MR Advanced Acceleration Technology

Welcome to the MR Advanced Acceleration Technology (AAT) resource, provided by the IPEM MR AAT Task and Finish Group. This resource provides information and user experience on three MR pulse sequence innovations that are collectively referred to as AAT: compressed sensing (CS), simultaneous multi-slice (SMS), and artificial intelligence (AI). Information and user experience for GE Healthcare, Philips and Siemens is currently available.

The aims of this resource include:

Providing up-to-date information about available MR AAT by manufacturer, across a range of anatomies and pulse sequences. This will be helpful for departments making MR software specification decisions on new or upgraded scanners
Providing detailed, concise manufacturer specific information regarding the implementation of AAT on MR scanners, covering where the technology works well, the challenges encountered and any notes, including user tips for deploying each AAT
Providing user experiences on the implementation of each technology in relation to their impact on MR image quality and/or patient throughput

The resource is a work in progress and reflects the current level of collated user knowledge and experience in this new and exciting field. The resource will be reviewed and updated regularly as more experience is gathered across the MR community. The IPEM MR T&F group welcome contributions from clinical scientist, radiographer, academic and radiologist colleagues who are involved in the deployment of MR AAT.

Please send any experience you believe is suitable for the resource to: Michael.hutton4@nhs.net and steven.jackson6@nhs.net

To use this resource, begin by choosing your MR scanner manufacturer from the first dropdown box. Then choose the specific AAT technique you wish to find out more information about. There are three categories to choose from: simultaneous multi-slice (SMS), compressed sensing (CS) and artificial intelligence (AI). A full glossary of acronyms used in the resource is available below the tables.

The information below is believed to be correct to the best of the T&F group’s knowledge at time of most recent update (March 2024). Users should assess available AAT locally and undertake their own optimisation.

Disclaimer: the information presented below reflects clinical user experiences shared through the IPEM AAT Task and Finish Group. Members of the IPEM MR AAT Task and Finish Group receive no funding from scanner manufacturers and have no conflicts of interest to declare.

;

Model & Software info	Name of technique on scanner?	Applicable to which sequence(s)	Where it works well	Challenges encountered	Notes
.E11 software (Aera, Skyra, including Aera fit/Skyra fit) .XA software (all 1.5T and 3T models excluding Sempra)	SMS	EPI (e.g. ep2d used for DWI)	• Works best with anatomy specific receive coils (e.g. head).	• Technique struggles with low slice number, thin slices and unfavourable coil geometry (e.g. when coil elements are not well separated in slice direction). • Not possible to have SMS and respiratory triggering	• Not expected to impact patient throughput due to use in specialised protocols only • Identified in the Siemens tree by 's2' in sequence name for an SMS factor 2 • Number of slices must be divisible by SMS factor, which may impact accompanying T2 image • SMS factor 2 most reliable
.XA software (all 1.5T and 3T models excluding Sempra)	SMS	2D TSE (including TSE Dixon)	• Works best with anatomy specific receive coils (e.g. head, transmit/receive knee, foot/ankle, shoulder). • Good for T1 and T2 whole body transverse imaging. • If a sequence has more than one concatenation before implementing SMS, this can be reduced 'for free' with SMS employed.	• Technique struggles with low number of slices, thin slices (e.g. prostate pi-rads), small FOV (e.g. gynae pelvis) and unfavourable coil geometry, for example when coil elements are not well separated in slice direction (e.g. sag spine). • It appears that oblique slices pose greater issues for the reconstruction algorithm than true axial, sagittal, coronal. • Can be higher SAR than non-SMS equivalent sequence due to novel RF pulses • Reconstruction time longer than TSE sequences (although high end computing and latest XA51 reduces this) • Not possible to have SMS and respiratory triggering	• Throughput increases possible if deployed across multiple 2D TSE sequences to reduce acquisition times in common protocols with favourable anatomy + coil combinations, e.g. mulsculoskeletal imaging (MSK) • Identified in the Siemens tree by 's2' in sequence name for an SMS factor 2 • If there is no scope to shorten TR or reduce concats then SMS may not lead to time savings • Cannot be used with DRG but can be used with DRB • Number of slices must be divisible by SMS factor • SMS factor 2 most reliable • On paper, no SNR penalty for using s2
.XA software (all 1.5T and 3T models excluding Sempra)	SMS	RESOLVE (DWI technique, not to be confused with Deep Resolve, the AI technique)	• Works best with anatomy specific receive coils (e.g. head).	• Technique struggles with low slice number, thin slices and unfavourable coil geometry (e.g. when coil elements are not well separated in slice direction).	• Not expected to impact patient throughput due to use in specialised protocols only • Identified in the Siemens tree by 's2' in sequence name for an SMS factor 2 • Number of slices must be divisible by SMS factor, which may impact accompanying T2 image • SMS factor 2 most reliable.
.XA software (all 1.5T and 3T models excluding Sempra)	SMS	BOLD (Blood oxygen level dependent)	• Main application is to increase sampling rate of the fMRI time series by reducing TR. This is desirable particularly for resting state fMRI. • High SMS factors (up to 8) can be used for whole brain coverage when using high density receive coils (e.g. 32 channel head coil) if not using GRAPPA. SMS factors need to be constrained (to 2) if using GRAPPA=2. • SMS can be used to increase nuber of slices and/or reduce slice thinkness to improve coverage without time penalty.	• At higher SMS factors susceptibility induced artefacts are enhanced due to the longer EPI readout when GRAPPA is not in use. • The higher the slice acceleration factor (smaller distance between simultaneously excited slices), the higher the potential for slice leakage artefacts.	• Minimal impact on patient throughput expected at non-neuro specialist centres

