基于胸部正面X光的肺结节检测任务因结节较小、肋骨遮挡等原因检测难度较大,需要在保证高敏感度的前提下,尽可能地减少假阳性样本比率.目前大多数肺结节检测方法一般分为3个步骤:肺部区域分割;候选区域生成;通过进一步分类,减少假阳性结...基于胸部正面X光的肺结节检测任务因结节较小、肋骨遮挡等原因检测难度较大,需要在保证高敏感度的前提下,尽可能地减少假阳性样本比率.目前大多数肺结节检测方法一般分为3个步骤:肺部区域分割;候选区域生成;通过进一步分类,减少假阳性结果.这类方法存在一些问题,每一步的结果都依赖于前一步的性能,整个流程往往会使用多个模型、多次处理以提升效果,算法复杂而且计算量大.同时,会有些结节因为器官遮挡不在肺部分割的区域内,肺部分割会漏掉一些结节.针对这个问题,本文使用一个端到端的目标检测网络来完成肺结节检测任务,X光片经过图像预处理后输入网络,直接得到肺结节的预测结果.此方法基于卷积神经网络(Convolutional Neural Network,CNN)的目标检测模型,同时在分类任务中融合位置和尺寸信息,实验证明这些信息有助于模型判断.在公开数据集--日本放射技术学会(Japanese Society of Radiological Technology,JSRT)数据集的实验结果显示,本文方法在平均每张图像4. 5个假阳性结果时敏感度为92%,2个假阳性结果时敏感度为88%,在较低的假阳性率的情况下,超出了先前的研究成果.展开更多
AIM: To ensure the diagnostic value of computer aided techniques in diabetic retinopathy(DR) detection based on ophthalmic photography(OP). METHODS: PubM ed, EMBASE, Ei village, IEEE Xplore and Cochrane Library databa...AIM: To ensure the diagnostic value of computer aided techniques in diabetic retinopathy(DR) detection based on ophthalmic photography(OP). METHODS: PubM ed, EMBASE, Ei village, IEEE Xplore and Cochrane Library database were searched systematically for literatures about computer aided detection(CAD) in DR detection. The methodological quality of included studies was appraised by the Quality Assessment Tool for Diagnostic Accuracy Studies(QUADAS-2). Meta-Di Sc was utilized and a random effects model was plotted to summarize data from those included studies. Summary receiver operating characteristic curves were selected to estimate the overall test performance. Subgroup analysis was used to identify the efficiency of CAD in detecting DR, exudates(EXs), microaneurysms(MAs) as well as hemorrhages(HMs), and neovascularizations(NVs). Publication bias was analyzed using STATA. RESULTS: Fourteen articles were finally included in this Meta-analysis after literature review. Pooled sensitivity and specificity were 90%(95%CI, 85%-94%) and 90%(95%CI, 80%-96%) respectively for CAD in DR detection. With regard to CAD in EXs detecting, pooled sensitivity, specificity were 89%(95%CI, 88%-90%) and99%(95%CI, 99%-99%) respectively. In aspect of MAs and HMs detection, pooled sensitivity and specificity of CAD were 42%(95%CI, 41%-44%) and 93%(95%CI, 93%-93%) respectively. Besides, pooled sensitivity and specificity were 94%(95%CI, 89%-97%) and 87%(95%CI, 83%-90%) respectively for CAD in NVs detection. No potential publication bias was observed. CONCLUSION: CAD demonstrates overall high diagnostic accuracy for detecting DR and pathological lesions based on OP. Further prospective clinical trials are needed to prove such effect.展开更多
文摘基于胸部正面X光的肺结节检测任务因结节较小、肋骨遮挡等原因检测难度较大,需要在保证高敏感度的前提下,尽可能地减少假阳性样本比率.目前大多数肺结节检测方法一般分为3个步骤:肺部区域分割;候选区域生成;通过进一步分类,减少假阳性结果.这类方法存在一些问题,每一步的结果都依赖于前一步的性能,整个流程往往会使用多个模型、多次处理以提升效果,算法复杂而且计算量大.同时,会有些结节因为器官遮挡不在肺部分割的区域内,肺部分割会漏掉一些结节.针对这个问题,本文使用一个端到端的目标检测网络来完成肺结节检测任务,X光片经过图像预处理后输入网络,直接得到肺结节的预测结果.此方法基于卷积神经网络(Convolutional Neural Network,CNN)的目标检测模型,同时在分类任务中融合位置和尺寸信息,实验证明这些信息有助于模型判断.在公开数据集--日本放射技术学会(Japanese Society of Radiological Technology,JSRT)数据集的实验结果显示,本文方法在平均每张图像4. 5个假阳性结果时敏感度为92%,2个假阳性结果时敏感度为88%,在较低的假阳性率的情况下,超出了先前的研究成果.
基金Supported by National Key R&D Program of China (No.2018YFC1314900 No.2018YFC1314902)+2 种基金Nantong “226 Project”Excellent Key Teachers in the “Qing Lan Project” of Jiangsu Colleges and UniversitiesJiangsu Students’ Platform for Innovation and Entrepreneurship Training Program (No.201910304108Y)
文摘AIM: To ensure the diagnostic value of computer aided techniques in diabetic retinopathy(DR) detection based on ophthalmic photography(OP). METHODS: PubM ed, EMBASE, Ei village, IEEE Xplore and Cochrane Library database were searched systematically for literatures about computer aided detection(CAD) in DR detection. The methodological quality of included studies was appraised by the Quality Assessment Tool for Diagnostic Accuracy Studies(QUADAS-2). Meta-Di Sc was utilized and a random effects model was plotted to summarize data from those included studies. Summary receiver operating characteristic curves were selected to estimate the overall test performance. Subgroup analysis was used to identify the efficiency of CAD in detecting DR, exudates(EXs), microaneurysms(MAs) as well as hemorrhages(HMs), and neovascularizations(NVs). Publication bias was analyzed using STATA. RESULTS: Fourteen articles were finally included in this Meta-analysis after literature review. Pooled sensitivity and specificity were 90%(95%CI, 85%-94%) and 90%(95%CI, 80%-96%) respectively for CAD in DR detection. With regard to CAD in EXs detecting, pooled sensitivity, specificity were 89%(95%CI, 88%-90%) and99%(95%CI, 99%-99%) respectively. In aspect of MAs and HMs detection, pooled sensitivity and specificity of CAD were 42%(95%CI, 41%-44%) and 93%(95%CI, 93%-93%) respectively. Besides, pooled sensitivity and specificity were 94%(95%CI, 89%-97%) and 87%(95%CI, 83%-90%) respectively for CAD in NVs detection. No potential publication bias was observed. CONCLUSION: CAD demonstrates overall high diagnostic accuracy for detecting DR and pathological lesions based on OP. Further prospective clinical trials are needed to prove such effect.
