Quantitative multiparametric magnetic resonance imaging and its application on endometrial cancer

Abstract

Non-invasive preoperative assessment of endometrial cancer (EC) is valuable for the personalized treatment planning. However, quantitative Magnetic Resonance Imaging (MRI), including T1, T2 and proton density (PD) maps, have not been applied to evaluate the histopathological characteristics of EC. In addition, diffusion quantification has not been integrated into the conventional quantitative multi-parametric MRI (qMP-MRI) framework. The purpose of this thesis was to investigate the diagnostic performance of qMP-MRI-based quantitative analysis, and simultaneously quantify the relaxation times and diffusion coefficient based on the framework of Magnetic Resonance Fingerprinting (MRF).

Firstly, the association of histopathological characteristics of EC and qMP-MRI metrics, including T1, T2, PD from synthetic MRI and apparent diffusion coefficient (ADC) from conventional diffusion-weighted imaging (DWI), was investigated. It was found that aggressive histopathological features of EC, such as the presence of myometrial invasion (MI) and lymphovascular invasion (LVSI), high tumor grade, and high International Federation of Gynecology and Obstetrics (FIGO) stage, were correlated to lower T2, PD and ADC values. How-ever, tumor heterogeneity was not considered in the first study, since the mean value of signal intensities inside the VOI was used to represent the whole tumor’s characteristic.

Secondly, radiomics has been widely applied to the quantitative analysis of routine MRI to capture the spatial heterogeneity of tumor. Radiomics features were extracted from T1, T2 and PD maps, and further selected on some criteria to build a random forest model for the preoperative assessment of EC. Limited by the small sample size, the diagnostic performance was represented by the average evaluation scores from cross-validation sets. It was found that the model demonstrated clinically acceptable performance in differentiating histopathological characteristics, including MI, tumor grade and LVSI.

Lastly, a partial diffusion-weighted MRF (DW-MRF) method was designed to measure T1, T2, PD and ADC in a single sequence. The pulse sequence was composed of two parts: conventional MRF for the first 700 time points, and diffusion-weighted SSFP (DW-SSFP) for the last 300 time points. The subspace re-construction was applied to conventional MRF and DW-SSFP acquisitions separately to generate full-resolution image series in all time points. The temporal basis used in subspace reconstruction was estimated from low-resolution images, which were reconstructed from raw data in the fully sampled central k-space. It was found that the measurement accuracies were relatively high in the phantom experiment. High-resolution images can be generated in the in-vivo experiments without the impact of shot-by-shot phase variations.