Discovering Eye Tracking Biomarkers of ASD with Diagnostic and Prognostic Power

平均而言，患有自闭症的儿童直到4岁左右才被发现并接受治疗，比 首发症状和体征。即使当幼儿被诊断为自闭症时，父母和临床医生也知之甚少 以指导治疗决定或预测孩子未来几年的早期病程。需要研究来 发现客观的生物标记物，在早期高精度地检测ASD，指示相关的障碍亚型 以确定可定义的临床资料，并传达预后信息。这类生物标志物的发现 难以捉摸，部分原因是大多数研究使用了小样本量，没有包括非ASD延迟对比组 这对增强特异性至关重要。我们的提案计划通过利用眼球跟踪技术来填补这一空白 为了确定极小(12-36个月)ASD和非ASD幼儿的大样本的视觉注视模式 (n=225)可用于发现ASD的眼球跟踪生物标记物概况。这个 我们的眼球跟踪生物标志物的预后能力将通过将初始眼球跟踪分数与临床联系起来来确定 1-2年后的情况。考虑到ASD的异质性，单眼跟踪测试不太可能检测到 都是蹒跚学步的孩子。在这里，我们计划通过测试9个适合发育的电池的效用来解决这个问题， 短时间(每次约1分钟)的眼球跟踪测试，每次都能触及ASD症状的一个基础域，包括： 视觉社交注意、视线转移和听觉社交注意。我们的测试将客观地量化关键指标 例如社会形象与非社会形象中的整体注视水平、关节期间凝视变化的频率 注意力和凝视测试之后，使用独特的凝视应急技术，一个蹒跚学步的孩子 更喜欢听有韵律的、动情的、母亲式的演讲。我们的初步发现表明，有几个 我们建议的眼球跟踪测试具有极高的诊断准确率。然而，研究结果有时是 没有在科学上复制，我们通过提出并发的、独立的和准确的来主动地解决这个问题 (设备、软件、范例、程序)重复测试我们的每项眼球跟踪测试 华盛顿大学的独立幼儿队列(n=90名幼儿)。因此，在目标1中，我们将发现眼球跟踪 在早期(12-36个月)使用人工神经网络高精度检测ASD的生物标记物 UCSD，然后我们将在华盛顿大学独立测试其性能的复制。为了增强 可解释性，我们的方法将结合每个眼球跟踪测试的度量模式，以产生 自闭症风险评分(ARS)的范围从0到100，代表幼儿患自闭症的风险水平。在AIM 2中，使用 丰富的临床电池，捕捉每个幼儿的社交、语言、认知和症状严重程度 从标准化、家长报告和自由发挥测试的组合中，我们将识别出临床上有意义的 使用无偏网络聚类方法基于眼球跟踪的ASD亚型。在目标3中，我们将研究 通过识别基于眼球跟踪的个人资料的程度，我们的眼球跟踪测试电池的预后效用 在12-36个月时，可以预测1-2年后的社交、语言、认知和注意力能力。

Children with ASD, on average, are not identified and treated until around age 4‐years, several years beyond the first signs and symptoms. Even when toddlers are diagnosed as ASD, parents and clinicians have little information to guide treatment decisions or predict the early course of that child's next few years. Research is needed to discover objective biomarkers that detect ASD at early ages with high accuracy, indicate disorder subtypes linked to definable clinical profiles, and convey prognostic information. The discovery of such biomarkers have been elusive, in part, because most studies utilize small sample sizes and fail to include non‐ASD delay contrast groups which are essential to enhance specificity. Our proposal plans to fill this gap by leveraging eye tracking technology to determine if visual fixation patterns in a large sample of very young (12‐36 months) ASD and non‐ASD toddlers (n=225) from the general population can be used to discover an eye tracking biomarker profile of ASD. The prognostic power of our eye tracking biomarkers will be determined by linking initial eye tracking scores to clinical profiles 1‐2 years later. Given the heterogeneity in ASD, it is unlikely that a single eye tracking test would detect all toddlers. Here we plan to remedy this by testing the utility of a battery of 9 developmentally appropriate, short (~1‐minute each), eye tracking tests that each tap into a foundational domain in ASD symptoms including: visual social attention, gaze shifting, and auditory social attention. Our tests will objectively quantify key metrics such as overall fixation levels within social versus non‐social images, the frequency of gaze alterations during joint attention and gaze following tests and, using unique gaze contingent technology, the degree to which a toddler prefers to listen to prosodic, emotionally valent, motherese speech. Our preliminary findings suggest that several of our proposed eye tracking tests have extremely high diagnostic accuracy. Findings, however, are sometimes not replicated in science, and we proactively address this by proposing concurrent, independent and exact (equipment, software, paradigms, procedures) replication testing of each of our eye tracking tests within an independent cohort of toddlers at U. Washington (n=90 toddlers). Thus, in AIM 1 we will discover an eye tracking biomarker that detects ASD at early ages (12‐36 months) with high accuracy using artificial neural networks at UCSD, and then we will independently test replication of its performance at U Washington. To enhance interpretability, our approach will incorporate patterns of metrics from each eye tracking test to produce an Autism Risk Score (ARS) scaled from 0‐100 that represents the level of ASD risk for a toddler. In AIM 2, using a rich clinical battery that captures each toddler's social, language, cognitive and symptom severity profile derived from a combination of standardized, parent report, and free‐play testing, we will identify clinically meaningful eye‐tracking based subtypes of ASD using unbiased network clustering approaches. In AIM 3 we will examine the prognostic utility of our eye tracking test battery by identifying the degree to which eye tracking‐based profiles at ages 12‐36 months can predict social, language, cognitive and attention abilities 1‐2 years later.