摘要
目的探讨静息态脑活动及功能连接特征在分类乳腺癌化疗相关认知障碍中的价值。材料与方法收集乳腺癌(breast cancer,BC)化疗患者基线期(P0期)40人,治疗结束后一周(P1期)33人,治疗结束后半年(P2期)19人以及健康对照(healthy control,HC)组44人,均进行静息态功能磁共振成像(resting-state functional magnetic resonance imaging,rs-fMRI)和神经精神学量表评估。功能图像经DPARSF和PRoNTo软件处理后得到低频振幅(amplitude of low frequency fluctuation,ALFF)、低频振幅比率(fractional amplitude of low frequency fluctuation,fALFF)、局部一致性(regional homogeneity,ReHo)和海马功能连接四种数据特征集,利用PRoNTo 2.1工具箱将四种数据特征分别作为机器学习算法的输入,使用二分类法建模。基线期临床数据指标组间比较采用独立样本t检验,随访组组内采用单因素ANOVA。结果与化疗前相比,乳腺癌患者化疗后一周在中文听觉词语学习测验(Auditory Verbal Learning Test,AVLT)、抑郁及焦虑评分中差异有统计学意义(P<0.05)。在四组数据特征中,分类化疗相关认知障碍准确率最高的数据特征为fALFF。对模型具有显著贡献的脑区为:右侧枕上回、右额中上回、右侧后扣带回及楔前叶周围。在HC组与BC组以及BC组内各期分类中,准确率最高的为P0期vs.HC组(准确率86.90%,P<0.001),在P0、P1、P2组间,P0与P1组分类准确率高于其他两组(准确率76.27%,P<0.001),在fALFF数据特征分类中,基于决策函数的权值图的分布大部分覆盖了基于体素分析的统计学差异显著的脑区。结论基于rs-fMRI数据特征的机器学习法可以有效区分乳腺癌化疗相关认知障碍患者,为早期诊断提供影像学参考。
Objective:The aim of this study was to investigate the diagnostic value of resting brain activity and functional connectivity characteristics in classifying chemotherapy-related cognitive impairment in breast cancer.Materials and Methods:A total of 40 patients with breast cancer(BC)treated with chemotherapy at baseline(P0),33 survivors assessed one week following treatment(P1),19 survivors assessed six months after treatment completion(P2),and 44 female volunteers as the healthy control(HC)group were recruited in this study and underwent resting-state functional magnetic resonance imaging(rs-fMRI)examination and neuropsychological test.After data processing by DPARSF and PRoNTo software,4 types of rs-fMRI measurements,including low-frequency fluctuations(fALFF),regional homogeneity(ReHo),and hippocampal functional connectivity were obtained.Using the PRoNTo 2.1 toolbox,the four data features were used as inputs to the machine learning algorithm,and the binary classification method was used for modeling.Independent sample t test was used for comparison of clinical data indicators at baseline,and single factor ANOVA was used between groups Results:Compared with P0,the BC group showed significantly statistical significance in auditory verbal learning test,self-rating depression scale(SDS)and self-rating anxiety scale at P1(P<0.05).The fALFF feature gave the highest accuracy in classifing chemotherapy related cognitive impairment among these groups.The specific results demonstrated the highest accuracy of classification was between P0 and HC groups(accuracy 86.90%,P<0.001).Among the group P0,P1,and P2,the classification accuracy between the P0 and P1 groups(accuracy 76.27%,P<0.001)was higher than that of other classifications.In all classifications,the regions showing high feature importance calculated by the decision function within the algorithm largely overlapped with those showing significant differences during the comparisons between fALFF maps in t-tests.Conclusions:The machine learning algorithm based on multiple types of rs-fMRI measurements can effectively identify breast cancer patients with chemotherapy-related cognitive impairment and provide imaging reference for early diagnosis.
作者
冯云
郭莉莉
鲍艺文
黄伟
柏根基
FENG Yun;GUO Lili;BAO Yiwen;HUANG Wei;BAI Genji(Department of Medical Imaging,Huai'an First Hospital Affiliated of Nanjing Medical University,Huai'an 223300,China)
出处
《磁共振成像》
CAS
CSCD
北大核心
2024年第11期83-89,共7页
Chinese Journal of Magnetic Resonance Imaging
基金
淮安市自然科学研究计划项目(编号:HAB202103)
淮安市级自然科学研究计划(联合专项)卫生健康类项目(编号:HAB202235)。
关键词
乳腺癌
化疗
认知障碍
支持向量机
功能连接
磁共振成像
breast cancer
chemotherapy
cognitive impairment
support vector machine
functional connectivity
magnetic resonance imaging
作者简介
通信作者:柏根基,E-mail:hybgj0451@163.com。
