Classification method and apparatus

Patent metadata

Field	Value
Publication number	US-10163040-B2
Application number	US-201615216006-A
Country	US
Kind code	B2
Filing date	Jul 21, 2016
Priority date	Jul 21, 2016
Publication date	Dec 25, 2018
Grant date	Dec 25, 2018

Abstract

A medical image data processing apparatus comprises processing circuitry configured to: receive a plurality of sets of medical imaging data; and train a classifier for use in classification, wherein the training of the classifier comprises, for each of the plurality of sets of medical imaging data: selecting a first part and a second part of the respective set of medical imaging data, wherein the first part and the second part are representative of different regions of the same subject; and training the classifier for use in classification based on the first part of the set of medical imaging data and the second part of the set of medical imaging data.

First claim

The invention claimed is:

1. A medical image data processing apparatus comprising processing circuitry configured to:
receive a plurality of sets of medical imaging data; and
train a classifier for use in classification, wherein
the training of the classifier comprises, for each of the plurality of sets of medical imaging data:
selecting a first part and a second part of the respective set of medical imaging data, wherein the first part and the second part are representative of different regions of the same subject; and
training the classifier for use in classification based on the first part of the set of medical imaging data and the second part of the set of medical imaging data, and

the training of the classifier for use in classification based on the first part of the set of medical imaging data and the second part of the set of medical imaging data comprises training the classifier for use in determining presence or absence of at least one abnormality in the first part of the set of medical imaging data based at least partially on the second part of the set of medical imaging data and ground truth data associated with the first part of the set of medical imaging data.

2. An apparatus according to claim 1 , wherein the first part and the second part of the set of medical imaging data are representative of at least one of: substantially symmetrical regions; or regions having substantially the same shape.

3. An apparatus according to claim 1 , wherein:
the first part of the set of medical imaging data is representative of a region on one side of a line of symmetry of anatomical structure represented by the medical imaging data;
the second part of the set of medical imaging data is representative of a region on an opposite side of the line of symmetry; and
wherein the first part of the set of medical imaging data and second part of the set of medical imaging data are representative of corresponding anatomical features on opposite sides of the line of symmetry.

4. An apparatus according to claim 3 , wherein the training of the classifier for use in determining the presence or absence of at least one abnormality in the first part of the medical imaging data based at least partially on the second part of the set of medical imaging data is such as to at least partially normalise data values of the first part of the set of medical imaging data relative to the second part of the set of medical imaging data.

5. An apparatus according to claim 3 , wherein the processing circuitry is configured to, for each of the sets of medical imaging data, perform a folding process about said line of symmetry in respect of one of the first and second parts of the medical imaging data set such that imaging data of the first part at least partially overlays imaging data of the second part.

6. An apparatus according to claim 1 , wherein the processing circuitry is configured to select boundaries of the first and second parts based on at least one of ground truth data, expected position of one or more abnormalities, or a position of at least one selected anatomical feature.

7. An apparatus according to claim 3 , wherein the processing circuitry is configured to modify the ground truth data for at least some positions adjacent to regions of abnormality indicated by the ground truth data, and to use the modified ground truth data in training the classifier.

8. An apparatus according to claim 7 , wherein the processing circuitry is configured to adjust the importance of at least some of the modified ground truth data in training the classifier.

9. An apparatus according to claim 1 , wherein the processing circuitry is further configured to at least one of:
use atlas data in training the classifier;
use expected or actual position of at least one anatomical feature in training the classifier;
align at least one of the medical imaging data sets with at least one atlas data set to obtain registration data, and use the registration data in training the classifier.

10. A medical image data processing method comprising:
receiving a plurality of sets of medical imaging data; and
training a classifier for use in classification, wherein
the training of the classifier comprises, for each of the plurality of sets of medical imaging data:
selecting a first part and a second part of the respective set of medical imaging data, wherein the first part and the second part are representative of different regions of the same subject; and
training the classifier for use in classification based on the first part of the set of medical imaging data and the second part of the set of medical imaging data, and

the training of the classifier for use in classification based on the first part of the set of medical imaging data and the second part of the set of medical imaging data comprises training the classifier for use in determining presence or absence of at least one abnormality in the first part of the set of medical imaging data based at least partially on the second part of the set of medical imaging data and ground truth data associated with the first part of the set of medical imaging data.

11. A method according to claim 10 , wherein the first part and the second part of the set of medical imaging data are at least one of: representative of substantially symmetrical regions; or regions having substantially the same shape.

12. A method according to claim 10 , wherein:
the first part of the set of medical imaging data is representative of a region on one side of a line of symmetry of anatomical structure represented by the medical imaging data;
the second part of the set of medical imaging data is representative of a region on an opposite side of the line of symmetry; and
wherein the first part of the set of medical imaging data and second part of the set of medical imaging data are representative of corresponding anatomical features on opposite sides of the line of symmetry.

13. A method according to claim 12 , wherein the training of the classifier for use in determining the presence or absence of at least one abnormality in the first part of the medical imaging data based at least partially on the second part of the set of medical imaging data is such as to at least partially normalise data values of the first part of the set of medical imaging data relative to the second part of the set of medical imaging data.

14. A method according to claim 12 , further comprising, for each of the sets of medical imaging data, performing a folding process about said line of symmetry in respect of one of the first and second parts of the medical imaging data set such that imaging data of the first part at least partially overlays imaging data of the second part.

15. A method according to claim 10 , further comprising selecting boundaries of the first and second parts based on at least one of ground truth data, expected position of one or more abnormalities, or a position of at least one selected anatomical feature.

16. A method according to claim 12 , further comprising modifying the ground truth data for at least some positions adjacent to regions of abnormality indicated by the ground truth data, and using the modified ground truth data in training the classifier.

17. A method according to claim 16 , further comprising adjusting the importance of at least some of the modified ground truth data in training the classifier.

18. A method according to claim 10 , further comprising at least one of:
using atlas data in training the classifier;
using expected or actual position of at least one anato

Assignees

• Toshiba Medical Sys Corp

Inventors

• Poole Ian
• Lisowska Aneta
• Beveridge Erin
• Wang Ruixuan

Classifications

• G06V10/7788 Primary

  the supervisor being a human, e.g. interactive learning with a human teacher · CPC title

• G06V10/774

  Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title

• G06F18/2415

  based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate · CPC title

• G06F18/41

  Interactive pattern learning with a human teacher · CPC title

• G06F18/214

  Generating training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title