deepCEST: Non-invasive molecular MRI signatures from ultra-high-field to clinical translation

脑部的临床磁共振成像(MRI)对肿瘤的分期、活动性和治疗反应的洞察力仍然有限，因为传统的MRI只能检测到形态组织的变化。然而，在可见的形态变化发生之前，病变组织的特征是细胞微环境和新陈代谢发生改变。无创性MRI对这些分子变化的检测可能会给各种疾病的诊断、定性和治疗监测带来革命性的变化。有趣的是，新的MRI方法，如化学交换饱和转移(CEST)MRI产生了一系列的生理信息。CEST成像产生与pH、蛋白质含量和结构以及不同代谢物浓度相关的对比度。然而，到目前为止，由于标准临床MRI扫描仪的光谱选择性要低得多，并且MR信号减少，所以到目前为止，这种种类繁多的孤立信号可以在罕见的超高磁场(UHF)研究扫描仪中产生得最好。拟议项目的目标是通过生成一个全面的CEST数据库，并采用利用在超高频获得的先验知识的创新机器学习方法，克服这一磁场强度差距。这一进步将使功能丰富的CEST成像能够转化为临床领域的优势。为此，(I)将创建一个全面的多场强CEST人脑数据库，(Ii)将开发创新的机器和深度学习算法来对3T的CEST数据进行去噪和去卷积，(Iii)独特的信号模式识别将在脑肿瘤患者的临床试点研究中识别可利用的非侵入性MR生物标记物信号。这种临床MRI从形态成像到分子成像的扩展有可能提供更早、更具体的诊断，并使个性化治疗成为可能，以改善治疗策略和成功。

Clinical magnetic resonance imaging (MRI) of the brain yields still limited insight with respect to stage, activity and therapy response of tumors, because conventional MRI merely detects morphological tissue changes. However, before visible morphological changes occur, diseased tissue is characterized by an altered cellular micro-environment and metabolism. The detection of these molecular changes by non-invasive MRI could revolutionize the diagnosis, characterization and therapy monitoring of various diseases. Interestingly, novel MRI methods like chemical exchange saturation transfer (CEST) MRI yield a whole range of physiological information. CEST imaging generates contrasts that correlate with pH, protein content and structure, as well as concentration of different metabolites. However, this vast variety of isolated signals can until now be best generated at rare ultra-high magnetic field (UHF) research scanners due to the much lower spectral selectivity and reduced MR signal in standard clinical MRI scanners. The objective of the proposed project is to overcome this magnetic field strength gap by generating a comprehensive CEST data base, and by employing innovative machine learning approaches which exploit prior knowledge acquired at UHF. This advancement will allow feature–rich CEST imaging to be translated to clinical field strengths. To do so (i) a comprehensive multi-field strength CEST database oft he human brain will be created, (ii) innovative machine- and deep-learning algorithms will be developed to denoise and deconvolve CEST data at 3T, and (iii) unique signal pattern recognition will identify exploitable non-invasive MR biomarker signatures with respect to disease type, stage, and activity, as well as therapy response in clinical pilot studies of brain tumor patients. This extension of clinical MRI from morphologic to molecular imaging has the potential to provide much earlier and more specific diagnostics and it can empower personalized therapy to improve treatment strategy and success.