Applications of Machine Learning to Improve Diagnosis, Advance Treatment, and Identify Causal Factors for Mental Disorders.
Applications of Machine Learning to Improve Diagnosis, Advance Treatment, and Identify Causal Factors for Mental Disorders.
复制标题
应用机器学习来改善诊断、推进治疗并确定精神疾病的病因。
DOI：
10.1016/j.bpsc.2022.04.002
复制
发表时间：
2022
期刊：
影响因子：
--
通讯作者：
Hudson,JamesI
中科院分区：
文献类型：
--
作者：
Brennan,BrianP;Hudson,JamesI
机器学习技术主要起源于计算机科学和工程领域，已经改变了许多科学学科中多元建模的格局。虽然机器学习最初是为涉及信息技术和大数据的非学术应用中的预测模型开发的，并且迄今为止主要用于涉及信息技术和大数据的非学术应用中，但机器学习已越来越多地用于增强甚至取代学术领域内某些应用中的多元建模，包括涉及规模较小的数据集的应用。在本期《生物精神病学：认知神经科学和神经影像学》中，Kalmady 等人 (1) 将集成学习框架应用于强迫症 (OCD) 的诊断。他们的 EMPaSchiz（用于精神分裂症预测的多个分区的集成算法）模型通过使用广泛接受的基于死后细胞结构的脑图谱和分区、源自扩散成像的解剖连接性以及源自静息态和基于任务的功能磁共振成像的功能连接性，结合了先前的脑功能和解剖学神经生物学知识。因此，与不使用领域知识（即有关所研究领域的科学信息）的所谓不可知模型相反，Kalmady 等人（1）使用了知识知情方法，这意味着他们使用领域知识为模型添加结构。（这两种分类方法如图 1 的第一列和第二列所示）。 EMPaSchiz 的性能优于基于神经网络机器学习方法的不可知模型 (2)。
Machine learning techniques, originating largely in the fields of computer science and engineering, have transformed the landscape of multivariate modeling in many scientific disciplines. Although initially developed, and thus far primarily used, for prediction models in nonacademic applications involving information technology and big data, machine learning has been increasingly used to augment or even replace multivariate modeling in certain applications within academic fields, including applications involving datasets of more modest size.In the current issue of Biological Psychiatry: Cognitive Neuroscience and Neuroimaging, Kalmady et al.(1) apply an ensemble learning framework to the diagnosis of obsessive-compulsive disorder (OCD). Their EMPaSchiz (Ensemble algorithm with Multiple Parcellations for Schizophrenia prediction) model incorporated prior neurobiological knowledge of brain function and anatomy by using widely accepted brain atlases and parcellations based on postmortem cytoarchitecture, anatomic connectivity derived from diffusion imaging, and functional connectivity derived from resting-state and task-based functional magnetic resonance imaging. Thus, in contrast with a so-called agnostic model that uses no domain knowledge (ie, scientific information about the domain under study), Kalmady et al.(1) used a knowledge-informed approach—meaning that they used domain knowledge to add structure to the model.(These two approaches for classification are depicted in the first and second columns of Figure 1). EMPaSchiz outperformed an agnostic model based on the machine learning method of neural networks (2).
DOI：
10.1016/j.bpsc.2018.07.014
发表时间：
2019-01
期刊：
Biological psychiatry. Cognitive neuroscience and neuroimaging
影响因子：
--
作者：
Brennan BP;Wang D;Li M;Perriello C;Ren J;Elias JA;Van Kirk NP;Krompinger JW;Pope HG Jr;Haber SN;Rauch SL;Baker JT;Liu H
通讯作者：
Liu H
影响因子：
25.8
作者：
Goodkind M;Eickhoff SB;Oathes DJ;Jiang Y;Chang A;Jones-Hagata LB;Ortega BN;Zaiko YV;Roach EL;Korgaonkar MS;Grieve SM;Galatzer-Levy I;Fox PT;Etkin A
通讯作者：
Etkin A