Browse Prior Art Database

Q-SPACE MISSING DATA RECONSTRUCTION FOR ARBITRARY SAMPLING SCHEMES

IP.com Disclosure Number: IPCOM000240731D
Publication Date: 2015-Feb-24
Document File: 7 page(s) / 280K

Publishing Venue

The IP.com Prior Art Database

Abstract

The present invention provides a q-space missing data reconstruction technique that does not require q-space coordinates of data points as inputs. The technique is termed as q space deep learning (q-DL). The technique obtains scalar measures from advanced models at twelve-fold reduced scan time. The technique is based on inferring missing data from sampled data by means of a trained machine learning algorithm. Deep learning and a group of cutting-edge algorithm based on training artificial neural networks reduce pipeline to a single step. As a result, the technique provides solution to numerous problems of conventional diffusion MRI. In addition, q-DL is applicable for directing model-free tissue segmentation and lesion detection, compatibility with non-diffusion image contrasts, and generalization of missing q-space data reconstruction to arbitrary acquisition schemes.

This text was extracted from a Microsoft Word document.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 27% of the total text.

Q-SPACE MISSING DATA RECONSTRUCTION FOR ARBITRARY SAMPLING SCHEMES

BACKGROUND

The present invention relates generally to diffusion magnetic resonance imaging (diffusion MRI) and more particularly to a technique for providing q-space missing data reconstruction for arbitrary sampling schemes.

Diffusion magnetic resonance imaging (diffusion MRI) is a non-invasive method for assessing microstructure of materials and biological tissues. Diffusion MRI is generally based on diffusion of gases and liquids, primarily water. Advanced diffusion MRI models, such as diffusion kurtosis imaging (DKI) and neurite orientation dispersion and density imaging (NODDI) provide more accurate characterization of tissue microstructure. The models are more elaborate as compared to conventional model, such as, diffusion tensor imaging. As a result, the DKI and NODDI provide improved sensitivity. However, the models require significantly longer acquisition times.  

Further, data processing pipeline in quantitative diffusion MRI includes several steps, typically, model fitting of diffusion space or q space measurements, scalar parameter calculation, and abnormality detection. However, certain steps are prone to instabilities. As a result, the pipeline relies on considerable amounts of partly redundant input data which requires long acquisition time. This results in high scan costs and makes advanced diffusion models inapplicable for children and adults who are uncooperative, uncomfortable or very ill.

In diffusion MRI, a number of diffusion-weighted images (DWIs) for different diffusion weightings and directions constituting three-dimensional q-space are acquired. Signal intensity in these images includes information regarding diffusion properties. The task in quantitative diffusion MRI is to find a mapping from a limited number of noisy signal samples to rotationally invariant scalar measures that quantify microstructural tissue properties. This inverse problem is solved in each image voxel.

Conventional techniques of q-space missing data reconstruction are based on compressed sensing and use regularizations discretized for specific arrangements of data in q-space. As a result, the techniques impose data smoothness. Each of the techniques is designed for a specific structured q space sampling scheme, such as, Cartesian or radial sampling.

For example, a conventional technique uses model fitting approach for estimating scalar measures. Data processing pipeline of the technique includes fitting a diffusion model and calculating rotationally invariant measures from fitted model parameters.

Another conventional technique is considered if closed-form analytical solutions exist. For the diffusion model of DKI which requires approximately 150 DWIs it is observed that for certain DKI-based measures much fewer DWIs are sufficient. Further, these measures are analytically calculated from data in a single step. As a result, the technique helps in assuming that for many other sca...