基于神经网络集成模型的宫颈细胞病理计算机辅助诊断方法

针对宫颈细胞病理图像自动筛查问题,本文提出一种基于人工智能技术的计算机辅助诊断方法。该方法通过对宫颈细胞病理图像采用自适应双阈值法进行初步检测,再采用改进Chan-Vase模型进行精确分割,提取出细胞(粘连簇团)中的不同区域。然后,结合病理诊断专家规则,构建相应的正交特征集。在此基础上,使用神经网络集成模型进行正常、疑似病变二分类识别,完成计算机辅助诊断。实验表明,本文方法能够有效完成宫颈病理细胞(粘连簇团)的分类识别,具有较高的正确率(84%)与较低的误判率(2.1%)。满足了在保证判断正确率的条件下,尽量降低将疑似病变样本误判为正常样本的实际病理诊断要求。

Computer-aided diagnosis of cervical cytopathology based on neural network ensemble model

Aiming to the automatic screening of cervical cytopathological images, an artificial intelligence based automatic diagnosis-assisted method was proposed. First of all, adaptive dual threshold method was used to detect the cervical cytopathological images initially. Secondly, improved Chan-Vase model was used to precisely extract different areas of adhesive cell cluster. After that, the related feature set was built according to the diagnostic rules of pathological experts. At last, neural network ensemble was applied to normal or suspected lesions two-classification recognition. The result of the experiment showed that cervical cell lesions could be effectively distinguished according to classification with this method, which had high accuracy (84%) and low rate of misjudgment (2.1%), meeting the practical requirement of pathological diagnosis, which is reducing the miscalculating of the suspected lesions to normal ones, meanwhile assuring the diagnostic accuracy.