Model & Software info	Name of technique on scanner?	Applicable to which sequence(s)	Where it works well	Challenges encountered	Notes
.XA software (all 1.5T and 3T models excluding Sempra)	Compressed Sensing	SPACE (3D TSE)	• 3D volume scanning with moderate acceleration factors (2-5) • MRCP SPACE breathold (24x acceleration) found to be acceptable for some difficult patients, quick triggered sequence (up to 20x acceleration) found to be reliable.	• High denoising factors can introduce artefactual features into images • Isometric voxels can’t go much faster than conventional parallel imaging (e.g. CAIPIRINHA x4) without encountering image quality issues • Long reconstruction times (up to 1 minute for large volumes) due to iterative reconstruction on XA31	• Not expected to impact patient throughput due to use in specialised protocols only • Identified by 'cs' in Siemens tree. • Where the option is available, a high denoising factor (1-1000 is available) can lead to synthetic looking image. Use lowest de-noising factor that gives acceptable images (e.g. 5-20).
.XA software (all 1.5T and 3T models excluding Sempra)	Compressed Sensing	SEMAC	• SEMAC used for metal artefact reduction in 2D sequences, pre-sets available for hip and knee • Big reduction in imaging time possible with CS SEMAC compared to non-CS SEMAC (e.g. T2 knee with metalwork), SAR much easier to control than non-CS SEMAC	• T1 implementation of the sequece has an echo train length of 9, which provides poor T1-weighting outside of pre-set anatomies. • Long reconstruction times on XA31 (quicker on XA51)	• Not expected to impact patient throughput due to use in specialised protocols only • Identified by 'cs' in Siemens tree. • High SEMAC factors possible in short TA
.XA software (all 1.5T and 3T models excluding Sempra)	Compressed Sensing	TOF (time of flight)	• Works well for circle of Willis scanning with CS acceleration factors of 5 and 7.5 available in Siemens tree.	• May miss small vessel details with cs factors higher than 6	• Not expected to impact patient throughput due to use in specialised protocols only
.XA software (all 1.5T and 3T models excluding Sempra)	Compressed Sensing	Cardiac	• Capable of acquiring a full short axis stack in one breath hold	• CS only available for cine sequences, Siemens have stated that in the future it will be available for first-pass perfusion sequence. • For patients with arrythmia any changes of RR may result in slices not being interpolated resulting in varying segments • Issues with analysis of ejection fraction, in some patients not all phases are generated or are truncated	• Not expected to impact patient throughput due to use in specialised protocols only
.XA software (all 1.5T and 3T models excluding Sempra)	Compressed Sensing	GRASP-VIBE	• Motion- insensitive, free-breath dynamic volume for abdominal imaging in patients unable to breath-hold • Motion artefact free imaging in prostate and neck	• Potential for radial streaking artefact if radial coverage is too low. This is mitigated by adding spokes to the acquisition at the cost of reducing temporal resolution.	• Not expected to impact patient throughput due to use in specialised protocols only

