Synopsis

Keywords: Musculoskeletal: Cartilage

In the last decade musculoskeletal MRI has unprecedented advances. With 3 Tesla field strength high performance gradients and rapid radiofrequency pulse transmission, advanced multichannel receiver coil and surface coil technologies, quantitative MRI techniques and radiomics, MR fingerprinting, synthetic contrast generation, parallel imaging, compressed sensing, simultaneous multislice acquisition techniques have been developed. Deep learning and artificial intelligence algorithms can enhance MRI with unparalleled gains in image speed and quality without signal-to-noise loss. These advances brought challenges too. In this educational lecture we will look around these issues centering around the fundamentals including definitions, advantages, and clinical applications.

Computational MRI

1. Quantitative MRI
Quantitative MRI techniques are variable methods to find the information about microstructural tissue characteristics, those are not accessible by conventional qualitative MRI (1). The microscopic changes of tissues cause the changes of the T1rho, T2, T2* relaxation times and so on and it can be quantitatively measured and visualized on the parameter maps with these techniques. With evaluation of these disease-driven changes, we can evaluation the cartilage and muscle composition, and joint inflammation objectively (1). There are some challenges for adopting and using quantitative MRI in musculoskeletal MRI, including long scan time. However, thanks to the recent advances of MR science quantitative MRI is promising.

2. Radiomics
Radiomics is the automated process to extract features from digital medical images to support decision-making (2). When we extract large amounts of features from MR images and combine them with clinical, biological, genetic data, we can build diagnostic, prognostic, or predictive model with them. Radiomics makes segmentation very fast and precisely so it can reduce our huge work of quantitative analysis. Owing to no need of additional scans for radiomics, it can be easily integrated into the workflow. Radiomics can uncover hidden patterns of specific information of MR images those human eyes cannot perceive (2, 3). MRI-based radiomics model with a machine learning is also promising (3).

3. MR Fingerprinting
MR Fingerprinting is an image generation framework for acquisition of multiple quantitative property maps with a single scan (4). With MRF technique we can quantify multiple tissue properties like proton density, T1, T2, T2* relaxation times, and field inhomogeneity at once (5). Using a various flip angles and repetition times we can get the transient signal of parameters and generate a unique ‘fingerprint’ with it, and pattern matching with a predefined dictionary of simulated signal evolutions is efficient and accurate to determine various tissue properties (4. 5). Recently there have been many technical developments for MRF, including sequence optimization, improved reconstruction algorithms, partial volume separation, and deep learning (6).

4. Synthetic contrast generation
Synthetic contrast generation is emerging promising technique in musculoskeletal MRI; making additional conventional contrast images from one or more sequences without requiring additional acquisition time, enhancing conventional contrast images to be easier to detect pathology, and putting quantitative imaging into clinical routine without requiring additional scan time are possible (7). There are three methodologies: mathematical image transformation, MR physics-based approach, and data-driven synthetic contrast generation (7). Quantitative MRI can be accelerated with synthetic contrast generation, and independent of the used MR scanner, coil, and acquisition techniques (7).

Conventional Accelerated MR Reconstruction

5. Optimization of parameters in 3T MRI
3T MR system has many advantages including high-performance gradients systems and radiofrequency pulse transmission technology, advanced multichannel receiver technology, and high-end surface coils (8). Compared with 1.5 T, 3.0-T MRI systems can achieve approximately 2-fold higher signal-to-noise ratios, enabling 4 times faster data acquisition or double the matrix size (8). Thanks to this powerful 3T MRI, we can get better quantitative MRI. The first and most widely accessible techniques of acceleration of MRI in musculoskeletal imaging is to optimize the pulse sequence parameters (9).

6. Parallel Imaging technique
Parallel imaging is one of the most accessible advanced acceleration techniques for musculoskeletal MRI examinations (10). With sampling only every second phase-encoding step we can save 50% of acquisition time (10). With use of parallel imaging technique and reduced repetition times and spatial resolution, twofold acceleration of 2D FSE and TSE pulse sequences can make abbreviated 5-minute knee and shoulder MRIs possible (11). It also makes 3D FSE and TSE pulse sequences clinically feasible with acquisition times of 5 minutes and less (11).

7. Compressed Sensing technique
Data redundancy of MR images can be used to reduce the sampling rate of phase-encoding direction. When it be compactly represented in some transform domains, it is close to the concept of “sparsity” (12). Compressed sensing technique allows accurate reconstruction from sparsely sampled k-space data (12). Compressed sensing technique uses nonlinear undersampling patterns and recovers missing data during image reconstruction (11). Although parallel imaging acceleration is usually limited to acceleration factor of 3, compressed sensing enables acceleration factor of 8 (11). Compressed sensing has become the essential tools in musculoskeletal MRI (11).

8. Simultaneous Multislice acquisition
Acceleration in data acquisition using Simultaneous Multislice acquisition is the same to the number of simultaneously excited slices (13). This technique has only a marginal intrinsic signal loss and at fixed echo time full acceleration is achievable (13). Combination of simultaneous multislice acceleration and parallel imaging is desirable because it has minimal signal-to-noise ratio penalty but increases the specific absorption rate, whereas parallel imaging results in signal loss but specific absorption rate does not increase (11). And this combination permits the design of a 5-minute knee MRI protocol (11).

Deep-Learning Application in Musculoskeletal MRI

9. Deep Learning-based Fast Superresolution MRI
In the last decade the advances in 3T MRI hardware and potent pulse sequences techniques achieved amazingly reduced acquisition time of conventional MRI in musculoskeletal imaging (14). But at the same time, we encountered the signal-to-noise limits of temporal, spatial, and contrast resolution because of the high-level acceleration (15). Emerging novel deep learning image reconstruction methods can enhance MR images with unparalleled gains in image speed and quality without signal-to-noise loss (14). The uncountable potential of deep learning–based image reconstruction promises to fast super-resolution musculoskeletal MRI such as a total acquisition time of 2–3 minutes for entire MRI examinations of joints without sacrificing spatial resolution or image quality (15).

10. Challenges and Future Directions
Deep learning-base MRI technologies are not yet widely used in clinical practice. For this, widespread implementation and clinical validation is required (14). According to increase of the MRI workflows with big data, artificial intelligence, and sophisticated image analysis applications, the use of data set availability and network stability is becoming increasingly important for both clinical and research purposes (15). Newly developing vendor-neutral sequences of quantitative MRI will enhance the reliability of multicenter clinical trials (16).

Acknowledgements

No acknowledgement found.