Keywords: AI/ML Image Reconstruction, Diffusion/other diffusion imaging techniques

Motivation: Current deep learning methods for fast dMRI signal estimation are limited in the accuracy and imaging speed. 

Goal(s): Our goal is to enhance the quality of signal estimation  and imaging speed for dMRI, by introducing a new deep learning method.

Approach: Our approach fully utilizes the information in both the diffusion gradient domain and spatial domain to design a joint sparse sampling optimization and reconstruction deep learning framework, along with a specifically designed loss function.

Results: The proposed method achieved up to 15x acceleration while maintaining high estimation accuracy, increasing SSIM by 7% compared with  other q-space learning approaches.

Impact: The dMRI signal estimation performance of our method is promising, as it incorporates domain knowledge into the deep learning process. This approach improves the acquisition and reconstruction workflow of dMRI, benefiting clinical applicability.

Keywords: AI/ML Image Reconstruction, Prostate, AI, DWI

Motivation: While DWI is crucial for prostate MRI, it suffers from low SNR and CNR.

Goal(s): To use AI-based image reconstruction for DWI and compare image quality, TA and diagnostic certainty with the standard DWI.

Approach: Patients underwent standard prostate MRI protocol and received an extra DWI sequence with less averages and AI-based reconstruction. Image quality and PI-RADS were assessed by two blinded radiologists. ROI-based SNR, CNR and ADC values in lesions were calculated.

Results: TA of AI-DWI was reduced by 57 %, while image quality improved. Lesion ADC values and PIRADS assessment remained the same regardless of reconstruction. AI-DWI outperforms the standard DWI.

Impact: AI-based reconstruction of DWI shows promising results for further improving the accessibility and quality of prostate MRI while reducing scan time.  

Keywords: AI/ML Image Reconstruction, Pediatric

Motivation: High angular resolution diffusion imaging has great potential but is time-consuming so is limited in pediatric clinical studies.

Goal(s): To assess the utility of novel deep learning techniques for predicting non-acquired brain diffusion MRI for equivalent HARDI analyses.

Approach: A multilayer perceptron (MLP) and convolutional neural network (CNN) were trained to predict b=2000s/mm² data from b=750s/mm² data. The neurite orientation dispersion and density index (NODDI) outcomes were computed with quality evaluated with PSNR and SSIM.

Results: Both deep learning methods achieved the goal but the CNN outperformed the MLP.

Impact: By applying a competitive neural network method, high angular resolution diffusion imaging can be made possible for the pediatric population in a typical clinical setting based only on half of the data typically required.

Keywords: AI/ML Image Reconstruction, Machine Learning/Artificial Intelligence, Transfer learning, Head and neck

Motivation: Propeller and multi shot EPI sequences offer DWI head and neck images with reduced distortions, but suffer from longer scanning time and reduced SNR.

Goal(s): Shorten the scan time while improving SNR and without impacting ADC values.
 

Approach: Apply transfer learning to retrain denoising deep learning algorithm pretrained on different anatomy and vendor scanner, and using limited amount of data.

Results: Retraining has improved quantitative metrics, resulting in higher quality denoised images using a limited dataset of 5 cases.

Impact: Offsett extra time needed by distortion-reducing protocols to enhance the efficiency of MRI for cancer applications.

Keywords: AI/ML Image Reconstruction, Diffusion/other diffusion imaging techniques

Motivation: Diffusion-weighted imaging (DWI) has been widely reported for detection, staging, and treatment response prediction of rectal cancer. High number of excitations for sufficient SNR was normally used in DWI for rectal cancer but with lengthy acquisition.

Goal(s): Investigate the role of deep learning reconstruction (DLR) in rapid rectum DWI by comparing with standard protocol.

Approach: Forty primary rectal cancer patients were enrolled. Each patient was imaged with standard and rapid DWI. Image quality and diagnostic performance were compared. 

Results: Rapid DWI with DLR reduced 1/2 scan time without sacrificing image quality and improved the diagnostic performance for distinguishing T1/T2 from T1/T4 patients.

Impact: The application of DLR would be beneficial for rectum DWI not only for scan time but also for tumor boundary delineation being important for staging.    

Keywords: AI/ML Image Reconstruction, Neuro, k-space, clinical sensitivity

Motivation: The long acquisition time of diffusion-weighted (DW) imaging hinders its adoption in the clinic for studying pathological microstructural changes in vivo.

Goal(s): The goal of this study was to reduce these long acquisition times by performing downsampling in k-space while maintaining clinical sensitivity. 

Approach: Deep learning was used to transform k-space data to DW images. Experiments were performed eliminating 30% of k-space lines using different methods.

Results: DW images obtained with k-space down-sampling showed a reduction in artefacts, while fractional anisotropy images fitted from the network output appeared to have increased clinical sensitivity.

