Normal tidal volume for clin sims1/3/2024 13 proposed a 4D MRI reconstruction framework for liver MRI with arbitrary breathing motion. On the downside, most of these methods have long beforehand acquisition times of up to 60 min and are not real-time capable due to long reconstruction times of tens of seconds 12. 11 train a deep learning network without ground truth to remove reconstruction artifacts from under-sampled phase-resolved 4D MRI.ĤD MRI methods of the other class can reconstruct arbitrary/irregular breathing and are mainly based on clinically available MRI sequences. 10 use the diaphragm in sagittal slices as an anatomic feature to sort axial slices into ten breathing phases. 9 cluster data slices without using navigator slices by comparing different surrogate signals. 8 acquire coronal slices and extract an image-based self-sorting signal performing rigid registration of the diaphragm to sort the image data into ten respiratory phases retrospectively. 7 repeatedly sample the k-space center line as a self-gated motion surrogate and retrospectively bin k-space data into different respiratory phases. 6 implemented a continuous spoiled gradient echo sequence with 3D radial trajectory and 1D self-gating for respiratory motion detection to retrospectively sort data into different respiratory phases. 5 removed the need for navigator frames by directly comparing neighboring slices using mutual information to reconstruct one breathing cycle. 4 acquire sagittal or coronal slices and retrospectively stack them in a two-pass approach into ten respiratory phase volumes. They reconstruct four respiratory states of one breathing cycle. 3 use single-shot acquisition with parallel imaging and partial k-space imaging to improve acquisition speed. 2 retrospectively sort axial slices into respiratory phases using the body area as an image-based internal respiratory surrogate. They are mainly based on sequence programming and unique k-space sampling designs, and the acquisition usually takes around 5 min. The former can reconstruct a fixed number of phases of a single breathing cycle (usually 10 or fewer phases) and can not account for arbitrary/irregular breathing. Related work 4D MRI methods can be classified as either respiratory phase-resolved or time-resolved (see Table 1). That could soon change with further advances in deep learning, as we will show in our work. Consequently, the effective application of 4D MRI in the intervention room remains challenging. Additionally, there are limits to the specific absorption rate (SAR) allowed during MRI imaging, and these limits are likely to be exceeded during prolonged imaging. Although a recent study demonstrated promising results using a deep learning (DL) approach with only 24 min of training data 1, this timeframe is still impractical for routine clinical settings where time is crucial. However, acquiring real-time 4D MRIs of a large target region during an intervention is currently not feasible due to the need for a significant amount of reference data beforehand. Real-time 4D MRI imaging in MRI-guided procedures holds the potential to address this issue. Insufficient compensation for irregular organ motion during image-guided interventions is a significant problem that can lead to inaccuracies in instrument navigation and compromised treatment outcomes. This shows that TL + Ens significantly reduces beforehand acquisition time and improves reconstruction quality, rendering it a key component in making 4D MRI clinically feasible for the first time in the context of 4D organ motion models of the liver and beyond. The smaller the target domain data amount, the larger the effect. Comparing ensemble of fine-tuned models (N = 10) with directly learned models, we report significant improvements (P < 0.001) of the root mean squared error (RMSE) of up to 12% and the mean displacement (MDISP) of up to 17.5%. For that the data base was split into 16 source and 4 target domain subjects. We evaluate four approaches: pre-trained models from the source domain, models directly trained from scratch on target domain data, models fine-tuned from a pre-trained model and an ensemble of fine-tuned models. This work shows that transfer learning (TL) combined with an ensembling strategy can help alleviate this key challenge. Deep learning-based (DL) 4D MRI approaches promise to overcome these shortcomings but are sensitive to domain shift. Current 4D reconstruction techniques are unsuitable for most interventional settings because they are limited to specific breathing phases, lack temporal/spatial resolution, and have long prior acquisitions or reconstruction times. Time-resolved volumetric magnetic resonance imaging (4D MRI) could be used to address organ motion in image-guided interventions like tumor ablation.
0 Comments
Leave a Reply.AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |