基于卷积神经网络的白带中白细胞的自动检测

摘要

As the most common gynecological examination items,leucorrhea routine examination has a wide application and an important position in clinical testing.In view of the importance of leukocytes in clinical medicine and the many deficiencies of current detection methods,this paper focuses on the automatic detection of leukocytes.Microscopic images were obtained in a local hospital through a leucorrhea automatic detector.After filtering,the images were enhanced and segmented.The sample library was established,and the feature extraction and classification were done based on the convolution neural network.Finally,the validity of the method was verified by cross validation.In the automatic detection of leukocytes,for a dataset consisting of twenty thousand samples,our proposed method achieved 95％ in sensitivity,84％ in specificity and 89.5％ in accuracy,which meet the requirement of medical clinical testing.The digital image processing technology and the convolution neural network are applied to the detection of leukocytes in medical microscopic images.The proposed method solves the key problem of characteristic expression,verifies the feasibility of automatic identification,and improves the quality and efficiency of detection.%作为妇科常规的检查项目,白带常规检查有着相当广泛的应用,在临床检验中具有非常重要的地位.鉴于白带中白细胞在临床医学中的重要意义和现行检测方法的诸多不足,对白细胞的自动检测技术进行研究.依托于白带自动检测仪,与本地医院进行协作,采集得到白带显微图像.对滤波增强后的图像进行分割,建立样本库,基于卷积神经网络完成特征提取和分类,最终采用交叉验证证实该方法的有效性.在白细胞的自动检测中,对于由2万个样本组成的数据集,运用该方法实现95％的敏感性、84％的特异性、89.5％的准确率,达到医学临床检验的要求.将数字图像处理技术和卷积神经网络综合应用于医学显微图像中白细胞的检测,解决特征表达的关键问题,验证白细胞自动检测的可行性,实现检测质量和检测效率的提升.