摘要
目的:分析基于全数字化乳腺X线摄影(FD)的全卷积单阶段乳腺AI算法及基于数字乳腺断层摄影(DBT)的三维多影像融合AI算法在乳腺肿瘤影像诊断中的效能。方法:用FD AI算法(AI-FD)及DBT AI算法(AI-DBT)对常州市妇幼保健院乳腺外科2022年经病理证实的469例(515个病灶)乳腺疾病患者的影像数据进行计算。以病理结果为基准,分别采集记录两种算法结果的阳性数及与病理结果符合数,比较两种算法对乳腺癌的诊断灵敏度、特异度、阳性预测值及阴性预测值;以病理诊断为金标准,比较AI算法、超声(US)及钼靶X线人工诊断(MG人工)等诊断方法的受试者工作特征曲线(ROC)和曲线下的面积(AUC)。结果:AI-DBT的诊断阳性率(67.81%)高于AI-FD (49.17%),差异有统计学意义(χ^(2)=35.01,P<0.05);AI-DBT的诊断准确率(44.33%)稍低于AI-FD (46.90%),差异无统计学意义(χ^(2)=0.42,P>0.05)。两种AI算法比较,AI-DBT对乳腺癌的诊断灵敏度高,但特异性较弱,AI-FD的诊断特异性更佳。AUC值方面,MG人工的最高,为0.804,两种AI算法次之,但均略高于US。结论:基于乳腺DBT三维断层图像与FD二维图像的AI算法在乳腺肿瘤尤其是在乳腺癌的诊断方面具有一定的准确性,但其诊断效能与MG人工诊断之间仍有一定的差距,故目前尚不能完全取代人工诊断。
Objective:Analyze the efficacy of the fully convolutional single-stage breast AI algorithm based on fully digital mammography(FD)and the 3D multi image fusion AI algorithm based on digital breast tomography(DBT)in the diagnosis of breast tumor images.Methods:The FD AI algorithm(AI-FD)and DBT AI algorithm(AI-DBT)were used to calculate the imaging data of 469 cases(515 lesions)with the pathologically confirmed breast diseases in breast surgery department of Changzhou Maternal and Child Health Hospital in 2022.Based on the pathological results,collect and record the positive number and coincidence number of the results of the two algorithms,and compare the diagnostic sensitivity,specificity,positive predictive value and negative predictive value of the two algorithms for breast cancer.With pathological diagnosis as the gold standard,the receiver operating characteristic curve(ROC)and the area under the curve(AUC)of diagnostic methods such as AI algorithm,ultrasound(US),and manual diagnosis of molybdenum target X-ray(manual MG).Results:The positive data of AI-DBT group was 67.81%,higher than the AI-FD group(49.11%),the difference was statistically significant(χ^(2)=35.01,P<0.05).The coincidence data of AI-DBT group was 44.33%,slightly lower than the AI-FD group(46.90%),there was no statistical difference between groups according to the meaning(χ^(2)=0.42,P>0.05).Comparing between the two algorithms,AI-DBT has high sensitivity but weak specificity in the diagnosis of breast cancer,while the specificity of AI-FD group was better.In terms of AUC value,manual diagnosis(MG)was the highest,0.804,followed by the two AI algorithms,which were both slightly higher than ultrasound.Conclusion:The AI algorithm based on three-dimensional tomographic images of breast DBT and two-dimensional images of FD has certain accuracy in the diagnosis of breast tumors,especially breast cancer.However,there is still a certain gap of the diagnostic efficacy between it and manual MG.Therefore,it cannot completely replace manual diagnosis at present.
作者
陆壹子
冯鑫钰
张望
张凤丽
魏赟
陈楚楚
LU Yizi;FENG Xinyu;ZHANG Wang;ZHANG Fengli;WEI Yun;CHEN Chuchu(Department of Radiology,Changzhou Maternal and Child Health Care Hospital,Changzhou 213000,Jiangsu,China)
出处
《暨南大学学报（自然科学与医学版）》
CAS
北大核心
2024年第4期438-446,共9页
Journal of Jinan University(Natural Science & Medicine Edition)
基金
中关村精准医学基金会临床科研专项资助基金项目(320.1140.2023.0905.14)。
