QUANTIMA: Quantitative imaging platform for the diagnosis, subtyping, staging and outcome prognosis in dementia

分析改变微结构组织完整性和细胞排列的条件，如痴呆症，对于个性化患者治疗和改善结果至关重要。目前，通常只有在疾病处于非常严重的晚期后才能识别这些受试者，这使得预后很差。在这个奖学金项目中，我的目标是开发一个创新的AI启用的磁共振(MR)图像分析平台，称为QUANTIMA，它能够生成新的非侵入性脑微结构生物标记物，并为改进诊断信息提供非侵入性工具，其功能与有创性和/或昂贵的技术如脑脊液腰椎穿刺术和PET一样强大。了解大脑各个组成部分的设计(形态)和排列(组织微结构)是破译其结构和功能的关键，更重要的是了解其在疾病中的退化/失调。该平台将利用基于扩散加权磁共振成像(DW-MRI)的先进微结构建模技术，提供微米级组织微结构的间接但非侵入性探头。然而，组织的微观结构高度复杂，而DW-MRI信号非常简单，因此从信号到微观结构的映射是不适定的。当前旨在克服这一挑战的计算建模技术使用数学模型，将DW-MRI信号映射到潜在的组织属性，通过拟合DW-MRI数据的每个体素的模型来估计这些属性。然而，这些方法受到许多限制，这些限制限制了它们的诊断能力和临床应用，例如对复杂细胞形态特征的低敏感性，对长MRI扫描时间的要求和临床环境中不常见的最先进的硬件，对试图模拟健康组织的细胞排列和生物学的预定义模型的依赖，因此限制了这些方法对疾病过程(特别是未观察到的)的适用性，以及未量化的模棱两可。为了克服这些限制，所提出的平台将使用先进的基于人工智能的优化和加速，能够生成组织微结构的定量估计，可与微结构计算建模、在克服其局限性的同时，通过：i)依靠临床上可应用于常用MR硬件(1.5T和3T扫描仪)的MR采集协议，ii)提供完全依赖于使用扩散MR信号进行观察的无模型方法，iii)能够估计不确定性、量化模糊性和结果的重要性。当诊断这些情况时，除了成像，医生还依赖从患者的病史、体检、实验室测试以及思维、日常功能和行为的特征变化中收集的信息。提供统一的解决方案，该平台还将能够进行多模式和多参数数据融合，利用来自这些数据源的信息(例如，认知测试)，允许更早和准确地进行痴呆症诊断、分型、分期和疾病轨迹预测；因此，支持个性化的治疗选择和改善的结果。该项目还将导致开发一种针对痴呆症的多参数优化(临床使用)MRI扫描协议，最大化通过短扫描获得的信息，并满足我打算开发的新型人工智能微结构建模技术的要求。在与伦敦AI Center for Value Based Healthcare、曼彻斯特大学、大曼彻斯特精神健康NHS基金会信托基金、GSK和索尔福德皇家NHS基金会信托基金的合作下，该平台将使用多方面的方法进行验证：使用回顾和预期患者的数据，并通过临床试点研究。

Analysis of conditions altering microstructural tissue integrity & cellular arrangement, such as dementia, is vital to personalise patient treatment & improved outcomes. Presently, it is often the case that these subjects are only identified after the disease is at a grossly advanced stage, making prognosis poor.In this fellowship I aim to develop an innovative AI-enabled magnetic resonance (MR) image analysis platform, known as QUANTIMA, capable of generating new non-invasive biomarkers of brain microstructure & providing non-invasive tools for improved diagnostic information as powerful as invasive and/or expensive techniques such as CSF lumbar puncture & PET. Understanding the design (morphology) & arrangement (tissue microstructure) of the brain's individual components is the key to deciphering both its structure & function, and more importantly its degeneration/dysregulation in diseases. The platform will make use of advanced diffusion-weighted magnetic resonance imaging (DW-MRI) based microstructure modelling techniques, providing an indirect but non-invasive probe of the tissue microstructure at the micrometre scale. However, tissue microstructure is highly complex while the DW-MRI signal is quite simple, so the mapping from signal to microstructure is ill-posed. Current computational modelling techniques aimed at overcoming this challenge use mathematical models, mapping the DW-MRI signal to underlying tissue properties, to estimate those properties by fitting the models per voxel of the DW-MRI data. Nevertheless, these methods suffer from a number of limitations that have restricted their diagnostic power & clinical adoption, such as poor sensitivity to features of complex cellular morphologies, requirements for long MRI scan times & state-of-the-art hardware not commonly available in clinical settings, reliance on predefined models of cellular arrangement & biology trying to mimic healthy tissue, therefore limiting the methods' applicability to disease processes (especially unobserved), & unquantified ambiguities.To overcome these limitations, the proposed platform will use advanced AI-based optimisations & accelerations, capable of generating quantitative estimates of tissue microstructure, comparable with the state-of-the-art in microstructure computational modelling, whilst overcoming their limitations by: i) relying on clinically achievable MR acquisition protocols applicable on commonly available MR hardware (1.5T & 3T scanners), ii) providing a model free approach that solely relies on the observations made using the diffusion MR signal, iii) is capable of estimating uncertainty, quantifying ambiguity & the significance of the results.When diagnosing these conditions, aside from imaging, doctors rely on information gathered from the patient's medical history, physical examination, laboratory tests, and the characteristic changes in thinking, day-to-day function, & behaviour. Offering a unified solution, the platform will also be able to perform multimodal & multi parametric data fusion, utilising information from such data sources (e.g.,cognitive tests), allowing for earlier & accurate dementia diagnosis, subtyping, staging, and disease-trajectory prediction; therefore, enabling personalised treatment selection & improved outcomes.The project will also lead to the development of a multiparametric dementia-specific optimised (for clinical use) MRI scan protocol, maximising information obtained over short scans & satisfying the requirements of the novel AI-enabled microstructure modelling technique that I intend to develop.In collaborations with the London AI Centre for Value Based Healthcare, University of Manchester, Greater Manchester Mental Health NHS Foundation Trust, GSK, and Salford Royal NHS Foundation Trust, the platform will then be validated using a multifaceted approach: using both retrospective & prospective patients data, and through a clinical pilot study.