Smith Lukasz Tyszczuk, Levita Liat, Amico Francesco, Fagan Jennifer, Yek John H, Brophy Justin, Zhang Haihong, Arvaneh Mahnaz
Annu Int Conf IEEE Eng Med Biol Soc. 2020 Jul;2020:5004-5007. doi: 10.1109/EMBC44109.2020.9176304.
Depression is the leading cause of disability worldwide, yet rates of missed- and mis-diagnoses are alarmingly high. The introduction of objective biomarkers, to aid diagnosis, informed by depression's physiological pathology may alleviate some of the burden on strained mental health services. Three minutes of eyes-closed resting state heart rate and skin conductance response (SCR) data were acquired from 27 participants (16 healthy controls, 11 with major depressive disorder (MDD)). Various classifiers were trained on state-of-the-art and novel features. We are aware of no previous studies analysing the utility of multimodal vs. individual modalities for classification. We found no improvement using multimodal classifiers over using heart rate variability (HRV) alone, which achieved 81% test accuracy. The best multimodal and SCR only classifiers were only slightly less accurate at 78%. Despite not improving depression detection, SCR features did show stronger correlation with suicidal ideation than HRV. SD1/SD2 is a novel HRV feature proposed in this paper, similar to the commonly used ratio SD1/SD2 but with more marked separation between classes, having the largest Rank Biserial Correlation of all examined features (p-value = 0.002, RBC = -0.73). We recommend further studies in this area.
抑郁症是全球致残的主要原因，然而漏诊和误诊率高得惊人。根据抑郁症的生理病理学引入客观生物标志物以辅助诊断，可能会减轻紧张的心理健康服务的一些负担。从27名参与者（16名健康对照者、11名重度抑郁症（MDD）患者）中获取了三分钟闭眼静息状态下的心率和皮肤电反应（SCR）数据。使用最先进的和新颖的特征训练了各种分类器。我们所知，之前没有研究分析多模态与单模态在分类中的效用。我们发现，与单独使用心率变异性（HRV）相比，使用多模态分类器并没有提高准确率，HRV单独使用时测试准确率达到81%。最佳的多模态分类器和仅使用SCR的分类器准确率略低，为78%。尽管没有提高抑郁症检测率，但SCR特征与自杀意念的相关性确实比HRV更强。SD1/SD2是本文提出的一种新颖的HRV特征，类似于常用的比率SD1/SD2，但类别之间的分离更明显，在所有检查特征中具有最大的等级双列相关性（p值 = 0.002，RBC = -0.73）。我们建议在该领域进行进一步研究。