Model & Software info	Name of technique on scanner?	Applicable to which sequence(s)	Where it works well	Challenges encountered	Notes
.XA software (Sola, Vida, Altea, Viato but not Aera, Skyra. Consult manufacturer for other models up to date information)	Deep Resolve Gain (and Sharp)	2D TSE and TSE	• Use to reduce acquisition time through reducing averages, adding parallel imaging, reducing base resolution • Improved image sharpness through use of DR Sharp • Use higher GRAPPA factors (up to 3) to reduce acquisition time in 2D imaging • Works across a wide range of anatomies, regardless of receive coil	• Not compatible with multi-phase acquisitions, multi-echo acquisitions, inline image filters or true inversion recovery reconstructions • Modified image noise appearance compared to non-DRG images • Movement and flow artefacts enhanced in comparison to TSE imaging	• Throughput increases possible if deployed across multiple 2D TSE sequences to reduce acquisition times in common protocols • Identified in Siemens tree by heading 'DRG+DRS' (not included in sequence name) • Uses targeted de-noising, not AI, but included in Deep Resolve package • By default DRS applied with DRG, but DRS can be turned off • Applicable to (non-turbo) SE and Dixon sequences on XA51 • Can reduce acquired phase matrix by 20-30% and use phase percentage 75-80% when applying DRS with no impact on image sharpness • Can reduce TA by up to 50% with minimal impact on image quality in favourable anatomies • DRG settings (denoising factor and edge enhancement) have modest effect on non-FS images • Not available for BLADE sequences
.XA software (Sola, Vida, Altea, Viato but not Aera, Skyra. Consult manufacturer for other models up to date information)	Deep Resolve Boost (and Sharp)	2D TSE	• Use to reduce acquisition time through reducing averages, adding parallel imaging, reducing acquired phase matrix • Use higher GRAPPA factors (up to 4) with increased phase oversampling to reduce acquisition time in 2D imaging • Works across a wide range of anatomies, best with anatomy specific coils e.g. knee, head, foot/ankle	• Not compatible with multi-phase acquisitions, multi-echo acquisitions, inline image filters or true inversion recovery reconstructions • Not compatible with Dixon • Movement and flow artefacts enhanced in comparison to TSE imaging • Prone to unconventional wrapping artefacts when using spine coil and anterior body coils e.g. spine, pelvic imaging • Ensure correct calibration scan is enabled to minimise artefact ('TSE/separate' for brain/prostate and 'Integrated' for all other anatomies)	• Throughput increases possible if deployed across multiple 2D TSE sequences to reduce acquisition times in common protocols • Identified in Siemens tree by heading 'DRB+DRS' (not included in sequence name) • By default DRS also applied with DRB, but DRS can be turned off • Reduce TA by >50% with minimal impact on image quality in favourable anatomies • Effective when combined with View Angle Tilting (VAT) and high bandwidth for metal artefact reduction • Not available for Dixon, (non-turbo) SE or BLADE sequences • To reduce reshims with change in isocentre across protocol set shim to 'Local Range' using 'Table Position Strategy'