Impact: This work paves the way for the design of acquisition strategies for fast diffusion-imaging. Through deep learning, it was possible to downsample k-space data in several ways, while obtaining a reduction in artefacts, with a potential increase in clinical sensitivity.

Keywords: AI/ML Image Reconstruction, Pancreas, super-resolution convolutional neural network, high-resolution diffusion weighted imaging

Motivation: Image resolution achieved with compressed sensing was inferior compared to that obtained with sense technique.Current state of the art in Super-Resolution enables enhanced image resolution at a finer level of detail.

Goal(s): Objective is to enhance visualization of anatomical details in high-resolution pancreatic DWI by leveraging SR.

Approach: In our study, we employed integrating super-resolution convolutional neural network-compressed sensing (SR-CS) algorithm and integrating artiffcial intelligence-compressed sensing (AI-CS) algorithm for the reconstruction of pancreatic HR-DWI raw data.

Results: Images with SR-CS generally exhibit superior performance compared to traditional images in terms of tumor border delineation and reduction of background noise from  peritoneum and spine.

Impact: Current utilization of AI is extensive, while application of SR  in medical images remains rare. Utilization of SR  allows for execution of MRI within  a concise timeframe, while simultaneously considering the aspect of resolution.

Keywords: AI/ML Image Reconstruction, Neuro, Super-Resolution Reconstruction

Motivation: A deep learning-constrained algorithm has been integrated into MRI data acquisition and image reconstruction processes, encompassing compressed sensing, image denoising, and resolution upscaling techniques. Nonetheless, limited prospective studies are available that evaluate the application of this algorithm for brain diffusion-weighted imaging.

Goal(s): The primary objective of this study was to compare the recently developed deep learning-constrained algorithm with conventional compressed sensing reconstruction.

Approach: This study comprehensively assessed images, both qualitatively and quantitatively, employing rigorous methodologies and analytical tools.

Results: The results demonstrated that the newly developed deep learning-constrained algorithm significantly enhanced image sharpness while maintaining signal-to-noise ratio, thus advantaging clinical diagnosis.

Impact: Deep learning-constrained super-resolution reconstruction leads to a significant increase in image sharpness of brain DWI, which holds potential to improve clinical diagnosis of diseases, such as stroke and tumors.

Keywords: AI/ML Image Reconstruction, Pelvis

Motivation: Reduced field-of-view DWI (rDWI) can improve the image quality (IQ) and lesion conspicuity in rectal cancer (RC) while reducing signal-to-noise ratio. Deep learning reconstruction (DLR) can denoise images. The feasibility and performance of DLR in rDWI for RC remains unclear.

Goal(s): To compare the IQ and apparent diffusion coefficient (ADC) values between DLR-rDWI and noDLR-rDWI in RC.

Approach: Objective and subjective IQ analysis were performed. ADC values were calculated. Results of IQ analysis and ADC between DLR-rDWI and noDLR-rDWI were compared.

Results: The IQ of DLR-rDWI was better than noDLR-rDWI (except lesion conspicuity, P=0.157). ADC values was not affected by DLR.

Impact: DLR-rDWI could be considered for inclusion in routine rectal MRI protocols. However, whether DLR can improve lesion conspicuity in RC should be further investigated with smaller lesions. The ability of DLR-rDWI in diagnosis and prediction for RC could be studied. 

Keywords: AI/ML Image Reconstruction, Diffusion/other diffusion imaging techniques, Geometric distortion, ADC variability, Deformable Registration

Motivation: Trade-offs between acquisition time and precision of the apparent diffusion coefficient (ADC) hinder the adoption of diffusion-weighted (DW) MRI for biologically-adaptive MR-guided radiotherapy.

Goal(s): To obtain high quality DW images and precise ADC maps using deep learning while shortening acquisition times on an MR-Linac.

Approach: We trained U-net models to generate high quality DW images and ADC maps from only one average per b-value. Four models were trained using either trace-weighted or single DW direction images and with or without registration to the b0 image.

Results: Trained models effectively generated high-quality images from subsampled data. Registration reduced ADC variability.

Impact: Using deep learning we obtained high quality diffusion-weighted MRI from subsampled MR-Linac data. Shortened acquisitions and increased precision of the apparent diffusion coefficient facilitate integration into adaptive MR-guided radiotherapy workflows that use diffusion-weighted MRI for treatment response assessment and prediction.

Keywords: AI/ML Image Reconstruction, Brain

Motivation: Noncontrast computed tomography(NCCT), commonly used for its rapidity in acute  ischemic stroke(AIS) diagnosis, often fails to identify early ischemic changes, whereas MRI, despite its superior sensitivity , may introduce critical delays due to its lengthier image acquisition time.

Goal(s): This study aims to investigate the feasibility of converting NCCT images of AIS patients to diffusion-weighted (DW) images using deep learning techniques.

