宫颈上皮内瘤变(Cervical Intraepithelial Neoplasm,CIN)是宫颈浸润癌变相关度较高的癌前病变,准确检测CIN并对其分类处理有利于减少宫颈癌重症率。针对宫颈病变检测与分类准确率低等问题,文中提出一种融合多尺度特征与多注意力机制的Y...宫颈上皮内瘤变(Cervical Intraepithelial Neoplasm,CIN)是宫颈浸润癌变相关度较高的癌前病变,准确检测CIN并对其分类处理有利于减少宫颈癌重症率。针对宫颈病变检测与分类准确率低等问题,文中提出一种融合多尺度特征与多注意力机制的YOLOv5-CBTR(You Only Look Once version 5-Convolutional Block Transformer)宫颈病变图像检测方法。主干网络采用带有SENet(Squeeze-and-Excitation Networks)注意力机制的SE-CSP(SENet-BottleneckCSP)进行特征提取。引入Transformer编码器模块,融合多特征信息并放大,采用多头注意力机制增强病变区域的特征提取能力。在特征融合层引入卷积注意力模块,多尺度融合病变特征信息。在边界回归框计算中引入幂变换,加快模型损失函数的收敛,整体实现宫颈病变的检测与分类。实验结果表明,YOLOv5-CBTR模型对RGB(白光)宫颈病变图像检测与分类的准确率、召回率、mAP(mean Average Precision)和F值分别为93.99%、92.91%、92.80%和93.45%,在多光谱宫颈图像检测与分类中模型的mAP值和F值分别为97.68%和95.23%。展开更多
OBJECTIVE To investigate the available parameters in gynecological screening for cervical lesions by liquid-based cytology technology （ThinPrep Cytology Test, TCT） and The Bethesda System （TBS）, also with computer...OBJECTIVE To investigate the available parameters in gynecological screening for cervical lesions by liquid-based cytology technology （ThinPrep Cytology Test, TCT） and The Bethesda System （TBS）, also with computer image analysis. METHODS With application of the image analysis system, all grades of cervical lesion cells were detected quantitatively and sorted in atypical squamous cells of undetermined significance （ASCUS）, atypical squamous cells-cannot exclude HSIL （ASC-H）, low-grade squamous intraepithelial lesion （LSIL）, high-grade squamous intraepithelial lesion （HSIL） and cervical squamous cell carcinoma （SCC） with the mean optical density （MOD）, average grey （AG）, positive units （PU）, and nucleus to cytoplasmic ratio （N： C）. Differences between each group of cells were compared and analyzed statistically. RESULTS Apart from four stereologic parameters in LSIL and HSIL groups there were no differences among them, in the other groups, there was statistically significant in differences between MOD, AG and PU values. Differences between them in the ratio of nucleus to cytoplasm were highly statistically significant. CONCLUSION Stereological indexes may serve as a screening tool for cervical lesions. The image analysis system is expected to become a new means of cytological assisted diagnosis.展开更多
文摘宫颈上皮内瘤变(Cervical Intraepithelial Neoplasm,CIN)是宫颈浸润癌变相关度较高的癌前病变,准确检测CIN并对其分类处理有利于减少宫颈癌重症率。针对宫颈病变检测与分类准确率低等问题,文中提出一种融合多尺度特征与多注意力机制的YOLOv5-CBTR(You Only Look Once version 5-Convolutional Block Transformer)宫颈病变图像检测方法。主干网络采用带有SENet(Squeeze-and-Excitation Networks)注意力机制的SE-CSP(SENet-BottleneckCSP)进行特征提取。引入Transformer编码器模块,融合多特征信息并放大,采用多头注意力机制增强病变区域的特征提取能力。在特征融合层引入卷积注意力模块,多尺度融合病变特征信息。在边界回归框计算中引入幂变换,加快模型损失函数的收敛,整体实现宫颈病变的检测与分类。实验结果表明,YOLOv5-CBTR模型对RGB(白光)宫颈病变图像检测与分类的准确率、召回率、mAP(mean Average Precision)和F值分别为93.99%、92.91%、92.80%和93.45%,在多光谱宫颈图像检测与分类中模型的mAP值和F值分别为97.68%和95.23%。
基金This work was supported by a grant from the Natural Science Foundation of Nenan Province, China （No.102300410078）.
文摘OBJECTIVE To investigate the available parameters in gynecological screening for cervical lesions by liquid-based cytology technology （ThinPrep Cytology Test, TCT） and The Bethesda System （TBS）, also with computer image analysis. METHODS With application of the image analysis system, all grades of cervical lesion cells were detected quantitatively and sorted in atypical squamous cells of undetermined significance （ASCUS）, atypical squamous cells-cannot exclude HSIL （ASC-H）, low-grade squamous intraepithelial lesion （LSIL）, high-grade squamous intraepithelial lesion （HSIL） and cervical squamous cell carcinoma （SCC） with the mean optical density （MOD）, average grey （AG）, positive units （PU）, and nucleus to cytoplasmic ratio （N： C）. Differences between each group of cells were compared and analyzed statistically. RESULTS Apart from four stereologic parameters in LSIL and HSIL groups there were no differences among them, in the other groups, there was statistically significant in differences between MOD, AG and PU values. Differences between them in the ratio of nucleus to cytoplasm were highly statistically significant. CONCLUSION Stereological indexes may serve as a screening tool for cervical lesions. The image analysis system is expected to become a new means of cytological assisted diagnosis.