Model & Software info	Name of technique on scanner?	Applicable to which sequence(s)	Where it works well	Challenges encountered	Notes
All 3T models, 5.7 upwards	MultiBand SENSE	EPI readouts (fMRI/DTI/DWI)	• Works better with multiple reciever coils with a large number of elements in the slice direction. The 32 channel head coil is optimal for brain imaging. • BOLD: Whole head fMRI (faster sampling, ~1s ) for resting state fMRI. Can also be used to increase number of slices/ reduce slice thickness for improved slice coverage without time penalty. • Faster DTI acquisitions, but need to keep TR above 2500ms to avoid compromising SNR too much. • Areterial Spin Labelling: increased slice coverage for Cellular Blood Flow and arterial Cellular Blood Volume measurements.	• Technique struggles with low slice number, thin slices and unfavourable coil geometry (e.g. when coil elements are not well separated in slice direction). • Incomplete unwrapping artefacts are possible in DTI imaging. The number of slices divided by the multiband factor needs to be an odd number to avoid this artefact. • SENSE factors may need to be reduced when using slice accelearation, which will increase susceptibility-induded effects due to longer EPI readouts. • For BOLD, SENSE = 2 can only be combined with a maximim MultiBand factor of 2. Philips recommends a SENSE factor of 2 for DTI to with multiband SENSE=2 in order to compensate for loss of SNR due to shorter TR.	• Minimal impact on throughput expected at non-neuro specialist centres • MultiBand imaging is SNR efficient compared with conventional parallel imaging as there is no penalty related to the slice acceleration factor. The only SNR penalty is g-factor. • Reduction of TR which is sometimes made possible by multiband can also reduce the SNR. Flip Angle needs to be optimised to maximize available SNR for BOLD imaging. • Multiband SENSE can be combined with PI more flexibly than Siemens/GE as SENSE acceleration factor not required to be integer (e.g. combination of SENSE=1.5, MultiBand SENSE=4 works well for 2mm isotropic whole brain coverage resting state fMRI at 3T with 32-ch head coil) • Multiband SENSE can be combined with multi-shot EPI

Model & Software info	Name of technique on scanner?	Applicable to which sequence(s)	Where it works well	Challenges encountered	Notes
All models, 5.4 upwards	Compressed SENSE	Cartesian 2D and 3D scans of all contrasts and all anatomies.	• It works best for larger dimensions (3D, dynamic, flow) • Faster breathholds possible for body imaging e.g. liver, cardiac • Moderate time savings for 2D TSE sequences with C-SENSE factors slightly higher than SENSE	• Changing from SENSE to CS-SENSE can change the voxel size, which can impact SNR and diminish resolution. Be vigilant and adjust CS-SENSE factor to achive required resolution • Long reconstruction times (up to 1 minute for large number of slices), removes benefit of time saving if planning a sequence requires the previous one to be reconstructed, and when the final sequence needs to be checked before removing patient from the bore. • When present, motion artefacts are more severe than non-CS imaging, with significant numbers of bright ghosts	• Throughput increases possible if deployed across multiple 2D and 3D sequences to reduce acquisition times in common protocols • To enable on the scanner select 'CS' from acceleration dropdown • Needs an integer number of averages, which may be an issue when optimising an existing sequence • 3D T1 FFE needs to be multishot to allow CS, it won't run with single shot at present
From Release 11	Smartspeed MotionFree	Multi-VANE XD versions of multiple TSE and FFE contrast weighting, including IR, Dixon.	• Patients who are unable to keep still • Can be used to build alternative protocols for motion management	unable to comment - check back soon	• Not expected to impact patient throughput due to use in specialised protocols only • Allows MultiVANE XD radial sequences to be accelerated with CS • Equivalent BLADE/PROPELLER on Siemens/GE not currently compatible with AAT
From Release 11	SmartSpeed 3DFreeBreathing	Allows 3D T1 weighted FFE scans to be obtained without breath-holding, including fat sat techniques and mDixon.	• Where 3D T1 weighted FFE sequences are needed with a breathhold, for example post contrast liver imaging	unable to comment - check back soon	• Not expected to impact patient throughput due to use in specialised protocols only • Based on stack of stars technique with CS acceleration.
From Release 11	SmartSpeed Implant	Compatible with T1, T2, PD weighted TSE sequences and STIR	• Designed for MSK imaging around metallic implants	unable to comment - check back soon	• Not expected to impact patient throughput due to use in specialised protocols only • Allows O-MAR XD technique to be accelerated with CS
From Release 11	SmartSpeed Diffusion	Single shot EPI based DWI sequences (maximum of 3 directions)	• All DWI imaging, especially where coil geometry is unfavourable to image uniformity, e.g. pelvis • Increaed SNR allows for reduction in the number of averages at higher b-values	unable to comment - check back soon	• Not expected to impact patient throughput due to use in specialised protocols only • Allows EPI based diffusion to be accelerated with CS

