Synergistic EEG and machine learning enhanced low-dose PET for objective diagnostics of speech comprehension in cochlear implant users

拟议的资金将用于实现听觉系统功能研究的基本方法改进，PET/EEG（正电子发射断层扫描/脑电图）和解释人工耳蜗植入后的可变语音理解性能。为了改善听觉系统的O-15水PET激活研究，将确定在不影响诊断性能的情况下可以使用的最小放射性量。为此，将使用一个程序，该程序基于测量的PET研究生成符合性减少的数据集-模拟放射性量较低的研究。现代PET设备的高灵敏度表明，迄今为止，活动量显著低于通常水平的研究是可能的。此外，将使用机器学习技术在正弦图水平对低计数研究进行降噪。这将涉及数据的统计建模，而没有广泛的良好控制的数据用于训练（证据下限作为学习目标，ELBO），并将与深度神经网络（DNN）相结合。将使用统计参数映射（SPM）分析PET数据（有/无减少或去噪），并比较结果。此外，出于验证目的，将基于模拟脑体模（XCAT）生成具有和不具有减少或去噪的数据，并将比较相应的结果。除了提供明显减少辐射暴露的证据外，计划中的研究还将证明使用现代高分辨率PET设备对脑干/丘脑中的中枢听觉通路进行可靠成像。对于自动解剖学描绘，将基于组织形态学和MRI数据生成脑干和丘脑（例如下丘）中听觉区域的感兴趣体积（VOI）模板。多模态方法将客观地测量具有高空间分辨率（PET）和时间分辨率（EEG）的语音的中央处理。具体而言，将在句子辨别任务的刺激下测量晚期听觉诱发电位（N400），并在语音处理（PET）期间激活的区域的上下文中进行空间分配。这将使我们更好地了解不同言语理解能力的CI用户的言语处理资源分配和策略。背景噪声是影响CI使用者言语理解的主要因素，本文将研究背景噪声对言语可懂度和言语加工过程中皮层激活的影响。通过所有这些，将开发用于评估和优化人工耳蜗植入后听觉康复的客观程序。减少辐射暴露的证据应该鼓励在其他环境中使用PET。

The proposed funding will be used to achieve fundamental methodological improvements in functional studies of the auditory system with PET/EEG (positron emission tomography / electroencephalography) and explanations for variable speech understanding performance after cochlear implantation. To improve O-15 water PET activation studies of the auditory system, the minimum amount of radioactivity that can be used without compromising diagnostic performance will be determined. For this purpose, a program will be used which generates data sets with reduced number of coincidences based on measured PET studies - simulating studies with lower amounts of radioactivity. The high sensitivity of modern PET devices suggests that studies with significantly lower amounts of activity than has been usual so far are possible. In addition, the low-count studies will be denoised at the sinogram level using machine learning techniques. This will involve statistical modeling of the data without extensive well-controlled data for training (evidence lower bound as learning objective, ELBO) and will be combined with deep neural networks (DNN). The PET data (with/without reduction or denoising) will be analyzed using statistical parametric mapping (SPM) and the results compared. Furthermore, for validation purposes, data with and without reduction or denoising will be generated based on a simulated brain phantom (XCAT) and the respective results will be compared. In addition to providing evidence for a significantly reducible radiation exposure, the planned study will demonstrate the reliable imageability of the central auditory pathway in the brainstem/thalamus with modern high-resolution PET devices. For automated anatomical delineation, a volume of interest (VOI) template of auditory regions in brainstem and thalamus (e.g. inferior colliculus) will be generated based on histomorphometric and MRI data. The multimodal approach will objectively measure central processing of speech with high spatial (PET) and temporal resolution (EEG). Specifically, late auditory evoked potentials (N400) will be measured under stimulation with a sentence discrimination task and spatially assigned in the context of regions activated during speech processing (in PET). This should allow a better understanding of the resource allocation and strategies of speech processing in CI users with different speech comprehension performance. As a major factor in the speech comprehension of CI users, the influence of background noise on speech intelligibility and cortical activation during speech processing will be investigated. Through all this, objective procedures for the evaluation and optimization of auditory rehabilitation after cochlear implantation will be developed. Evidence of the reducibility of radiation exposure should encourage the use of PET in other settings.