Inter-scanner and inter-patient variability for large-scale MRI studies

核磁共振在NHS和其他卫生服务机构中被用于诊断和监测每一种主要疾病，包括癌症、心脏病、痴呆症和中风(英国四大杀手)。然而，临床MRI数据通常是定性的。这导致在不同扫描仪和不同时间点上获得的图像有相当大的变异性，这不是由于潜在的病理。这意味着重要的生物变异性或新疗法的效果被扫描仪之间的差异所掩盖--这是药物试验和基于人工智能的推理的一个主要困惑。扫描仪间的可变性极大地增加了所需的数据量，增加了研究成本和复杂性，在某些情况下还不必要地将有希望的方法和治疗方法一起排除。本项目旨在将扫描仪间的可变性与患者间的可变性分开，使用来自UCL和Alliance Medical的MRI图像和QA数据的大型数据库来表征每种方法的效果。我们将执行过程控制分析，以建立图像的时间相关校准，并利用这一点来校正扫描仪变异性的影响。然后，我们将对图像执行高级群集和可变性分析，以确定患者群体中剩余的变化驱动因素。这提供了一大批健康个体的基线，可以用来支持进一步标准化对照和患者图像之间的比较管道的工作。

MRI is used across the NHS and other health services in the diagnosis and monitoring of every major disease including Cancer, Heart Disease, Dementia, and Stroke (the four largest killers in the UK). Clinical MRI data, however, is typically qualitative. This leads to considerable variability in images acquired on different scanners and between different points in time which is not due to the underlying pathology. This means that important biological variability or the effect of a new therapy is obscured by differences between scanners - a major confound for drug trials and AI-based inference. Inter-scanner variability greatly increases the amount of data required, increasing study costs and complexity and in some cases needlessly eliminating promising approaches and therapies altogether.This project aims to separate inter-scanner variability from inter-patient variability, characterising the effect of each using large databases of MRI images and QA data from UCL and Alliance Medical. We will perform process control analysis to establish time-dependent calibration for the images and use this to correct for the effects of scanner variability. We will then perform advanced clustering and variability analysis on the images to identify the remaining drivers of change in the patient population. This provides baselining of a large cohort of healthy individuals which can be used to underpin further work in standardising the pipeline of comparison between control and patient images.