Synopsis

Keywords: Image acquisition: Fast imaging

The Fast Imaging: Parallel Imaging, Compressed Sensing & More lecture will provide a high-level introduction to fast imaging acquisition techniques, parallel imaging, and reconstruction methods including compressed sensing and the new field of deep learning. Each topic will be described using diagrams and images to illustrate the techniques as well as their strengths and limitations.

Abstract:

Fast MRI techniques are now available in all commercial scanners. Primarily designed to reduce the deleterious effects of physiological or patient motion, fast imaging methods are also useful in reducing the duration of the MRI examination and thus, improve patient comfort and throughput. Fast imaging methods typically rely on acquiring less data during the scan. The challenge of fast imaging is to maintain an adequate diagnostic quality which can be achieved by a combination of pulse sequence design, hardware technology, and novel reconstruction methods.
The lecture will provide a high-level introduction to fast imaging acquisition techniques, parallel imaging, and reconstruction methods including compressed sensing and the new field of deep learning. Each topic will be described using diagrams and images to illustrate the techniques as well as their strengths and limitations.
Target Audience: Clinicians and basic scientists who are new to the field of MRI as well as experts in the field who are interested in refreshing the concepts of fast MRI.
Fast imaging methods: An important component to achieve fast imaging is to have a pulse sequence designed to acquire data fast. This section will provide an overview of the imaging pulse sequences that are typically used to accelerate image acquisition. Clinical examples will be used to illustrate advantages and disadvantages of fast imaging methods.
Parallel imaging methods: Another step towards acceleration is to acquire less data than what is required by basic sampling principles to reconstruct an image. This process, called undersampling, will yield image artifacts that can impair diagnosis. These can be circumvented using the information acquired in “parallel” by the various pick-up coils used during data acquisition. This section will cover the basic concepts of “parallel imaging” including acquisition and reconstruction. Imaging examples will be used to illustrate the benefits of the technique as well as limitations in terms of residual artifacts and speed.
Compressed sensing methods: Higher acceleration can be achieved by increasing the degree of undersampling beyond what parallel imaging can tolerate. Reconstruction of images from “highly undersampled” data is possible based on the principles of “compressed sensing”. This section will introduce compressed sensing, show its improvement in acceleration, and discuss image characteristics and typical artifacts.
Deep learning methods for accelerating acquisition and reconstruction: With advances in computational power, the use of neural networks has reached the field of imaging. Neural networks can learn desired imaging features from high quality data sets and that information can be used to reconstruct data from accelerated acquisitions. Neural networks can be used to reduce artifacts typically found in image acceleration techniques. Neural networks can also reduce reconstruction speed. This section will illustrate how deep learning is being used in the context of fast imaging.
Fast quantitative MRI in a clinical setting: In current radiological practice, image interpretation is based on qualitative (i.e., visual) observations. Parametric or “quantitative” imaging (e.g., T1, T2 mapping) can provide information that is hard to extract visually. Quantitative MRI requires the acquisition of images at multiple contrasts to generate a parameter map. This section will show the advantages of combining Deep learning methods for accelerating acquisition and reconstructionmany of the principles outlined above for efficient parametric imaging in a clinical setting.

References

No reference found.