Quality evaluation of MR images****

磁共振成像（MRI）容易受到严重伪影的影响，这些伪影会影响图像质量，导致诊断价值降低、重复成像和增加手术成本。在MRI中出现伪影，部分原因是由于患者的运动，扫描仪的固有限制和不正确的扫描参数。运动（或运动）伪影是MRI技术固有的问题，它会降低图像质量，并可能导致误解，特别是在低信噪比的MRI采集中。然而，在获取期间或扫描完成后立即识别公共工件是可能的。****这个项目的目标是开发一个“MR图像质量评估工具”来自动检测MRI伪影。这种工具将通过纳入飞利浦质量评估产品和MRI系统增加实质性的临床价值。我们建议使用深度学习算法应用人工智能（AI）来识别和检测MRI伪影。文献中的一些可能的选择是使用预训练的卷积神经网络（CNN）进行迁移学习。另一种选择是开发一种生成对抗网络或GAN，它已被证明在有限数据存在下具有鲁棒性。****开发的模型将在医院和公开可用的数据集上进行测试，并使用人工诱发的工件。如果成功，深度学习解决方案可以在未来扩展到其他成像模式。******

Magnetic resonance imaging (MRI) is susceptible to severe artifacts that hinder image quality leading to reduced diagnostic value, repeated imaging, and increased procedure costs. Artifacts appear during MRI in part due to patient motion, inherent limitations of the scanner, and incorrect scanning parameters. Movement (or motion) artifacts are an inherent problem to MRI technology that degrades image quality and may lead to misinterpretation especially in MRI acquisitions with low signal-to-noise ratios. However, it may be possible to recognize common artifacts during acquisition or immediately after a scan has completed.****The goal of this project is to develop an "MR image quality assessment tool" to automatically detect MRI artifacts. Such a tool will add substantial clinical value through incorporation into Philips quality assessment products and MRI systems. We propose to apply artificial intelligence (AI) using a deep learning algorithm to identify and detect MRI artifacts. Some possible options from the literature are to use transfer learning using a pre-trained convolutional neural network (CNN). Another option is to develop a Generative Adversarial Network or GAN that has been shown to be robust in the presence of limited data. ****The developed model will be tested on both hospital and publicly available datasets with artificially induced artifacts. If successful, the deep learning solution can be extended to other imaging modalities in the future.******