Healthcare
❯
Image analysis
❯Medical Assistants
❯Accelerated acquisition of magnetic resonance images
Generation of computer tomography scans from magnetic resonance images
For:
Oncology hospitals, oncologists
Scope:
Train a model that generates CT images from MRI scans. Synthetic CT images may be used for radiation dose calculation in radiation therapy.
Goal:
Improved Employee Efficiency
Accelerated acquisition of magnetic resonance images
For:
Radiology departments, MRI vendors
Scope:
Innovations in MRI image formation
Goal:
Improved Employee Efficiency
AI platform for chest CT-scan analysis (early stage lung cancer detection)
For:
Healthcare authorities
Scope:
Detecting malignant neoplasms (lungs) on chest CT-scans
Goal:
Anticipate Risks, Improved Employee Efficiency
AI solution for end-to-end processing of cell microscopy images
For:
Biochemical, metabolomics and imaging branches of biomedicine
Scope:
Restoration of naturally distorted microscopy images for following visualization and analysis of meaningful patterns of protein formation inside living cells.
Goal:
Improved Product Development / R&D
Computer-aided diagnosis in medical imaging based on machine learning
Scope:
Detecting image anomalies.
Goal:
Improve Operation Efficiency
Accelerated acquisition of magnetic resonance images
For:
Radiology departments, MRI vendorsGoal:
Improved Employee EfficiencyProblem addressed
Developing new approaches to MRI image formation aimed at reducing image
acquisition time while maintaining the diagnostic image quality.
Scope of use case
Innovations in MRI image formation
Description
The excellent soft tissue contrast and flexibility of magnetic
resonance imaging (MRI) makes it a very powerful
diagnostic tool for a wide range of disorders, including
neurological, musculoskeletal, and oncological diseases.
However, the long acquisition time in the MRI machine,
which can easily exceed 30 min, leads to low patient
throughput, problems with patient comfort and compliance,
artefacts from patient motion, and high exam costs.
Increasing imaging speed has been a major ongoing research
goal since the advent of the MRI. By combining both
hardware developments (such as improved magnetic field
gradients) and software advances (such as new pulse
sequences), it has been possible to significantly reduce the
image acquisition times. One noteworthy development in
this context is parallel imaging, introduced in the 1990s,
which allows multiple data points to be sampled
simultaneously, rather than in a traditional sequential order
([238], [239]).
Compressed sensing ([240], [241]) techniques speed up the
MR acquisition by acquiring less measurement data than was
previously required to reconstruct diagnostic quality images.
Artefacts that are introduced by the violation of the Nyquist-
Shannon sampling theorem can be eliminated in the course
of image reconstruction. This can be achieved by
incorporating additional a priori knowledge during the
image reconstruction process.
The last two years have seen the rapid development of
machine learning approaches for MR image reconstruction,
which hold great promise for further acceleration of MR
image acquisition ([242], [243], [244]). To speed up the
algorithm development, public datasets are being provided
to the research community. For example, the fastMRI
challenge [245] introduced standardized evaluation criteria
and freely-accessible datasets to help the community make
rapid advances in state-of-the-art MR image reconstruction.
In machine learning-based approaches, the reconstruction
function is learned from the dataset of the input-output pairs
of samples drawn from a population. Such techniques also
leverage previous exam data to learn the spatial structure of
anatomy and typical image artefacts caused by under-
sampling. These attributes allow CNN-based methods to
reconstruct highly under-sampled data at higher fidelity
than CS schemes in certain cases [246].
The developed reconstruction algorithms may be deployed
either directly into the scanner console, or on the dedicated
reconstruction workstation or even on the cloud, depending
on the computational requirements.
The main challenge in clinical application of such deep
learning-based image formation algorithms is to guarantee
safety. For any device it is necessary to guarantee that the AI
system is not leading to diagnostic errors by removing or
introducing pathologies or other image features. It is also
necessary to guarantee image quality for all possible
combinations of MRI sequence parameters, anatomical
areas, and patient cohorts, or to be very conservative in
defining the limits of applicability.