Whole-brain structural magnetic resonance imaging-based classification of primary dysmenorrhea in pain-free phase: a machine learning study
Whole-brain structural magnetic resonance imaging-based classification of primary dysmenorrhea in pain-free phase: a machine learning study
复制标题
基于全脑结构 MRI 的无痛期原发性痛经分类：一项机器学习研究。
DOI：
10.1097/j.pain.0000000000001428
复制
发表时间：
2019-03-01
期刊：
影响因子：
7.4
通讯作者：
Liu, Jixin
中科院分区：
文献类型：
--
作者：
Chen, Tao;Mu, Junya;Liu, Jixin
开发一个机器学习模型，以研究原发性痛经（PDM）妇女和健康对照（HC）在无痛期的全脑灰质（GM）图像的区分能力，并进一步评估贡献特征在预测月经疼痛强度变化方面的预测能力。从当地大学招募60例PDM患者和54例匹配的女性HC。所有参与者在排卵期进行头部和盆腔磁共振成像扫描，以计算GM体积和子宫肌层表观扩散系数（ADC）。还进行了问卷评估。支持向量机算法被用来开发分类模型。通过排列检验确定模型性能的显著性。采用多元回归分析方法探讨月经痛的鉴别特征与疼痛强度的关系。人口统计学和基于子宫肌层ADC的分类未能通过排列检验。基于脑的分类结果表明，75.44%的被试者被正确分类，其中PDM患者的识别率为83.33%（P < 0.001）。在回归分析中，人口统计学指标和子宫肌层ADC占疼痛强度方差的29.37%。在回归这些因素后，GM特征解释了剩余方差的60.33%。我们的研究结果表明，GM体积可用于区分PDM和HC患者在无痛期，神经影像学特征可以进一步预测月经疼痛强度的变化，这可能为评估月经疼痛干预提供一个潜在的影像学标记。
To develop a machine leaming model to investigate the discriminative power of whole-brain gray-matter (GM) images derived from primary dysmenorrhea (PDM) women and healthy controls (HCs) during the pain-free phase and further evaluate the predictive ability of contributing features in predicting the variance in menstrual pain intensity. Sixty patients with PDM and 54 matched female HCs were recruited from the local university. All participants underwent the head and pelvic magnetic resonance imaging scans to calculate GM volume and myometrium-apparent diffusion coefficient (ADC) during their periovulatory phase. Questionnaire assessment was also conducted. A support vector machine algorithm was used to develop the classification model. The significance of model performance was determined by the permutation test. Multiple regression analysis was implemented to explore the relationship between discriminative features and intensity of menstrual pain. Demographics and myometrium ADC-based classifications failed to pass the permutation tests. Brain-based classification results demonstrated that 75.44% of subjects were correctly classified, with 83.33% identification of the patients with PDM (P < 0.001). In the regression analysis, demographical indicators and myometrium ADC accounted for a total of 29.37% of the variance in pain intensity. After regressing out these factors, GM features explained 60.33% of the remaining variance. Our results suggested that GM volume can be used to discriminate patients with PDM and HCs during the pain-free phase, and neuroimaging features can further predict the variance in the intensity of menstrual pain, which may provide a potential imaging marker for the assessment of menstrual pain intervention.