Model & Software info	Name of technique on scanner?	Applicable to which sequence(s)	Where it works well	Challenges encountered	Notes
Software Version 29	Air Recon DL	Applies to 2D sequences: - EPI: DW EPI - Fast Spin Echo: FSE, SSFSE - Gradient Echo: Fast GRE, Fast SPGR, GRE, SPGR - Spin Echo: Spin Echo	• Most effective in sequences with low SNR that require multiple signal averages. • Contrast to noise improvement greatest for objects with low contrast detectability. • Available in Low, Medium and High Recon DL strengths. • Applies Gibbs ringing suppression independent of Recon DL strengths. • Sharpness improvement across all sequences. • For respiratory gated sequences it can reduce breath-hold duration.	• Greatest effect seen when increasing resolution vs time reduction, except in cases with multiple NEX. • Optimization for each clinical sequence can require changing multiple factors (NEX, Matrix, and Slice thickness, receiver Bandwidth for FSE sequences, and parallel imaging factors ARC and ASSET). • Not able to change Recon DL strength retrospectively. • Amplification of flow and respiratory artefacts with ARDL on high due to denoising; artifacts at or just above the noise floor are rendered more conspicuous. Most noticeable on low SNR sequences such as DWI.	• Option key name: AIRReconDL2D. • Throughput increases possible if deployed across multiple 2D sequences to reduce acquisition times in common protocols • Is compatible with HyperBand for 2D DW EPI sequences. • For FAST SPGR 'real time' option is not supported with ARDL selected. • ARDL does not rely on conventional filters such as fermi filtering to balance Gibbs ringing and image sharpness. Instead, the convolutional neural network operates on raw, complex-valued imaging data.
Software Version 30.0	Air Recon DL	Applies to 2D radial sequences: - Fast Spin Echo: DW PROPELLER, PROPELLER	• Extension of Air Recon DL to PROPELLER sequences. PROPELLER is enabled for T1 FLAIR, T2, T2 FLAIR imaging in all planes, axial diffusion weighted imaging for brain, T2-weighted imaging for cervical spine, Body, and T2/PD-weighted imaging for MSK. • Works well to reduce motion artefact where breathing or patient movement is usually an issue, for example abdominal MR, difficult patients, shoulder		• Option key name: AIRReconDLPRP - Protocol Information: Enables DL Propeller when combined with airrcndl2d • Propeller is a motion insensitive radial acquisition that reduces the effects of motion artifact in routine scanning. DW Propeller reduces the effects of susceptibility artifacts at bone/tissue or air/tissue interfaces, or around tissue/metal interfaces. • Air Recon DL can be applied to denoise the image and increase image sharpness and resolution. • Recommended No Phase Wrap factors: Brain and Spine 1.5, MSK 1.5 (for Transmit Receive coils) or 2 (for Receive only coils) and Pelvis 2 to 2.5. • To gain SNR with Air Recon DL increase NEX rather than increase No Phase Wrap factors to avoid potential for streak artifacts. • To minimize chemical shift artifact, when using Air Recon DL, recommended receiver bandwidths are 3T: up to 125 kHz, 1.5T: up to 83 kHz.
Software Version 30.0	Air Recon DL	Applies to 3D sequences: - Fast Spin Echo: CUBE, CUBE DIR, CUBE T2 FLAIR, FRFSE-XL, FSE - Gradient Echo: Fast GRE, Fast SPGR, Fiesta, LAVA, VIBRANT, Inchance IFIR - APPLICATIONS: BRAVO, MP RAGE, T2 MAP, T1 MAP SPGR	• Extension of Air Recon DL to 3D sequences • Works well for shortening breath holds and denoising body imaging, i.e. LAVA.	• 3D Air Recon DL has longer reconstruction times for high resolution MRI	• Option key name: AIRReconDL3D. • Protocol Information: AIR Recon DL for 3D imaging • Is compatible with HyperSense on CUBE, LAVA, VIBRANT, BRAVO and MP RAGE sequences.
Software Version 30.1	Sonic DL	Cardiac MRI (2D Gradient Echo - FIESTA)	• Decrease scan time & exam duration • Improve temporal sharpness • Acquire rapid, free breathing or breath hold acquisitions		• Option key name: dlspdcine • Protocol Information: Sonic DL - Cine • Sonic DL is compatible with ZIP512, Square Pixel and Respiratory Gating/Triggering imaging options • Respiratory Triggering is compatible with 1 R-R, allowing for free breathing acquisitions • SCMR recommended acquired temporal resolution for Sonic DL sequences is 45 ms • SCENIC is not compatible with Sonic DL • Sonic DL is not compatible with ASSET