Approach: The proposed method utilizes an enhanced CycleGAN model to  generate synthetic DW images from CT scans.

Results: The synthetic DW images generated by our network achieved good performance with a PSNR of 28.30, an SSIM of 0.843, and an NMSE of 0.293.

Impact: This work might be translated to clinical settings to help physicians make clinical decisions for patients by providing with high-quality MR images in emergency situations.

Keywords: Analysis/Processing, Neuro, Brain, White Matter

Motivation: The DTI technique is used to analyze and evaluate the white matter microstructure; however, the acquisition time is too long for clinical practice and large-scale research. 

Goal(s): To reduce acquisition time by improving the low angular resolution diffusion parametric maps. 

Approach: The deep-learning framework was used to improve the low angular resolution diffusion parametric maps and image quality measured with PSNR, and NRMSE. 

Results: Our deep-learning framework improves low angular resolution diffusion parametric maps by effectively acquiring fiber information in FA map and enhancing overall image quality with increased PSNR and decreased NRMSE. 

Impact: The reduced acquisition time and improved quality of the low angular resolution diffusion parametric maps obtained with our deep-learning framework may benefit to clinicians and researchers who study in white matter microstructure in routine clinical practice and large-scale research. 

Keywords: Spectroscopy, Contrast Mechanisms, Spectroscopy, Neuro Tumors (Pre-Treatment), Surgery

Motivation: Molecular and pathological diagnosis requires fresh tissues. It is important to achieve precise sampling and preservation of the tumor is for maintaining the tumor integrity.
MRSI can accurately identify tumor hotspots and can be used for obtaining high quality samples.

Goal(s): Develop MRSI pipeline for precise intraoperative tumor sampling.

Approach: We processed 7T MRSI metabolic ratio maps and identified the tumor hotspots. We then correlated them with their quantitatively analyzed metabolic profiles.

Results: We were able to achieve high quality tumor samples based on our 7T MRSI maps. 

Impact: Due to high resolution 7T metabolic ratio maps we could identify and define tumor hotspots which resulted in precise tumor sampling. We were also able to preserve the tissue specimens and obtain high quality results.

Keywords: Spectroscopy, Brain, 7T, MRSI, Concentration Estimates

Motivation: In magnetic resonance spectroscopic imaging, generating concentration estimates is desirable. So far, our approach has been limited by the need for WM/GM separation and literature values, limiting it to healthy subjects. To solve this, it can be replaced by PD map-based references.

Goal(s): To test an approach of calculating concentration estimates within volunteers without relying on literature assumptions for reference.

Approach: Use quantitative proton density maps to calculate water concentration maps for metabolic imaging at 7T.

Results: We successfully used ME-GRE imaging to calculate water concentration maps. In healthy volunteers, these maps correspond well to the previous method.

Impact: Using tissue water maps for MRSI concentration estimation allows to not only apply the method to healthy brains, but also to pathologies like gliomas. This will make 7T MRSI a better tool for studies.

Keywords: Analysis/Processing, Spectroscopy, Wavelet Analysis, Signal Decomposition, Proton MRS

Motivation: Magnetic resonance spectroscopy (MRS) is currently limited by noise, low spatial resolution, and artifacts that compromise the accuracy of metabolite quantification.

Goal(s): This work aims to enhance MRS signal quality without compromising signal integrity, employing wavelet analysis for robust signal decomposition. 

Approach: A novel method utilizing wavelet analysis and a U-Net architecture creates masks that segment scalograms, effectively isolating individual metabolites in MRS signals. 

Results: The method has shown in simulations the ability to distinctly separate and characterize metabolite signals, offering a promising direction for refining MRS data analysis.

Impact: Provides a data-driven method for MRS signal decomposition based on wavelet analysis that shows success in extracting metabolite and baseline information. It holds the potential for accurate characterization of nuisance signals, which could lead to improved MRS fitting.

Keywords: Analysis/Processing, Machine Learning/Artificial Intelligence, Motion, MRSI, Brain, Quality assurance

Motivation: Quality assessment of whole-brain MRSI spectra is usually based on post-quantification analyses, which does not reflect if the estimated metabolite concentration is true.

Goal(s): Simulate effects of subject motion for the raw non-Cartesian MRSI kSpace data. Build a dataset of motion-corrupted MRSI data with a corresponding ground truth version. Train a classifier to assess the quality of MRSI data.

Approach: Translations and rotations were simulated in the kSpace domain. A classifier is trained in a supervised fashion with the thresholded deviation between the motion-affected and original data as the target.

Results: The classifier outperforms the CRLBs in the quality assessment of MRSI data.

Impact: Simulation of subject motion effects on raw non-Cartesian kSpace MRSI data allows us to assess the quality of MRSI spectra and can lead us toward the understanding of lipid artifacts, which is the main limiting factor of MRSI.