Sonic DL™ is expanding its capabilities to include 3D imaging sequences, promising similar benefits to those seen in cardiac imaging. While Sonic DL Cine undersamples the phase and temporal dimensions of k-space, Sonic DL 3D focuses on undersampling the phase and slice dimensions. A pre-trained neural network is employed to reconstruct the highly undersampled k-space, exceeding the capabilities of traditional acceleration methods.
Sonic DL is compatible with Cube (3D FSE) and 3D gradient echo-based sequences for neuro, spine, body and musculoskeletal (MSK) imaging. A comprehensive set of protocols will reside in the GE HealthCare library for compatible anatomies and pulse sequences.
A better acceleration solution
Autocalibrating Reconstruction for Cartesian (ARC) imaging is a parallel imaging technique to accelerate data acquisition and shorten scan times by utilizing a 3D kernel to synthesize missing data points. ARC leverages multiple receiver coils to accelerate MR scans, and its performance is influenced by the geometry of the coil array. It works by reducing the number of phase or slice encoding steps needed, which shortens scan times. The g-factor is a measure of how well the coil array can separate the signals from different spatial locations. Higher acceleration factors can lead to increased g-factor values, which in turn amplify noise and reduce the SNR.
At high acceleration factors, ARC can introduce several artifacts:
- Noise amplification: As the g-factor increases, noise is amplified, leading to a lower SNR.
- Signal loss: Variations in coil sensitivity can cause signal loss in certain regions of the image.
- Aliasing artifacts: These occur when the field of view (FOV) is smaller than the imaged object, causing parts of the image to wrap around or fold over in the phase or slice direction.
While ARC can provide adequate results, its performance deteriorates at higher acceleration factors due to increased noise and artifacts.
Sonic DL breaks down the barriers of conventional acceleration techniques, using a deep-learning engine to optimize every voxel by leveraging 6.5 million operations on the raw data, delivering astonishing new levels of acceleration. Because Sonic DL is not based on conventional acceleration techniques, it does not have the same limitations as ARC. Sonic DL is GE HealthCare’s most rapid MR acceleration technique yet.
Now available for volumetric imaging, Sonic DL supports over 75% of MR exams, delivering up to 12-times acceleration1 with up to 86% scan time reduction.2 There are advantages of using 3D acquisitions. A single volume acquisition can be reconstructed in other planes and thinner slice thickness can be utilized, leading to less tissue volume averaging and sharper images, particularly when imaging small lesions or small regions of interest such as cartilage.
Sonic DL can achieve significantly shorter scan times with fewer artifacts compared to traditional parallel imaging methods like ARC. With Sonic DL’s pre-trained neural network employed to reconstruct the highly undersampled k-space, it can intelligently fill in missing information.
Advantages over ARC include:
- Reduced artifacts: Sonic DL is less sensitive to the limitations of coil geometry and g-factor, resulting in fewer artifacts such as noise amplification, signal loss and aliasing.
- Higher acceleration factors: Sonic DL can support much higher acceleration factors, enabling rapid acquisitions with excellent image quality.
- Improved SNR: By leveraging deep-learning algorithms, Sonic DL can maintain a higher SNR even at high acceleration factors.
While ARC can provide adequate results, its performance deteriorates at higher acceleration factors due to increased noise and artifacts. Sonic DL, on the other hand, maintains high image quality even at rapid acquisition speeds, delivering an increase in resolution and a reduction in scan times. Users may be able to achieve greater reductions in scan times with Sonic DL compared to ARC.
How to apply Sonic DL
As mentioned, a comprehensive set of protocols will reside in the GE HealthCare protocol library and will be provided for compatible anatomies and pulse sequences. These will be available for all applicable anatomies on the MR31.0 software.
Should a user want to modify an existing protocol that uses a different parallel imaging technique like ARC, it can be modified via the imaging options page (Figure 1C). Within the imaging options page, the Sonic DL imaging option will be a selectable field. If ARC or ASSET is activated, deselect and Sonic DL will become available (Figure 1A).



Figure 1.
Sonic DL employs a single value selection for acceleration within the Acceleration tab.
GE HealthCare protocols will come with Sonic DL factors that will offer significant scan time reductions when compared to ARC. In each protocol, the resolution has been optimized to maximize quality with the appropriate Sonic DL factor preselected. Adjusting Sonic DL is simplified to selecting one value of acceleration, unlike ARC where the user must select phase acceleration, slice acceleration and/or HyperSense value.
Should you choose to modify the acceleration values to a higher value, there are some considerations that are similar to modifying a protocol with ARC. Sonic DL can be used with high acceleration factors, such as 12, although at times the reduction in the detectability of low contrast structures is not as obvious as what is commonly seen when over-accelerating with ARC.
From the Acceleration tab, the values can be selected, and non-integer and fractional values will be allowed (Figure 1B). To use non-integer accelerations, simply type in the value.
As shown in Figure 2, although there is a gradual increase in noise as the Sonic DL acceleration factor increases from a factor of 5 to a factor of 10 (Figures 2A-2C), the SNR is higher with better image sharpness and detail in the images acquired with Sonic DL compared to the image acquired with ARC and HyperSense (Figure 2D). As the user increases acceleration with Sonic DL, the decrease in SNR is not as abrupt as when using ARC and HyperSense.




