摘要
目的:探索基于深度学习的图像语义分割和图像识别技术在宫颈癌早筛中的应用,并构建基于全切片宫颈细胞涂片的宫颈癌智能辅助检测系统。方法:联合图像语义分割和图像识别技术构建宫颈癌智能辅助检测系统。首先通过语义分割完成对宫颈细胞核的分割定位,进而识别异常细胞。结果:在细胞核定位阶段,平均像素精确度(Mean Pixel Accuracy,MPA)达到87.85%,频权交并比(Frequency Weighted Intersection over Union,FWIoU)达到96.69%。在异常细胞识别阶段,灵敏度达到95.00%,准确率96.00%,与此同时,辅助检测系统还能实现细胞计数功能。结论:基于深度学习的宫颈癌智能辅助检测系统能够为临床宫颈癌早筛提供良好的辅助功能,降低检测成本,提升检测效率,减轻病理医生的阅片工作量。
Objective:Explore the application of deep learning image semantic segmentation and image recognition technology in the early screening of cervical cancer and construct a cervical cancer intelligent auxiliary detection system based on a full-slice cervical smear.Methods:This research proposes to combine image semantic segmentation and image recognition technology to construct an intelligent auxiliary detection system for cervical cancer.First,complete the segmentation and positioning of the cervical nucleus through semantic segmentation,and then identify abnormal cells.Results:In the cell nucleus localization stage,Mean Pixel Accuracy(MPA)of our system is 87.85%.Frequency Weighted Intersection over Union(FWIoU)is 96.69%.In the abnormal cell identification stage,the sensitivity reaches 95.00%and the accuracy rate is 96.00%.At the same time,our auxiliary detection system can also realize the cell counting function.Conclusion:The intelligent auxiliary detection system for cervical cancer based on deep learning can provide good assistance and empowerment for clinical early screening of cervical cancer,reduce detection costs,improve detection efficiency,and greatly reduce the workload of pathologists.
出处
《中国数字医学》
2021年第7期44-49,共6页
China Digital Medicine
基金
浙江省重点研发计划-智能医疗开放创新平台开发及应用示范(编号:2020C01022)
关键词
深度学习
宫颈细胞涂片
细胞核定位
deep learning
cervical smear
cell nucleus localization
