器;最后,使用一组真实SPECT全身骨显像数据,对构建的分类模型进行测试.实验结果表明,构建的分类器可有效检测关节病变,获得的准确率、AUC值、精度、召回率分别为0.926、0.986、0.921、0.934.Nuclear medicine SPECT is the main functional imaging modality, which plays a key role in diagnosing and treating the diseases including bone metastasis and joint degeneration.Arthritis is a common and multiple physiological disease, which is clinically prone to misdiagnosis between bone metastases, especially osteolytic metastases.In order to reliably identify arthritic lesions using SPECT images, a SPECT image classifier for automatic diagnosis of arthritis is developed by taking advantage of the automatic feature extraction ability that deep learning has.First, normalization and expansion of SPECT bone images were carried out, and the data was moderately expanded and converted to the data format required by the model.Second, arthritis classifiers with different depths were developed based on the standard VGG model.Last, a group of real-world whole-body SPECT bone images was used to test the developed classification models.The experimental results showed that our classifiers are workable and effective in identifying arthritis in SPECT images, obtaining values of 0.926,0.986, 0.921 and 0.934 for accuracy, AUC,precision and recall, respectively.

关 键 词：关节炎 SPECT成像 图像分类 深度学习 VGG模型 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]