Global tractography based on machine learning

Patent metadata

Field	Value
Publication number	US-12507959-B2
Application number	US-202217930467-A
Country	US
Kind code	B2
Filing date	Sep 8, 2022
Priority date	Sep 21, 2021
Publication date	Dec 30, 2025
Grant date	Dec 30, 2025

Abstract

One or more tractograms of a global tractography of a tissue of interest are determined. At least one instance of diffusion magnetic resonance imaging data of the tissue of interest is obtained. A trained machine-learning algorithm generates the one or more tractograms based on the at least one instance of the diffusion magnetic resonance imaging data.

First claim

The invention claimed is:

1 . A computer-implemented method for determining one or more tractograms of a global tractography of a tissue of interest, the method comprising:
obtaining at least one instance of diffusion magnetic resonance imaging (dMRI) data of the tissue of interest; obtaining, from a tractogram database storing predetermined tractograms associated with one or more tissue of interests, at least one predetermined tractogram associated with the at least one instance of the dMRI data; and generating, by a trained machine-learning (ML) algorithm, the one or more tractograms by inputting both the at least one instance of the dMRI data and the at least one predetermined tractogram into the trained machine-learning algorithm, wherein the trained ML algorithm comprises an encoder network and a generator network, the encoder network configured to input the at least one instance of the dMRI data and outputs a latent representation, the generator network configured to input the latent representation and output the one or more tractograms.

2 . The computer-implemented method of claim 1 ,
wherein obtaining of the at least one predetermined tractogram comprises: retrieving, from the tractogram database, one or more tractograms within a distance metric to a tractogram representing the at least one instance of the dMRI data, wherein the at least one predetermined tractogram is selected from the one or more of the tractograms within the distance metric.

3 . The computer-implemented method of claim 2 ,
wherein retrieving of the one or more of the tractograms within the distance metric comprises: extracting, from the at least one instance of the dMRI data, a feature map representing the at least one instance of the dMRI data; comparing the feature map representing the at least one instance of the dMRI data with each feature map representing each predetermined tractogram in the tractogram database.

4 . The computer-implemented method of claim 3 ,
wherein each predetermined tractogram in the tractogram database has multiple deformed versions and each of the multiple deformed versions has a feature map representing the respective deformed version of the corresponding predetermined tractogram.

5 . The computer-implemented method of claim 1 ,
wherein obtaining of the at least one predetermined tractogram comprises: applying a rule-based algorithm to the at least one instance of the dMRI data to obtain the at least one predetermined tractogram.

6 . The computer-implemented method of claim 1 , further comprising:
concatenating the at least one instance of the dMRI data and the obtained at least one predetermined tractogram; and applying the trained ML algorithm to the concatenated data.

7 . The computer-implemented method of claim 1 ,
wherein the encoder network comprises a first encoder branch and a second encoder branch, the method further comprising: feeding the at least one instance of the dMRI data to the first encoder branch to generate a first feature representing the at least one instance of the dMRI data; feeding the obtained at least one predetermined tractogram to the second encoder branch to generate a second feature representing the obtained at least one predetermined tractogram; combining the first feature and the second feature to obtain a combination of both features; and feeding the combination to the generator network to generate the one or more tractograms.

8 . The computer-implemented method of claim 7 ,
wherein the first feature and the second feature are combined using a pooling layer.

9 . The computer-implemented method of claim 1 , further comprising
generating, based on the at least one instance of the dMRI data, at least one of an apparent diffusion coefficient map, a fractional anisotropy map, and a track-density imaging map.

10 . The computer-implemented method of claim 9 ,
wherein the trained ML algorithm comprises a generator network and the generator network comprises multiple branches, wherein the multiple branches respectively generate the one or more tractograms, the apparent diffusion coefficient map, the fractional anisotropy map, and the track-density imaging map.

11 . A system comprising:
at least one memory configured to store program code; at least one processor configured to load the program code from the at least one memory and execute the program code to: obtain at least one instance of diffusion magnetic resonance imaging (dMRI) data of tissue of interest; obtain, from a tractogram database storing predetermined tractograms associated with one or more tissue of interests, at least one predetermined tractogram associated with the at least one instance of the dMRI data; input the at least one instance of the dMRI data and the at least one predetermined tractogram into a trained machine-learning (ML) algorithm, wherein the trained ML algorithm comprises an encoder network and a generator network, the encoder network configured to input the at least one instance of the dMRI data and outputs a latent representation, the generator network configured to input the latent representation and output one or more tractograms; and generate, by the trained machine-learning algorithm, the one or more tractograms.

12 . The system of claim 11 , wherein the at least one processor is configured to retrieve, from the tractogram database, one or more tractograms within a distance metric to a tractogram representing the at least one instance of the dMRI data, wherein the at least one predetermined tractogram is selected from the one or more of the tractograms within the distance metric.

13 . The system of claim 12 , wherein the at least one processor is configured to extract, from the at least one instance of the dMRI data, a feature map representing the at least one instance of the dMRI data and compare the feature map representing the at least one instance of the dMRI data with each feature map representing each predetermined tractogram in the tractogram database, the retrieval being based on the comparison.

14 . The system of claim 13 , wherein each predetermined tractogram in the tractogram database has multiple deformed versions and each of the multiple deformed versions has a feature map representing the respective deformed version of the corresponding predetermined tractogram.

15 . The system of claim 11 , wherein the at least one processor is configured to apply a rule-based algorithm to the at least one instance of the dMRI data to obtain the at least one predetermined tractogram.

16 . The system of claim 11 , wherein the encoder network comprises a first encoder branch and a second encoder branch, wherein the at least one processor is configured to feed the at least one instance of the dMRI data to the first encoder branch to generate a first feature representing the at least one instance of the dMRI data, feed the obtained at least one predetermined tractogram to the second encoder branch to generate a second feature representing the obtained at least one predetermined tractogram, combine the first feature and the second feature to obtain a combination of both features, and feed the combination to the generator network to generate the one or more tractograms; wherein the generator network comprises multiple branches, the multiple branches configured to respectively generate the one or more tractograms, an apparent diffusion coefficient map, a fractional anisotropy map, and a track-density imaging map.

Assignees

• Siemens Healthineers Ag

Inventors

• Mostapha Mahmoud
• Mailhe Boris
• Comaniciu Dorin
• Janardhanan Nirmal
• Arberet Simon
• Lim Hongki
• Nadar Mariappan S

Classifications

• G01R33/56341 Primary

  Diffusion imaging · CPC title

• A61B2576/026

  for the brain · CPC title

• A61B5/055

  involving electronic [EMR] or nuclear [NMR] magnetic resonance, e.g. magnetic resonance imaging · CPC title

• A61B5/0042

  for the brain · CPC title

• A61B5/7267 Primary

  involving training the classification device · CPC title