GLOSSARY

AI – artificial intelligence (vendor netural)

ARDL – Air^TM Recon DL, GE AI technique (GE)

ARC – Auto-calibrating Reconstruction for Cartesian Imaging, parallel imaging method (GE)

ASSET – Array Coil Spatial SensiTivity encoding ,parallel imaging method (GE)

BLADE – Radial acquisition technique to reduce sensitivity to movement (Siemens)

CAIPIRINHA – Controlled Aliasing in Parallel Imaging Results In Higher Acceleration, 3D Parallel imaging method (Siemens)

CS – compressed sensing, method of accelerating signal acquisition via incoherent undersampling and iterative de-noising. Vendor specific: Compressed sensing (Siemens), compressed SENSE (Philips), HyperBand (GE)

CS-AI – compressed SENSE and artificial intelligence, Philips implementation of AI into the CS reconstruction pathway (Philips)

DR – Deep Resolve, Siemens AI umbrella term (Siemens)

DRG – Deep Resolve Gain, Siemens intelligent de-noising technique (Siemens)

DRS – Deep Resolve Sharp, Siemens AI based image resolution enhancement (Siemens)

DRB – Deep Resolve Boost, Siemens Raw data to image deep learning technique, released after DRG, AI based de-noising, unlike DRG (Siemens)

EPI – echo planar imaging, T2* weighted sequenced used for rapid acquisitions in techniques such as DWI and fMRI (vendor neutral)

GRAPPA – GeneRalised Autocalibration Partial Parallel Acquisition, Parallel Imaging (Siemens)

HyperBand – see CS (GE)

HyperSense – see SMS (GE)

MultiBand SENSE – see SMS (Philips)

MultiVANE – Radial acquisition technique to reduce sensitivity to movement (Philips)

NSA/NEX – number of signal averages / number of excitations, number of times the same signal is sampled during an MR acquisition (vendor neutral)

O-MAR – Metal Artefact Reduction for Orthopedic implants (Philips)

PROPELLER – Radial acquisition technique to reduce sensitivity to movement (GE)

RESOLVE – Readout Segmentation Of Long Variable Echo trains, multi-shot technique for DWI (Siemens)

SEMAC – Slice Encoding Metal Artefact Correction, method to reduce the extent of image artefacts due to metal (vendor neutral)

SENSE – SENSitivity Encoding, parallel imaging (Philips)

SMS – simultaneous multi-slice, method of accelerating signal acquisition via acquiring image data from multiple slices at once. Vendor specific: Simultaneous multi-slice (Siemens), MultiBand SENSE (Philips), HyperSense (GE)

Model & Software info	Name of technique on scanner?	Applicable to which sequence(s)	Where it works well	Challenges encountered	Notes
Software Version 27	HyperSense	Limited Range of 3D sequences	• Works well for sequences with lower information content such as TOFs and MRCP exams with factors between 1.2 - 1.5.	• Compressed sense factors between 1.1-1.2 are acceptable on standard MRIs but higher levels of compressed sensing introduces blurring.	• Protocol Information: Provides scan time reduction technique while maintaining signal-to-noise ratio (SNR) through an innovative data compression algorithm for 3D • Not expected to have significant impact on patient throughput for HyperSense options available using 'HyperSense', additional sequences are available from software version 29 with option key 'HyperSense2.0'. • Not available for 2D sequences.
Software Version 29	HyperSense 2.0	Applies to 3D sequences: - Applications: BRAVO, DISCO, MPRAGE, COSMIC - Fast Spin Echo: CUBE, CUBE DIR, CUBE T2 FLAIR (Part of Hypercube package), FRFSE-XL, FSE - Gradient Echo: FIESTA-C, LAVA, MERGE, VIBRANT - Vascular: TOF-GRE, TOF-SPGR	• Extends HyperSense to 3D T1 Sequences		• Protocol Information: HyperSense 2.0 extends HyperSense to 3D T1 sequences

MR Advanced Acceleration Technology Experience Resource