马祯一　　钱萍　　俞芳　赵俊　　陈建清

[摘要] 目的 依靠深度學习的智能图像分析对IgA肾病精确检测及诊断，使临床采取有效的防治手段，是减少终末期肾病的发病率及病死率的关键。 方法 选取2016年1月至2019年6月我院肾病科收治的肾病病例共452例，排除因肾穿刺送检标本数量少，无法出具病理诊断;排除肾穿刺病理诊断疑似诊断者，排除未行免疫荧光检查患者后，确诊IgA肾病患者共135例进行图像分析，采用国际通用5级半定量法评价，选择传统图像处理方法分割提取荧光沉积区域。将输入图像进行颜色空间转换，在颜色和亮度两个特征维度，采用自适应阈值方法产生二值化图像。然后使用区域分离与合并，获得独立的沉积区域，再计算各个沉积区域的轮廓、面积和平均亮度，得出计算机自动识别荧光沉积强度和形状的过程。 结果 基于深度学习的人工智能图像分析能实现对IgA肾病免疫荧光结果的判读，与病理诊断医生结果相比符合率较高，IgA达88.9%，IgG达85.8%，IgM达83.8%，C3达88.6%，因此其可以协助病理诊断医生对IgA肾病免疫荧光的判读。 结论 充分利用计算机技术及网络技术改变病理工作流程，提高病理诊断医生的工作效率，减少因疲劳阅片发生的误诊率，使病理诊断更加精准及客观。

[关键词] IgA肾病;免疫荧光;智能图像分析;深度学习

[中图分类号] R737.9          [文献标识码] B          [文章编号] 1673-9701（2021）09-0147-05

Application of intelligent image analysis in immunofluorescence pathological diagnosis of IgA nephropathy

MA Zhenyi1   QIAN Ping1   YU Fang1   ZHAO Jun1   CHEN Jianqing2

1.Department of Pathology，Jiaxing Hospital of Traditional Chinese Medicine Affiliated to Zhejiang Chinese Medical University，Jiaxing   314001，China;2.Department of Information，Jiaxing Hospital of Traditional Chinese Medicine Affiliated to Zhejiang Chinese Medical University，Jiaxing   314001，China

[Abstract] Objective It is the key to reduce the incidence and mortality of end-stage renal disease to accurately detect and diagnose IgA nephropathy by relying on intelligent image analysis of deep learning，and to adopt effective prevention and treatment measures in clinic. Methods In this research，a total of 452 nephrotic cases admitted to the department of nephropathy in our hospital from January 2016 to June 2019 were selected. After excluding those because the number of specimens sent for renal puncture was small and pathological diagnosis could not be issued， those with suspected diagnosis of renal puncture pathology and those without immunofluorescence examination， a total of 135 patients with IgA nephropathy were diagnosed for image analysis. Meanwhile，they were evaluated by international five-level semi-quantitative method，and the fluorescence precipitation area was extracted by traditional image processing method. The input image was transformed into color space，and the binary image was generated by adaptive threshold method in the two feature dimensions of color and brightness. Then，independent deposition areas were obtained by region separation and combination，and then the contour， area and average brightness of each deposition area were calculated， and the process of automatic recognition of fluorescence deposition intensity and shape by computer was obtained. Results On the basis of deep learning，artificial intelligence image analysis can realize the interpretation of immunofluorescence results of IgA nephropathy. Compared with the results of pathological diagnosis doctors，the coincidence rate was higher， IgA reached 88.9%，IgG reached 85.8%，IgM reached 83.8% and C3 reached 88.6%. Therefore，it could assist pathological diagnosis doctors in the interpretation of IgA nephropathy immunofluorescence. Conclusion It is time to make full use of computer technology and network technology to change the pathological workflow，improve the work efficiency of pathological diagnosis doctors，reduce the misdiagnosis rate due to fatigue，and make pathological diagnosis more accurate and objective.

[Key words] IgA nephropathy; Immunofluorescence; Intelligent image analysis; Deep learning

IgA肾病（IgA nephropathy，IgAN）是以IgA为主的免疫复合物沉积，伴系膜细胞及基质增生为主和补体C3等膜攻击复合物（Membrane attack complex，MAC）沉积于肾小球系膜区的肾小球疾病。……