Robust quality-controlled quantitative stress perfusion cardiac MRI

博士项目的目的：灌注心脏MRI的全自动评估可能会因图像质量低、造影剂注射失败或对压力缺乏反应而出现可靠性问题。该项目将开发一系列支持AI的质量控制流程，使灌注心脏MRI在临床上更加强大和可靠。项目描述：负荷灌注心脏MRI未得到充分利用的主要原因是图像的视觉评估高度依赖于操作员的培训水平[1]。到目前为止，负荷灌注心脏MRI的定量分析仍然主要是一种研究工具，但其临床翻译将是有利的，因为它可以自动化，使准确和用户独立的心肌灌注评估。我们小组最近也证明了定量负荷灌注CMR的独立预后价值[2]。然而，这项工作仍然涉及手动交互的几个步骤，包括心肌的分割。定量负荷灌注CMR的自动化分析有可能彻底改变疑似冠状动脉疾病患者的管理。我们的团队率先使用深度学习来实现负荷灌注CMR的自动定量分析[3]。大规模部署该解决方案的问题是分析的目视检查和质量控制不可行。如果不清楚哪些病例分析失败，则临床医生可能会得到失败或不准确的测量结果，因此得出错误的结论。我们的目标是通过使用深度神经网络开发全自动图像质量控制（QC）工具来克服这一限制。深度学习已被广泛应用于医学成像领域，并已成为许多处理任务的事实标准。心脏MRI图像分析的趋势也遵循类似的路线，深度学习现在用于从重建到检测和分割任务，再到自动诊断和自动化的所有任务。为此，我们设想开发三个QC流程，以进行开发、验证并集成到临床工作流程中。动脉输入功能（AIF）分析：机器学习方法将用于对造影剂注射到患者体内是否正确进行分类。压力反应分析：部分患者对压力源药物的注射没有预期的反应。将对算法进行训练和验证，以识别这些患者。我们之前开发的自动化处理管道可能会失败，因为心脏被预测在错误的位置，或者心脏的预测形状不如预期。将开发进一步的神经网络来标记这些情况。

Aim of the PhD Project:Fully automatic assessment of perfusion cardiac MRI can suffer from reliability issues due to low image quality, failed contrast injections, or a lack of response to stress. This project will develop a range of AI-enabled quality control processes to make perfusion cardiac MRI more robust and reliable in the clinic.Project Description:The main cause of the underutilisation of stress perfusion cardiac MRI is that visual assessment of the images is highly dependent on the level of training of the operators [1]. As yet, quantitative analysis of stress perfusion cardiac MRI remains primarily a research tool but its clinical translation would be advantageous as it can be automated, enabling accurate and user-independent assessment of myocardial perfusion. Our group has also recently demonstrated the independent prognostic value of quantitative stress perfusion CMR [2]. However, this work still involved several steps of manual interaction, including the segmentation of the myocardium. The automated analysis of quantitative stress perfusion CMR has the potential to revolutionise the management of patients with suspect coronary artery disease.Our group has pioneered the use of deep learning to achieve the automatic quantitative analysis of stress perfusion CMR [3]. The problem of deploying this solution at large scale is the visual inspection and quality control of the analysis is not feasible. If it is not obvious for which cases the analysis has failed, then clinicians may be presented with failed or inaccurate measurements and, therefore, draw the wrong conclusions. We aim to overcome this limitation by developing a fully automated image quality control (QC) tool using deep neural networks. Deep learning has been widely adopted in the field of medical imaging and has become the de facto standard for many processing tasks. Trends in cardiac MRI image analysis have followed a similar route where deep learning is now used for everything from reconstruction to detection and segmentation tasks to automating diagnostics and prognostics.To this end, we envisage the development of three QC processes to be developed, validated, and integrated into the clinical workflow.Analysis of the arterial input function (AIF): Machine learning methods will be used to classify whether the injection of the contrast agent into the patient was done correctly.Analysis of stress response: Some patients do not respond as expected to the injection of the stressor drug. Algorithms will be trained and validated to identify these patients.Detection of failed cases. Our previously developed automated processing pipeline can fail, in that the heart is predicted in the wrong location or the predicted shape of the heart is not as expected. A further neural network will be developed to flag these cases.