Automated quality control of MRI scans with machine learning

通过机器学习对 MRI 扫描进行自动质量控制

• 批准号: 2852054
• 金额: --
• 依托单位: University College London
• 依托单位国家: 英国
• 项目类别: Studentship
• 财政年份: 2022
• 资助国家: 英国
• 起止时间: 2022 至 无数据
• 项目状态: 未结题
• 来源: https://gtr.ukri.org/projects?ref=studentship-2852054
• 关键词: Automated; quality; control; MRI; scans

Magnetic Resonance Imaging (MRI) is a widely used 3-D medical imaging technique in modern healthcare systems. However, MRI scans are more prone to corruption due to patient motion than other types of imaging, such as CT. As low-quality scans may reduce the accuracy of subsequent analysis, quality control (QC) process that can flag these scans is often required. Currently, this process is typically accomplished by human experts via visual inspection. With the volume of MRI scans rising rapidly, expert-based QC is becoming increasingly impractical. Hence, there is considerable interest in developing automated QC tools. The existing automated QC frameworks are usually machine learning (ML) based. However, due to a lack of training data, their performance is currently limited. This project aims to develop methods to simulate realistic MRI scans to address this challenge. These methods will provide a vast source of simulated yet realistic MRI scans to develop new and improved ML-based automated QC tools. The main potential impact is that the developed automated QC tools can allow researchers and clinicians alike to focus their valuable time on analysing high-quality MRI data. The aim of the project is to develop ML-based tools for automated MRI quality control. To address this aim, this project has the following two objectives: 1. To develop an MRI scan simulator that can produce synthetic but realistic MRI scans to increase data availability for training ML-based techiques. 2. To incorporate synthetic scans into training to improve the performance of ML-based MRI quality control frameworks.

磁共振成像(MRI)是现代医疗系统中广泛使用的三维医学成像技术。然而，与其他类型的成像(如CT)相比，MRI扫描更容易因患者运动而损坏。由于低质量扫描可能会降低后续分析的准确性，因此通常需要能够标记这些扫描的质量控制(QC)过程。目前，这一过程通常由人类专家通过视觉检查来完成。随着核磁共振扫描量的迅速增加，基于专家的质量控制正变得越来越不现实。因此，人们对开发自动化的质量控制工具很感兴趣。现有的自动化质量控制框架通常是基于机器学习的。然而，由于缺乏训练数据，他们目前的表现有限。该项目旨在开发模拟真实核磁共振扫描的方法来应对这一挑战。这些方法将提供大量模拟但真实的MRI扫描，以开发新的和改进的基于ML的自动化质量控制工具。主要的潜在影响是，开发的自动化质量控制工具可以让研究人员和临床医生将宝贵的时间集中在分析高质量的MRI数据上。该项目的目标是开发基于ML的工具，用于自动化MRI质量控制。为了达到这一目标，本项目有以下两个目标：1.开发一个MRI扫描模拟器，它可以产生合成的但真实的MRI扫描，以提高数据的可用性，以培训基于ML的技术。2.将合成扫描纳入训练，以提高基于ML的MRI质量控制框架的性能。