基于人工智能技术的动态增强磁共振成像直方图分析在乳腺癌术前分级诊断中的价值

Value of DCE-MRI histogram analysis based on AI in preoperative grading diagnosis of breast cancer

目的:探讨基于人工智能技术的动态增强磁共振成像(DCE-MRI)直方图分析在乳腺癌术前分级诊断中的价值。方法:连续纳入2020年9月至2022年9月河北生殖妇产医院收治的80例乳腺癌患者,分别进行分子分型[Luminal A型22例,Luminal B型44例,三阴型10例,人表皮生长因子受体2(HER-2)过表达型4例]和组织学分级(1级21例,2级20例,3级39例)。收集所有患者DCE-MRI检查资料,将图像传至图像后台工作站进行图像后处理,获取速率常数(K_(ep))、容积转移常数(K^(trans))以及血管外细胞外间隙容积比(V_(e))的平均值、10%位数、25%位数、75%位数和90%位数,并进行人工智能分析。结果:分子分型中非Luminal B型乳腺癌患者K_(ep)值的平均值、10%位数、25%位数、75%位数和90%位数高于Luminal B型乳腺癌患者,差异有统计学意义(t=23.203、14.305、10.706、10.257、19.754,P<0.05),K^(trans)值的平均值、10%位数、25%位数、75%位数和90%位数高于Luminal B型乳腺癌患者,差异有统计学意义(t=8.946、6.803、15.113、6.309、8.284,P<0.05),V_(e)值的平均值、10%位数、25%位数、75%位数和90%位数低于Luminal B型乳腺癌患者,差异有统计学意义(t=8.850、8.686、5.831、9.580、6.753,P<0.05)。组织学分级中3级乳腺癌患者K_(ep)值的平均值、10%位数、25%位数、75%位数和90%位数高于1~2级乳腺癌患者,差异有统计学意义(t=3.478、2.487、2.858、2.308、2.048,P<0.05),K^(trans)值的平均值、10%位数、25%位数、75%位数和90%位数高于1~2级乳腺癌患者,差异有统计学意义(t=2.103、2.075、2.063、2.116、2.042,P<0.05),V_(e)值的平均值、10%位数、25%位数、75%位数和90%位数低于1~2级乳腺癌患者,差异有统计学意义(t=8.925、2.368、6.545、3.370、2.008,P<0.05)。K_(ep)值的平均值和10%位数、K^(trans)值的平均值和10%位数与乳腺癌组织学分级呈显著正相关(r=0.541、0.425、0.481、0.469,P<0.05),V_(e)值的平均值与乳腺癌组织学分级呈显著负相关(r=-0.567,P<0.05)。结论:基于人工智能技术的DCE-MRI直方图分析可消除主观性和人为误差影响,提高乳腺癌术前分级诊断的客观性和一致性,帮助临床医生制定个性化治疗方案,具有临床推广价值。

Objective:To explore the value of histogram analysis of dynamic contrast enhanced-magnetic resonance imaging(DCE-MRI)based on artificial intelligence(AI)technique in preoperative grading diagnosis of breast cancer.Methods:A total of 80 patients with breast cancer admitted to Hebei Maternity Hospital from September 2020 to September 2022 were consecutively enrolled.They were divided into 22 cases were Luminal A type,and 44 cases were Luminal B type,and 10 cases were triple negative type and 4 cases were human epidermal growth factor receptor 2(HER-2)overexpression type according to molecular typing,and they also were divided into 21 cases were grade 1,20 cases were grade 2 and 39 cases were grade 3 according to histological grading.The DCE-MRI data of all patients were collected and were transmitted to backend workstation to conduct image post-processing.And then,the mean,10%,25%,75%and 90%values of rate constant(K_(ep)),volume transfer constant(K^(trans))and extracellular space volume ratio of blood vessels(V_(e))were obtained,and they were analyzed by AI.Results:As molecular typing,the mean,10%,25%,75%and 90%values of K_(ep) values patients with non-Luminal B type of breast cancer were respectively higher than those of patients with Luminal B type of breast cancer(t=23.203,14.305,10.706,10.257,19.754,P<0.05),and the mean,10%,25%,75%and 90%values of the K^(trans) values patients with non-Luminal B type of breast cancer were respectively higher than those of patients with Luminal B type of breast cancer(t=8.946,6.803,15.113,6.309,8.284,P<0.05),the differences were significant.In addition,the mean,10%,25%,75%and 90%values of the V_(e) values of patients with non-Luminal B type of breast cancer were significantly lower than those of patients with Luminal B type of breast cancer(t=8.850,8.686,5.831,9.580,6.753,P<0.05),respectively.As histological grading,the mean,10%,25%,75%and 90%values of K_(ep) values of patients with grade 3 of breast cancer were significantly higher than those of patients with grade 1-2 of breast cancer(t=3.478,2.487,2.858,2.308,2.048,P<0.05),respectively.The mean,10%,25%,75%and 90%values of K^(trans) values of patients with grade 3 of breast cancer were significantly higher than those of patients with grade 1-2 of breast cancer(t=2.103,2.075,2.063,2.116,2.042,P<0.05),respectively.The mean,10%,25%,75%and 90%values of V_(e) values of patients with grade 3 of breast cancer were significantly higher than those of patients with grade 1-2 of breast cancer(t=8.925,2.368,6.545,3.370,2.008,P<0.05),respectively.The mean and 10%value of K_(ep) value,and the mean and 10%value of K^(trans) value were significantly positively correlated with the histological grading of breast cancer(r=0.541,0.425,0.481,0.469,P<0.05),respectively,and the mean of V_(e) value was significantly negatively correlated with the histological grading of breast cancer(r=-0.567,P<0.05).Conclusion:DCE-MRI histogram analysis based on AI technique can eliminate the influence of subjectivity and human error,and improve the objectivity and consistency of preoperative grading diagnosis of breast cancer,and help clinicians to formulate personalized treatment plans,which has clinical promotion value.