Atypical perception in autism spectrum disorder:Combining computational models, functional neuroimaging and MR spectroscopy to understand aberrant perceptual mechanisms

自闭症谱系障碍（ASD）是一种以感知异常为特征的终身发育疾病。预测编码最近作为一个统一的框架出现，它不仅能够描述自闭症的病理生理，而且还可以在神经元和突触水平上创建自闭症感知改变的可测试假设。在这个框架内，ASD的核心异常被认为是异常贝叶斯推断的结果，或者是由于先验信念的低精度（“次先验”）和/或过于精确的似然分布。因此，自闭症的感知是由感官输入主导的，较少受自上而下的先前经验的正则化调节。在目前的项目中，我们的目标是揭示ASD中知觉推理改变的神经机制，特别是在分类过程方面。我们将使用多模态方法，结合计算建模，行为评估，功能神经成像和磁共振光谱。该项目由三个工作包组成。在工作包1中，我们使用计算建模方法研究人脸识别。面部识别障碍是ASD的一个主要行为缺陷。我们的目标是开发一个计算模型来描述在行为层面上对面部身份的学习。然后，我们将中心模型参数（例如预测误差信号）与神经活动相关联，以了解异常感知推理如何与神经系统级别的变化相关联（“基于模型的fMRI”）。在工作包2中，我们研究了ASD患者通常受损的情绪分类。我们将使用一种新开发的情绪范式，在这种范式中，受试者必须对可变模糊的面部情绪进行分类。通过神经网络分析，我们将研究从前额皮质到感觉面部和情绪感知区域的自上而下的调节是否在ASD患者中减弱，正如预测编码理论所预测的那样。在工作包3中，我们还采取了另一种方法，并测试了ASD中异常感知推理的假定神经生理学对应物的假设。在神经层面，预测编码框架将ASD的核心异常与局部gaba能中间神经元活动的改变联系起来。这一假设将在面部感知和情绪处理网络的核心区域进行核磁共振波谱测试。

Autism spectrum disorder (ASD) is a lifelong developmental condition that is characterized by aberrant perception. Predictive coding has recently emerged as a unifying framework that is not only able to describe the pathophysiology in ASD, but can also create testable hypotheses of altered autistic perception at the neuronal and synaptic level. Within this framework, core abnormalities of ASD are considered as a consequence of aberrant Bayesian inference, either due to a low precision of the prior belief ("hypo-Priors") and/or an overly precise likelihood distribution. Autistic perception is thus dominated by sensory input and less modulated by top-down regularizations of prior experience.In the present project, we aim to uncover the neural mechanisms underlying altered perceptual inference in ASD, in particular with regard to categorization processes. We will use a multimodal approach, combining computational modelling, behavioral assessments, functional neuroimaging, and MR spectroscopy. The project consists of three work packages. In work package 1, we study face identification using a computational modelling approach. Impaired face identification is a major behavioral deficit is ASD. We aim to develop a computational model that describes the learning of face identities at the behavioral level. We will then correlate central model parameters (e.g. prediction error signals) with neural activity to understand how aberrant perceptual inference is associated with alterations at the neural system level ("model-based fMRI"). In work package 2, we study the categorization of emotions, in which patients with ASD are typically impaired. We will use a newly developed emotion paradigm in which subjects have to categorize facial emotions of variable ambiguity. Using neural network analyses, we will investigate whether top-down modulations from prefrontal cortices to sensory face and emotion perception areas are attenuated in ASD patients, as predicted by the predictive coding theory. In work package 3, we yet take another approach and test hypotheses about the putative neurophysiological counterparts of aberrant perceptual inference in ASD. At the neural level, the predictive coding framework links core abnormalities of ASD with altered local GABAergic interneuron activity. This hypothesis will be tested with MR spectroscopy in core regions of the face perception and emotion processing network.