Systems and methods for deep learning based automated spine registration and label propagation
AssigneeGeneral Electric Company
InventorsRaghu Prasad, Jignesh Dholakia, Kavitha Manickam
Life of this patent
Estimating the term…
Abstract
Methods and systems are provided for whole-body spine labeling. In one embodiment, a method comprises acquiring a non-functional image volume of a spine, acquiring a functional image volume of the spine, automatically labeling the non-functional image volume with spine labels, automatically correcting the geometric misalignments and registering the functional image volume, and propagating the spine labels to the functional image volume. In this way, the anatomical details of non-functional imaging volumes may be leveraged to improve clinical diagnoses based on functional imaging, such as diffusion weighted imaging (DWI).
Claims18
A method for magnetic resonance imaging, comprising:
- automatically labeling a non-functional whole-body image volume of a spine with spine labels, wherein automatically labeling the non-functional whole-body image volume comprises inputting the non-functional whole-body image volume into a first trained neural network, wherein the first trained neural network automatically segments and labels the non-functional whole-body image volume;
- automatically registering a functional whole-body image volume of the spine to the non-functional whole-body image volume;
- and propagating the spine labels to the registered functional whole-body image volume.
A method, comprising:
- geometrically correcting, with a first trained neural network, a position of a spine in the functional whole-body MR image volume of the spine to generate a corrected functional image volume;
- propagating, with a second trained neural network, labels of the spine from a non-functional whole-body MR image volume of the spine to the corrected functional image volume;
- and
- displaying a view of the corrected functional image volume with the labels of the spine overlaid on the view;
- wherein the method further comprises automatically generating the labels of the spine in the non-functional whole-body MR image volume with a third trained neural network.
A system for automatically labeling a functional spine image, comprising:
- a first deep neural network configured to register a functional spine image by geometrically correcting the functional spine image according to a position of a spine in a non-functional spine image;
- and
- a second deep neural network configured to automatically propagate spine labels from the non-functional spine image to the geometrically corrected and registered functional spine image;
- wherein the system further comprises a third deep neural network configured to automatically generate the spine labels for the non-functional spine image.
Dependencies are inferred from the claim text.