A small number of abnormal brain connections predicts adult autism spectrum disorder

Abstract

Although autism spectrum disorder (ASD) is a serious lifelong condition, its underlying neural mechanism remains unclear. Recently, neuroimaging-based classifiers for ASD and typically developed (TD) individuals were developed to identify the abnormality of functional connections (FCs). Due to over-fitting and interferential effects of varying measurement conditions and demographic distributions, no classifiers have been strictly validated for independent cohorts. Here we overcome these difficulties by developing a novel machine-learning algorithm that identifies a small number of FCs that separates ASD versus TD. The classifier achieves high accuracy for a Japanese discovery cohort and demonstrates a remarkable degree of generalization for two independent validation cohorts in the USA and Japan. The developed ASD classifier does not distinguish individuals with major depressive disorder and attention-deficit hyperactivity disorder from their controls but moderately distinguishes patients with schizophrenia from their controls. The results leave open the viable possibility of exploring neuroimaging-based dimensions quantifying the multiple-disorder spectrum.

Advances in fMRI Real-Time Neurofeedback

“…One is that FCNef can be directly applied to rs-fc-based abnormal connectivity, or a biomarker, in networks [32–34]. For clinical purposes, connectivity-based neurofeedback techniques including FCNef need to be combined with the biomarkers that characterize psychiatric disorders based on rs-fcMRI.…”

“…Regrettably, early studies of biomarkers suffered from overfitting of machine learning algorithms and did not exhibit generalization capability for completely independent validation cohorts [70–74]. However, the recent development of sophisticated machine learning algorithms has overcome these technical problems and led to the production of rs-fcMRI-based biomarkers, which are capable of classifying patients from controls with high accuracy and generalization to completely independent cohorts [32–34,75]. Development of these robust rs-fcMRI-based biomarkers have allowed FCNef, which is also rs-fcMRI-based, to directly modify them.…”

“…Third, the development of multivariate analysis techniques, which allows for more sensitive neurofeedback [24–28], has been incorporated into neurofeedback [13,20,29]. Fourth, changes of connectivity in a targeted brain network [30,31], which could be a cause of mental abnormalities [32–34], have been incorporated into neurofeedback techniques. Combined the use of these techniques has provided insights into a causal relationship between the modified neural change and modified behavior.…”

Classification and characterisation of brain network changes in chronic back pain: A multicenter study

“…In terms of classification methods, the accuracy of the SVM is comparable with that seen in other machine learning-based disease biomarkers that have used independent validation cohorts ( Takagi et al , 2017; Yahata et al , 2016) (albeit less than that seen with classifiers for phasic BOLD responses to acute painful stimulation in healthy individuals ( Wager et al , 2013)). Here, we also applied a deep learning approach using deep convolutional neural networks, the utility of which has not previously been tested in chronic pain.…”

“…One way to tackle this is to use machine learning and deep learning methods, and a number of studies have shown how this can be used to successfully build biomarkers (i.e. classifiers) in a range of psychiatric disease ( Takagi et al , 2017; Watanabe et al , 2017; Yahata et al , 2016; Yamada et al , 2017). However, these methods need to be validated on genuinely independent data sets to be convincing, and current evidence of generalisable classifiers for chronic pain is lacking.…”