Making the Invisible Visible: a Multi-Scale Imaging Approach to Detect and Characterise Cortical Pathology

大脑的许多疾病，包括癫痫、痴呆症、多发性硬化症和精神疾病，都涉及大脑最外层的皮质。使用常规MRI作为诊断的一部分，或研究其病理生理学，一个关键的挑战是敏感性，即皮质异常可能很小和/或在形态上很细微-因此被遗漏。即使检测到异常信号，也不可能说出是什么驱动了这种信号的变化，例如，细胞大小/形状/密度的差异。目前，这种信息只能通过切开组织并在显微镜下进行检查来获得。然而，核磁共振物理学的最新进展有望直接检测和表征皮质中迄今不可见的组织异常。超强磁场提供高得多的分辨率图像，而超强的“梯度”提供了对组织微结构属性的敏感性，如细胞密度、大小和形状，这在传统磁共振成像中是看不到的。这些技术已经被应用于白质，但它们在皮质中的应用在很大程度上仍未被探索。在这里，我们将提供一个原则证明，通过结合MRI硬件、物理、显微镜、数学建模和人工智能(AI)的最新技术，我们不仅能够比以往任何时候都能看到更多患者的皮质异常，而且还能获得原本需要侵入性活检的相同类型的细胞构成信息。为此，我们将从现有的显微镜数据集开始，构建皮质组织的超逼真3D计算模型，并更改其属性以模拟我们在疾病中看到的情况，并了解这将如何在不同的设置下改变来自MRI扫描仪的信号。这将允许我们选择最大限度地提高对疾病敏感性的扫描仪设置，并了解我们的数学模型的哪些部分最能提供有关病理的信息，例如，考虑细胞大小/形状/密度。使用人工智能，我们将结合来自超强磁铁的空间分辨率和来自超强梯度的微观结构敏感性，创建具有前所未有细节的新的“混合”MRI图像。有了一个完全优化的磁共振成像方案&捕捉了在常规磁共振成像中不可见的关键皮质特征的模型，我们将扫描健康的个体，以了解每个特征中有多少典型的变异。我们假设皮质病理学将导致一些模型参数落在这个标准范围之外，使我们能够自动检测它们。为了测试我们的假设和验证我们的方法，我们将在一种癫痫形式的患者身上测试我们的技术，这种癫痫与大脑皮质结构的高度局部性异常相关，称为局灶性皮质发育不良(FCD)，在切除致痫组织的手术之前。这种组织将在实验扫描仪中进行长时间成像，比人类核磁共振扫描仪对组织微结构差异的敏感性更高。使用人工智能，我们将使用这些更详细的图像来增强在活着的人脑中收集的图像的细节。使用显微镜观察样本，然后我们将产生组织学‘地面真相’，并再次使用人工智能，更新我们的模型和采集协议，以最大限度地提高我们管道的灵敏度和准确性。最后，我们将在常规MRI上没有明显疾病(但症状与皮质异常一致)的患者身上测试我们的方法。在预测到异常组织的地方，我们将尝试用电子记录和显微镜对任何切除的组织进行验证。最终，对标准临床MRI不可见的病理的检测可能会指导更准确的网络询问，从而改善手术结果，扩大适合手术的患者群体，并洞察患有影响大脑皮质疾病的人的相关认知和行为并存。

Many diseases of the brain, including epilepsy, dementia, multiple sclerosis & mental health disorders, involve its outermost layer, the cortex. A key challenge in using conventional MRI as part of their diagnosis, or to study their pathophysiology, is sensitivity, i.e., cortical abnormalities may be small and/or subtle in their morphology & therefore missed. Even if abnormal signal is detected, it is impossible to say what drives such signal changes, e.g., differences in cell size/shape/density. Currently such information can only be obtained by cutting out the tissue & examining it under a microscope. Recent advances in MRI physics, however, hold the promise of detecting & characterising heretofore invisible tissue abnormalities directly in the cortex. Ultra-strong magnetic fields give much higher resolution images, while ultra-strong 'gradients' provide sensitivity to tissue 'microstructure' properties such as cell density, size & shape that cannot be seen on conventional MRI. Such technologies have been applied to white matter, but their use in cortex remains largely unexplored. Here, we will provide a proof-of-principle that, by combining the latest in MRI hardware, physics, microscopy, mathematical modelling & artificial intelligence (AI), we will not only be able to see cortical abnormalities in more patients than ever before, but also obtain the same kind of information about cellular make-up that would otherwise require invasive biopsy. To this end, beginning with existing microscopy datasets, we will build ultra-realistic 3D computational models of cortical tissue & change their properties to mimic what we see in disease, & learn how this would change the signals from the MRI scanners under different settings. This will allow us to select the scanner settings that maximise sensitivity to disease & to learn which parts of our mathematical models are most informative about the pathology, e.g., accounting for cell size/shape/density.Using AI, we will combine the spatial resolution from ultra-strong magnets & the microstructural sensitivity from ultra-strong gradients, to create new 'hybrid' MRI images with unprecedented detail. With a fully optimised MRI protocol & models that capture the key cortical features that are otherwise 'invisible' on conventional MRI, we will scan healthy individuals to learn how much typical variation there is in each feature. We hypothesise that cortical pathology will lead to some model parameters falling outside of this normative range, allowing us to detect them automatically.To test our hypothesis & validate our approach, we will trial our technique in patients with a form of epilepsy that is associated with highly localised abnormalities in the structure of the cortex, called 'focal cortical dysplasia' (FCD), prior to surgery to remove epileptogenic tissue. This tissue will undergo prolonged imaging in an experimental scanner with even greater sensitivity to differences in tissue microstructure than human MRI scanners. Using AI, we will use these more detailed images to enhance the detail of the images collected in the living human brain. Using microscopy of the sample, we will then produce a histological 'ground truth' and, again using AI, update our models & acquisition protocol to maximise sensitivity & accuracy of our pipeline. Finally, we will test our approach on patients with no visible disease on conventional MRI (but where symptoms are consistent with cortical abnormality). Where abnormal tissue is predicted, we will attempt validation with electrical recordings & microscopy of any resected tissue. Ultimately, the detection of pathology invisible to standard clinical MRI may direct more accurate network interrogation thereby improving surgical outcomes, expand the population of patients suitable for surgery, & yield insight into associated cognitive and behavioural co-morbidities in people with diseases affecting the cortex.