作者:彭晨辉，刘皓晨，高玮宁，何勇军
Authors:PENGChenhui,LIUHaochen,GAOWeining,HEYongjun摘要:智能病理辅助诊断系统可以辅助医生诊断宫颈癌 , 筛除低质量的样本图像能够有效减少智能病理辅 助诊断系统的漏诊和误诊 ,提高诊断的效率和准确率 。 目前智能辅助诊断领域没有完整的细胞病理图像质量评价 的方法 , 因此提出 一种细胞病理图像质量评价方法 。利用病理诊断知识和医生经验 , 归纳总结出图像质量评价指 标 ,包括图像有效视野、栅格样成像、清晰度、染色标准、鳞状上皮细胞数量、细胞团面积和脏污面积占比等 。首先 针对图像有效视野和栅格样成像利用图像分割和图像纹理特征提取的方法进行评价 ;再采用引入通道注意力的 ResNet-34 模型对清晰度进行评价 ;然后对图像进行颜色反卷积处理 ,获得染色剂通道灰度值 ,用于判断细胞的染 色是否标准 ;再通过目标检测模型 Yolov5s ,对鳞状上皮细胞、细胞团和脏污进行检测 ; 最后利用回归模型为细胞图 像质量评分 ,并将结果应用到智能辅助诊断项目中 。实验表明 ,方法规范了细胞病理图像质量评价的流程 ,避免由 于主观评价带来的差异 ,提高了病理诊断效率和准确率。
Abstract:The intelligent pathological auxiliary diagnosis system can assist doctors in diagnosing cervical cancer , and screening out low-quality sample images can effectively reduce the missed diagnosis and misdiagnosis of the intelligent auxiliary diagnosis system , and improve the efficiency and accuracy of diagnosis. At present , there is no complete cytopathological image quality evaluation method in the field of intelligent auxiliary diagnosis , so we propose a cytopathological image quality evaluation method. Using the knowledge of pathological diagnosis and the experience of doctors , the image quality evaluation indicators were summarized , including the effective field of view , grid-like imaging , sharpness , staining standard , the number of squamous epithelial cells , the area of cell clusters and the proportion of dirty areas. First , the effective field of view and grid-like imaging are evaluated by image segmentation and image texture feature extraction; the ResNet-34 model with channel attention is used to evaluate the sharpness; then the color deconvolution is performed on the image. The gray value of the dye channel is obtained , which is used to judge whether the staining of cells is standard. Then , the target detection model Yolov5s is used to detect squamous epithelial cells , cell clusters and dirt. Finally , the regression model is used to score the cell image quality , and the results are applied to the intelligent auxiliary diagnosis project. Experiments show that the method in this paper standardizes the process of cytopathological image quality evaluation , avoids differences caused bysubjective evaluation , and improves the efficiency and accuracy of pathological diagnosis.

PDF全文下载地址:

可免费Download/下载PDF全文