In-vivo high-resolution χ-separation (chi-separation) at 7T

1866 

Abstract Submission 

Jiye Kim¹, Minjun Kim¹, Sooyeon Ji¹, Kyeongseon Min¹, Seong-Gi Kim^2,3, Jongho Lee¹

¹Department of Electrical and Computer Engineering, Seoul National University, Seoul, Korea, Republic of, ²Center for Neuroscience Imaging Research (CNIR), Institute for Basic Science (IBS), Suwon, Korea, Republic of, ³Department of Biomedical Engineering, Sungkyunkwan University, Suwon, Korea, Republic of

Jiye Kim  
Department of Electrical and Computer Engineering, Seoul National University
Seoul, Korea, Republic of

Minjun Kim  
Department of Electrical and Computer Engineering, Seoul National University
Seoul, Korea, Republic of

Sooyeon Ji  
Department of Electrical and Computer Engineering, Seoul National University
Seoul, Korea, Republic of

Kyeongseon Min  
Department of Electrical and Computer Engineering, Seoul National University
Seoul, Korea, Republic of

Seong-Gi Kim  
Center for Neuroscience Imaging Research (CNIR), Institute for Basic Science (IBS)|Department of Biomedical Engineering, Sungkyunkwan University
Suwon, Korea, Republic of|Suwon, Korea, Republic of

Jongho Lee  
Department of Electrical and Computer Engineering, Seoul National University
Seoul, Korea, Republic of

χ-separation¹ can separate paramagnetic and diamagnetic susceptibility distributions related to iron and myelin, respectively². A combination of this new technique and 7T imaging can benefit from increased SNR and susceptibility effects, potentially providing sub-millimeter resolution maps for brain structures^3-4. However, χ-separation requires an R2 map, which is challenging to acquire due to high SAR, B1 inhomogeneities, and long scan time (> tens of minutes). Recently, a neural network, χ-sepnet-R2*, was developed to remove the necessity for the R2 map but it is designed to work for the input of 3T R2*.⁵ Therefore, further work is required to utilize only 7T data (i.e., R2* and local field).
This study aims to produce in-vivo high-resolution χ-separation maps at 7T. To achieve this, we introduce a new pipeline that includes a novel deep neural network, an R2* 7T-to-3T conversion network, which converts 7T R2* to match 3T R2*, and QSMnet, χ-sepnet-R2*, and resolution generalization approach^6-7.

Ten healthy volunteers were scanned at 3 Tesla and 7 Tesla MRI (Siemens Tim Trio and Magnetom Terra, Erlangen, Germany). Sequences include multi-echo GRE, multi-echo SE, and MPRAGE at 3T, and multi-echo GRE at 7T (IRB-approved).
Since χ-sepnet-R2* is trained with 3T R2*, a 7T R2* map needs to be converted to that of 3T. This is achieved by an R2* 7T-to-3T conversion network. A dataset of ten R2* pairs at 3T and 7T was used (train:validation: test = 5:1:4). The neural network is trained to take patches of 7T R2* maps as input and patches of 3T R2* maps as labels, utilizing a 3D U-net with L1 and gradient loss as a loss function.
The pipeline for high-resolution χ-separation from 7T data is as follows: Initially, a QSM map is reconstructed from a local field map using QSMnet with the resolution generalization method. Secondly, the 7T R2* map is converted using the R2* 7T-to-3T conversion network. Lastly, the QSM map and converted R2* map are applied to χ-sepnet-R2* with resolution generalization, creating χ-separation maps.
For χ-separation maps, three methods are compared: i) proposed pipeline, ii) χ-sepnet-R2* with a linearly-scaled 7T R2* map (by B0) as R2* input, and iii) χ-sepnet-R2* with a 3T R2* map as R2* input (not practical because 3T data required). These results were evaluated with respect to the χ-separation-COSMOS maps at 3T as the reference. The comparison utilizes NRMSE, SSIM, and χ-separation atlas-based ROI anlaysis. The laminar profile of a middle frontal sulcus is conducted in the χpara, χdia, QSM, and R2* maps.

The proposed pipeline shows contrasts comparable to those of χ-separation-COSMOS. In contrast, maps from the linearly-scaled R2* input exhibit notably different contrasts and worse metrics.
Figure 2 illustrates the capability of the high-resolution χ-separation map to delineate fine structures in the in-vivo human brain such as lamina structures in the globus pallidus, primary visual cortex, transverse pontine fiber and fissures of the cerebellum. These structures are not as distinct at 3T, highlighting the advantage of 7T imaging. Additionally, high-resolution maps enable layer-wise cortex analysis. The results demonstrated consistency with previous findings, the discrepancy between the locations of the peak of QSM and the peak of χpara.

In this study, we introduced a novel deep neural network, R2* 7T-to-3T conversion network, to mitigate the discrepancy of R2* along the field strength and utilize χ-sepnet-R2*. By integrating the R2* 7T-to-3T conversion network, QSMnet, χ-sepnet-R2*, and resolution generalization, we generated high-resolution χ-separation maps from GRE data at 7T. Utilizing the proposed high-resolution χ-separation method, it is feasible to delineate more precise brain structures. The proposed method can be applied to layer-wise analysis and examine detailed structures related to iron and myelin accumulation.

Methods Development ¹

Cortical Cyto- and Myeloarchitecture ²

White Matter Anatomy, Fiber Pathways and Connectivity

Imaging Methods Other

Cortical Layers

HIGH FIELD MR

Machine Learning

MRI

Myelin

WHITE MATTER IMAGING - DTI, HARDI, DSI, ETC

Other - Susceptibility