Accelerated respiratory‐resolved 4D‐MRI with separable spatio‐temporal neural networks

Abstract

BackgroundRespiratory‐resolved four‐dimensional magnetic resonance imaging (4D‐MRI) provides essential motion information for accurate radiation treatments of mobile tumors. However, obtaining high‐quality 4D‐MRI suffers from long acquisition and reconstruction times.PurposeTo develop a deep learning architecture to quickly acquire and reconstruct high‐quality 4D‐MRI, enabling accurate motion quantification for MRI‐guided radiotherapy (MRIgRT).MethodsA small convolutional neural network called MODEST is proposed to reconstruct 4D‐MRI by performing a spatial and temporal decomposition, omitting the need for 4D convolutions to use all the spatio‐temporal information present in 4D‐MRI. This network is trained on undersampled 4D‐MRI after respiratory binning to reconstruct high‐quality 4D‐MRI obtained by compressed sensing reconstruction. The network is trained, validated, and tested on 4D‐MRI of 28 lung cancer patients acquired with a T1‐weighted golden‐angle radial stack‐of‐stars (GA‐SOS) sequence. The 4D‐MRI of 18, 5, and 5 patients were used for training, validation, and testing. Network performances are evaluated on image quality measured by the structural similarity index (SSIM) and motion consistency by comparing the position of the lung‐liver interface on undersampled 4D‐MRI before and after respiratory binning. The network is compared to conventional architectures such as a U‐Net, which has 30 times more trainable parameters.ResultsMODEST can reconstruct high‐quality 4D‐MRI with higher image quality than a U‐Net, despite a thirty‐fold reduction in trainable parameters. High‐quality 4D‐MRI can be obtained using MODEST in approximately 2.5 min, including acquisition, processing, and reconstruction.ConclusionHigh‐quality accelerated 4D‐MRI can be obtained using MODEST, which is particularly interesting for MRIgRT.

Accelerating 4D image reconstruction for magnetic resonance-guided radiotherapy

“…In particular regarding time-resolved 4D-MRI, the use of randomized projection-encoding [49] could be beneficial, but wouldn’t allow for straight-forward inter-slice parallelism. Compared with AI-based reconstruction schemes [50] , [51] , our proposed implementations provided an excellent agreement with the reference implementation.…”

Automatic lung segmentation of magnetic resonance images: A new approach applied to healthy volunteers undergoing enhanced Deep-Inspiration-Breath-Hold for motion-mitigated 4D proton therapy of lung tumors

“…While 4D-MRIs play an important role in the analysis, modeling and image guidance of dynamic respiratory motion [10] , [11] , [12] , [13] , [14] , motion artifacts in MRI lung reconstructions can be critical [15] . Although artificial intelligence has advanced motion compensation [16] , a preferred method for limiting image artifacts is MRI acquisition under breath-hold conditions, where “oxygen administration and patient coaching […] can increase breath-hold capability” [17] .…”

Paired conditional generative adversarial network for highly accelerated liver 4D MRI

“…Lately, Xu et al designed a 2D CNN-assisted RST), a variant of Video Swin Transformers (Liu et al 2021), supervised with a combination cost function of L1, peak signal-to-noise ratio (PSNR), and multi-scale structure similarity index measurement (MS-SSIM) (Wang et al 2003) for 4D MRI reconstruction and validates the algorithm on a 9× accelerated cardiac dataset (Xu et al 2023a). The RST architecture showed promising results in dynamic relationship learning and faster inference (<1 s), but struggled to retain finer spatial anatomies (Wang et al 2021, Terpstra et al 2023, Zhi et al 2023.…”