Figure 2.
Axial T2 Cube prostate imaging with AIR™ Recon DL, comparing Sonic DL at (A) a speed factor of 5 acquired in 4:06 min., (B) a speed factor of 8 acquired in 3:38 min., and (C) a speed factor of 10 acquired in 2:13 min., versus (D) ARC speed factor of 2 x 1.3 and HyperSense 1.2 acquired in 4:11 min.(A-C) Although there is a gradual increase in noise as the Sonic DL acceleration factor increases from a factor of 5 to a factor of 10, the SNR is higher with better image sharpness and detail in the images acquired with Sonic DL compared to (D) the image acquired with ARC and HyperSense. As the user increases acceleration with Sonic DL, the decrease in SNR is not as abrupt as when using ARC and HyperSense.
Speed and image quality
Similar to other traditional parallel imaging techniques, higher Sonic DL speed factors can produce images with lower SNR when compared to lower acceleration values. However, Sonic DL is compatible with AIR Recon DL to help increase SNR and reduce Gibbs ringing artifact, so users can expect the same benefits when using AIR Recon DL in combination with Sonic DL.
In neuro imaging, quantifying white matter and gray matter volumes is essential for monitoring neurodegenerative diseases. Sonic DL has been built to maintain consistency in these measurements, particularly when using segmentation tools for longitudinal assessments. By following an approach similar to ARC, Sonic DL allows data to remain reliable across time points, enabling confident interpretation. In Figure 3, the image acquired with Sonic DL factor of 5 has higher SNR and depicts more fine details in a 3:06 minute scan time compared to the image acquired with ARC in a 5:19 minute scan time, representing a 41% reduction in scan time with Sonic DL.


Figure 3.
Axial MP-RAGE neuro imaging with AIR Recon DL, comparing (A) ARC 2 x 1 speed factor acquired in 5:19 min. and (B) Sonic DL speed factor of 5 acquired in 3:06 min. The image acquired with Sonic DL has a 31% shorter scan time.
Even at high speed factors of 12, Sonic DL may still provide diagnostic quality examinations in a fraction of the time, as shown in Figure 4.



Figure 4.
Sagittal Cube STIR spine imaging with AIR Recon DL, 1 x 1 x 1 mm, with (A) ARC=2 x1 acquired in 4:29 min., (B) Sonic DL speed factor of 4 in 3:31 min., and (C) Sonic DL speed factor of 12 in 1:27 min. Even with a high speed factor of 12, Sonic DL delivers a diagnostic exam in a fraction of the time compared to ARC and Sonic DL with lower speed factors.
Coils, SNR and resolution
Additionally, and also similar to ARC, increased channel count coils can provide more SNR and are beneficial when accelerating with higher Sonic DL speed factors. In the comparison shown in Figure 5, 5A acquired with the 19-channel Head Neck Unit (HNU) demonstrates sufficient SNR with a Sonic DL speed factor of 8. Figure 5B has an appearance of lower SNR with increasing the Sonic DL speed factor to 10 when using the same coil. Figure 5C acquired with the 48-channel Head Coil has an increase in SNR using the same Sonic DL speed factor as 5B.



Figure 5.
Sagittal T2 Cube FLAIR acquired with AIR Recon DL and Sonic DL at the same resolution, 0.9 x 0.9 x 0.9 mm, 340 slices, demonstrating differences in SNR and image sharpness when using different coils and speed factors. (A) 19-channel HNU with Sonic DL speed factor of 8 acquired in 3:50 min., (B) 19-channel HNU with Sonic DL speed factor of 10 acquired in 3:12 min. and (C) 48-channel Head Coil with Sonic DL speed factor of 10 acquired in 3:12 min.
While reducing resolution is also a strategy to reduce scan times, it could potentially lead to reductions in sharpness or blurred images. In Figure 6, 6A is acquired at 1.2 mm isotropic resolution and 6B is acquired at 0.9 mm isotropic resolution. With lower spatial resolutions, Figure 6A demonstrates a lack of sharpness of the deep brain structures. Although this is an exaggerated example with a 58% change in resolution, it shows the impact on image quality when using high speed factors in combination with low spatial resolution.


Figure 6.
Sagittal T2 Cube FLAIR, coronal reformat, acquired with AIR Recon DL and Sonic DL at different resolutions. (A) Sonic DL speed factor of 10, 1.2 x 1.2 x 1.2 mm, 2:09 min., 270 slices and (B) Sonic DL speed factor of 10, 0.9 x 0.9 x 0.9 mm, 3:12 min., 340 slices.


Figure 7.
The combination of AIR Recon DL and Sonic DL deliver high-resolution MSK imaging in a sagittal Cube proton density FatSat acquisition at 0.5 x 0.5 x 0.5 mm in 3:37 min.
Two solutions should be considered to remedy this concern—reduce the Sonic DL speed factor or increase the spatial resolution. When starting with a protocol from the GE HealthCare protocol library, conservative changes to speed or resolution should be a consideration to produce adequate results.
Summary
Although Sonic DL is less sensitive to the limitations of coil geometry and g-factor, resulting in fewer artifacts such as noise amplification, signal loss and aliasing, it is best to improve spatial resolution in combination with using high Sonic DL factors. Not doing so can lead to suboptimal image quality. Use higher channel count coils whenever possible. As with any 3D volume acquisition, ensure adequate oversampling parameters are considered when using Sonic DL.
References
- Very high acceleration factors could impact the detectability of low contrast structures.
- Compared to fully sampled MR datasets.

