AI-based atlas mapping slice localizer for deep learning autosegmentation

Patent metadata

Field	Value
Publication number	US-12141966-B2
Application number	US-202117488208-A
Country	US
Kind code	B2
Filing date	Sep 28, 2021
Priority date	Sep 28, 2021
Publication date	Nov 12, 2024
Grant date	Nov 12, 2024

Abstract

Embodiments described herein provide a method for generating training data for Al based atlas mapping slice localization, and a system and method for using the training data to train a deep learning network. Training data development maps each slice of an input medical image to a position in a full body reference atlas along the longitudinal body axis. The method constructs a landmarking table of 2D slices indicating known anatomic landmarks of a reference subject, and interpolated slices. A final step for obtaining training data uses regression analysis techniques to create a vector of longitudinal axis coordinates of all slices from the input image. The training data is used to train a deep learning model to create an AI-based atlas mapping slice localizer model. The trained AI-based atlas mapping slice localizer model can be applied to generate mapping inputs to autosegmentation models to improve efficiency and reliability of contouring.

First claim

What we claim is:

1. A method comprising:
executing, by a processor, a machine learning model that receives an input of a first image of an anatomical region of a patient depicting a first organ having an outline to cause the machine learning model to predict boundary information of the anatomical region within a reference image;
selecting, by the processor, a computer model from a library of computer models based on the boundary information of the anatomical region, the computer model configured to execute a contouring protocol identifying outlines of organs within anatomical regions; and
transmitting, by the processor, the boundary information of the anatomical region of the first image to the computer model based on selecting the computer model from the library of computer models, whereby the computer model executes the contouring protocol to identify the outline of the first organ within the anatomical region.

2. The method of claim 1 , wherein the computer model is a second machine learning model configured to ingest the first image and identify the contour within the anatomical region.

3. The method of claim 1 , wherein the machine learning model outputs a mapping of a slice from the first image to the reference image.

4. The method of claim 1 , wherein the machine learning model is trained based on a set of second images of a set of second anatomical regions depicting a set of second organs having a set of second outlines.

5. The method of claim 4 , wherein at least one second image within the set of second images is labeled based on a reference point within the at least one second image to a corresponding anatomical region within the reference image.

6. The method of claim 1 , wherein the boundary information comprises a first limit and a second limit.

7. A method, comprising:
receiving, by a processor, a set of medical images comprising at least a first image and an atlas image, each image containing a plurality of two dimensional (2D) anatomic landmarks;
calculating, by the processor, longitudinal axis coordinates for each of the plurality of 2D anatomic landmarks in a first image space of the first image of the set of medical images and for each of the corresponding landmarks of the plurality of 2D anatomic landmarks in a reference image space of the atlas image;
executing, by the processor, a regression analysis between each of the 2D anatomic landmarks of the first image space and the corresponding 2D landmarks of the reference image space, wherein the regression analysis is applied to a plurality of regression segments corresponding to 2D slices from the set of medical images;
generating, by the processor, a vector of longitudinal axis coordinates for each slice of the medical images of the set of medical images;
mapping, by the processor, each vector of longitudinal axis coordinates of the first image to the atlas image; and
providing, by the processor, data associated with each vector of longitudinal axis coordinates to an analytics server to train a localizer model.

8. The method of claim 7 , wherein each of the 2D anatomic landmarks is one of the 2D slices from the first image that matches the corresponding 2D landmark defined on the atlas image.

9. The method of claim 7 , wherein the atlas image is a full body atlas, and wherein the longitudinal axis coordinates in the reference image space of the atlas image are coordinates within a total cranial-caudal range along a longitudinal body axis in the full body atlas.

10. The method of claim 7 , wherein the regression analysis between the 2D anatomic landmarks and the corresponding 2D landmarks is a piecewise linear regression.

11. The method of claim 10 , wherein the piecewise linear regression applies the regression analysis to the plurality of regression segments corresponding to the 2D slices from the set of medical images.

12. The method of claim 7 , wherein the vector of longitudinal axis coordinates for each of the medical images encompasses all 2D slices from the first image.

13. The method of claim 7 , wherein the vector of longitudinal axis coordinates for each of the medical images is mapped to the atlas image via estimated regression coefficients.

14. The method of claim 7 , further comprising training, by the processor, a deep learning model to determine a vector of longitudinal axis coordinates in an image space of an image analyzed by the model, wherein the deep learning model receives the set of medical images and the vector of longitudinal axis coordinates for each of the medical images as training inputs.

15. A system comprising:
a server comprising a processor and a non-transitory computer-readable medium containing instructions that when executed by the processor causes the processor to perform operations comprising:
execute a machine learning model that receives an input of a first image of an anatomical region of a patient depicting a first organ having an outline to cause the machine learning model to predict boundary information of the anatomical region within a reference image;
select a computer model from the library of computer models to execute a contouring protocol, whereby the contouring protocol identifies the outline of the first organ within the anatomical region; and
transmit the boundary information of the anatomical region of the first image to the computer model based on selecting the computer model form the library of computer models, whereby the computer model executes the contouring protocol to identify the outline of the first organ within the anatomical region.

16. The system of claim 15 , wherein the machine learning model is trained based on a set of second images of a set of second anatomical regions depicting a set of second organs having a set of second outlines.

17. The system of claim 16 , wherein at least one second image within the set of second images is labeled based on a reference point within the at least one second image to a corresponding anatomical region within the reference image.

18. The system of claim 15 , wherein the boundary information comprises a first limit and a second limit.

Assignees

• Varian Med Sys Inc
• Siemens Healthineers Int Ag

Inventors

• Genghi Angelo
• Siroki-Galambos Anna
• Coradi Thomas
• Fartaria Mário Joao
• Fluckiger Simon
• Haas Benjamin M
• Franco Fernando

Classifications

• G06T2207/20084

  Artificial neural networks [ANN] · CPC title

• G06T2207/20081

  Training; Learning · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G16H30/40

  for processing medical images, e.g. editing · CPC title

• G06T7/13

  Edge detection · CPC title