基于深度学习的乳腺X线摄影肿块检测系统的应用

Application of Deep Learning in Mass Detection Based on Mammography

目的探讨基于深度学习(DL)的乳腺X线摄影肿块检测系统的临床应用价值。资料与方法回顾性分析2019年4—12月深圳市人民医院1755例接受乳腺X线摄影检查的患者资料。由DL系统和2名初级职称医师采用盲法独立阅片,比较DL系统与2名医师对肿块病灶检出的敏感度,并分析患者年龄、乳腺密度分类、乳腺影像报告和数据系统分类、肿块形态、边缘、密度、大小对于DL系统及2名初级职称医师肿块检出准确度的影响。结果共检出肿块324例(618个肿块),2名医师及DL系统分别检出肿块277例(519个肿块)、268例(482个肿块)、284例(533个肿块)。DL系统及2名医师对于肿块检出的敏感度分别为86.25%、83.98%、77.99%,DL系统对于不同乳腺密度分类的肿块检出差异无统计学意义(χ^(2)=3.04,P>0.05),而2名医师的差异有统计学意义(χ^(2)=9.39、8.96,P<0.05)。DL系统对于不同患者年龄、肿块形态、边缘、密度、大小及乳腺影像报告和数据系统分类的肿块检出差异均有统计学意义(χ^(2)=15.28、41.70、58.67、53.22、28.83、53.75,P<0.05)。结论基于DL的乳腺X线摄影肿块检测系统对肿块病变检测不受乳腺密度的影响,可辅助医师减少因腺体致密造成的肿块漏诊。

Purpose To explore the clinical application value of deep learning(DL)in mass detection based on mammography.Materials and Methods A retrospective analysis was performed on 1755 cases of mammography in Shenzhen People’s Hospital from April to December 2019.Independent imaging reading was evaluated by a DL system and was reviewed by two junior doctors via blinded methods.The results of a senior doctor combined with clinical data,relevant imaging examinations and pathological results were used as the reference standard to compare the sensitivity between DL system and two junior doctors for the detection of mass.The effects of patient age,breast density,breast imaging reporting and data system(BI-RADS),morphology,edge,density,size on the DL system and two junior doctors were analyzed.Results A total of 1755 cases of mammography were enrolled in this study,324 cases were detected(n=618 masses in total).277 cases(519 masses),268 cases(482 masses),and 284 cases(533 masses)were detected by two junior doctors and DL systems,respectively.The sensitivities of the DL system and the two junior doctors for the detection of masses were 86.25%,83.98%,and 77.99%,respectively.The DL system had no significant difference in the detection of breast masses with different density(χ^(2)=3.04,P>0.05),while two junior doctors had significant differences in the detection of breast masses with different density(χ^(2)=9.39,8.96,P<0.05).The DL system and two junior doctors had statistically significant differences in the detection of tumor with different age,morphology,edge,density and size,and BI-RADS(χ^(2)=15.28,41.70,58.67,53.22,28.83,53.75,all P<0.05).Conclusion The DL-based mammography mass detection system is not affected by breast density,and it can assist doctors to reduce the missed diagnosis caused by dense glands.