IMAGE RECONSTRUCTION

A BREAKTHROUGH IN SCAN ACCELERATION

A major limit of MRI has always been the long scan times. An advance in scan acceleration would be a major advance in medicine.

Deep learning based image reconstruction approaches show unprecedented performance in generating images from highly accelerated data.

AIRS’s approach is to achieve stability and generalizability in clinical practice by enforcing physics models in addition to the deep neural network inferences.

This allows us to reconstruct ground-truth-quality images from highly accelerated scans without generating additional artifacts.

QUALITY ASSESSMENT

ENSURING CLINICAL RELIABILITY

A new paradigm for evaluating image quality is needed in the era of nonlinear image reconstruction.

We are creating an image quality assessment system to objectify MRI image quality and locate unwanted artifacts.

This will play a key role in developing reliable AI reconstruction solutions, and more generally in improving quality control of MRI images.

IMAGE ENHANCEMENT

AUTOMATED REMOVAL OF ARTIFACTS

Once the image quality assessment system displays unwanted artifacts in medical images, a quality enhancement technique can remove these artifacts.

​We are developing a solution applicable to images with motion artifacts, metal artifacts, and hardware imperfections.

​Both physics models and data-driven knowledge are integrated into the technology to ensure enhanced outcomes.

SEQUENCE OPTIMIZATION

AUTOMATED PROTOCOL SELECTION

Optimal scan parameter settings are automatically suggested using a user defined trade-off between acceleration and image quality.

Data on scan parameters is integrated to develop a standardization to further improve sequences.

We are also working on applying our technology to multi-contrast MRI reconstruction sequences.