基于改进的Faster R-CNN模型的异常鳞状上皮细胞检测

Atypical Squamous Cells detection based on Improved Faster R-CNN

宫颈癌是目前世界上最常见的妇科恶性肿瘤,患者死亡率非常高。新柏氏液基细胞学检测(TCT)是宫颈癌筛查的基本方法,病理医生在显微镜下观察子宫颈脱落的鳞状上皮细胞,查看是否存在异常鳞状上皮细胞进行诊断。TCT对宫颈癌的检出率为100%,同时还可以发现部分癌前病变和微生物感染。目前国内的病理医生只有10000人左右,而且培养周期长,需求缺口极大。本文使用经过病理医生标注的数字病理图像,训练目标检测模型。设计了基于Faster R-CNN的网络结构改进的模型,引入了可形变卷积网络和特征金字塔网络,实现了对宫颈数字病理图像进行自动识别,为临床宫颈疾病诊断提供辅助参考。实验结果表明,改进后的模型能快速收敛,在测试集上的测试结果m AP(mean Average Precision)可以达到0.29,已经基本满足辅助病理医生诊断的需求(实际医院临床使用的模型m AP为0.32)。

Cervical cancer is the most common gynecological malignant tumor in the world,with a very high mortality rate.Thinprep Cytologic Test(TCT)is the basic method for cervical cancer screening.Pathologists observe squamous epithelial cells shed from the cervix under a microscope to see if there are abnormal squamous epithelial cells for diagnosis.The detection rate of cervical cancer by TCT is 100%,and some precancerous lesions and microbial infections can also be found.At present,there are only about10000 pathologists in China,and the training cycle is long,with a huge demand gap.In this paper,the object detection model is trained by using digital pathological images labeled by pathologists.An improved network structure model based on Faster R-CNN is designed,and a deformable convolutional network and a feature pyramid network are introduced to realize automatic detection of cervical digital pathological images,providing an auxiliary reference for clinical diagnosis of cervical diseases.The experimental results show that the improved model can converge rapidly,and the mean Average Precision on the test set can reach 0.29,which has basically met the needs of auxiliary pathologists for diagnosis(the model mAP used in actual hospital clinical practice is 0.32).