Computational Models to Study Auditory Processing and Learning Disorders in Children

这项工作对学习障碍儿童的听觉处理及其相应的治疗进行了建模，建立、验证和实验了人脑听觉神经系统中音频信号的处理的计算模型。在这项工作中，将完成以下任务：建模和验证从听神经纤维到听觉皮质的通路上的“单一神经元通道(SNC)”将SNC扩展到可能包括听觉通路上的所有神经元，并在高性能计算机上验证所得到的高度并行的模型，以及沿着通路的反馈；在计算模型中将听觉通路的疾病(障碍)建模为神经元回路中的故障，并研究其对通路沿途信号处理的影响；以及确定“计算疾病”模型的神经元活动与通过对学习障碍儿童进行实验观察到的人类脑干反应的相关性。该项目的长期目标是将言语和听力科学家通过人类听觉系统实验获得的与声学处理流有关的知识捕获到计算模型中，该项目旨在实现以下短期目标：理解听觉处理的哪些部分应该包括在模型中；与言语和听力科学的研究人员磋商，建立神经网络模型；以及通过对实验观察到的数据运行计算机模拟来验证模型。这样的模型有望预测脑干对通过人耳进入的音频信号的反应，并在耳蜗中将其划分为更精细的频谱分量，脉冲从耳蜗处通过听神经和脑干中几个专门的核团传递到听觉皮质。音频信号分成两个不同的“什么”和“哪里”流在听觉皮质或在通往它的途中，这一事实推动了这项研究。虽然存在许多听觉系统的计算模型，但提出的研究的新颖性在于。在计算模型中包括腹侧流和背侧流。通过在患有学习障碍的儿童身上观察到的听觉脑干反应来验证该模型。对不同听觉通路的理解的研究在治疗各种言语和沟通障碍方面具有广泛的意义，例如儿童的语言障碍。广泛的影响：经过验证的听觉皮质计算模型具有执行基于计算机的实验和试验的潜力，以了解拟议的治疗此类障碍的影响。这样的实验可以是快速的，相对便宜的，并且可以作为进行涉及人和动物的实验的先驱。此外，学生还可以获得该领域的专业知识。这项研究扩展了神经元计算建模的学科，应该有助于理解儿童学习障碍的原因，从而能够开发出相应的治疗方法。然而，目前还不清楚这种计算模型需要什么样的训练来模拟实验观察到的行为，然后提出可以通过实验验证的行为。计算模型将用来回答“如果”的问题。

This work, modeling auditory processing in children with learning disabilities and its corresponding treatments, builds, validates, and experiments with computational models the processing of audio signals in the auditory nervous system of the human brain. The following tasks will be completed during this work:. Modeling and validation of a "single neuronal channel (SNC)" along the pathway from the auditory nerve fiber all the way to the auditory cortex;. Expansion of the SNC to include possibly all neurons along the auditory pathway and validation of the resulting highly parallel model, with feedback along the pathway, on a high performance computer; . Modeling of diseases (disorders) of the auditory pathway as faults in the neuronal circuitry in the computational model and studying their impact on signal processing along the pathway; and. Determining the correlation of the neuronal activity of the "computationally diseased" model with the human brainstem response observed through experimentation with learning-impaired children.With a long-term goal for capturing into a computational model the knowledge related to the acoustic processing streams acquired by speech and hearing scientists through experiments with the human auditory system, the project aims to accomplish the following short-term goals: . Understanding what portions of the auditory processing ought to be included in the model;. Building a neural network model in consultation with researchers in speech and hearing sciences; and. Validating the model by running computer simulations against experimentally observed data.Such a model was expected to predict brainstem responses to audio signals that come in via the human ear and partition into its finer spectral components in the cochlea from where the impulses travel via the auditory nerve and several specialized nuclei in the brainstem to the auditory cortex. The fact that the audio signals split into two distinct "what" and "where" streams in the auditory cortex or en-route to it motivates the research. Although many computational models of the auditory system exist, the novelty of the proposed research lies in. Including the ventral and dorsal streams in the computational model and. Validating the model against auditory brainstem responses observed in children with learning disorders.Research into the understanding of the different auditory pathways has wide ranging implication in the treatment of various speech and communications disorders such as language impairment in children.Broader Impacts: Validated computational models of the auditory cortex carry the potential for performing computer-based experiments and trials to understand the impact of proposed treatments for such disorders. Such experiments can be fast, relatively inexpensive, and serve as a precursor to conduct experiments involving humans and animals. Additionally, students also gain expertise in the area. The research expands the discipline of neuron-computational modeling and should contribute in understanding the causes of learning disabilities in children, thus enabling development of the corresponding treatment. However, it is unclear as to what type of training will be needed for such computational model to mimic the experimentally observed behaviors and then to suggest behaviors that can be verified experimentally. The computational model will be used to answer "what if" questions.