Feature-space clustering for physiological cycle classification

Patent metadata

Field	Value
Publication number	US-2022398717-A1
Application number	US-202117303868-A
Country	US
Kind code	A1
Filing date	Jun 9, 2021
Priority date	Jun 9, 2021
Publication date	Dec 15, 2022
Grant date	—

Abstract

Systems and methods are disclosed for performing operations comprising: receiving a plurality of training images representing different phases of a periodic motion of a target region in a patient; applying a model to the plurality of training images to generate a lower-dimensional feature space representation of the plurality of training images; clustering the lower-dimensional feature space representation of the plurality of training images into a plurality of groups corresponding to the different phases of the periodic motion; and classifying a motion phase associated with a new image of the target region in the patient based on the plurality of groups of the clustered lower-dimensional feature space representation of the plurality of training images.

First claim

What is claimed is:

1 . A system comprising:
a memory; and one or more processors that, when executing instructions stored in the memory, are configured to perform operations comprising:
receiving a plurality of training data representing different phases of a periodic motion of a target region in a patient;
applying a model to the plurality of training data to generate a lower-dimensional feature space representation of the plurality of training data;
clustering the lower-dimensional feature space representation of the plurality of training data into a plurality of groups corresponding to the different phases of the periodic motion; and
classifying a motion phase associated with a new data sample of the target region in the patient based on the plurality of groups of the clustered lower-dimensional feature space representation of the plurality of training data.

2 . The system of claim 1 , wherein applying the model comprises:
generating a Principal Component Analysis (PCA) representation of the plurality of training data; obtaining motion direction information for the periodic motion; and applying a correction factor to the PCA representation based on the motion direction information to generate a scaled PCA representation.

3 . The system of claim 2 , wherein applying the correction factor to the PCA representation comprises scaling a first PCA component of the PCA representation by the correction factor, wherein the motion phase associated with a new data sample is classified based on the scaled PCA representation.

4 . The system of claim 3 , wherein clustering the lower-dimensional feature space comprises:
splitting a range of a first PCA component of the PCA representation into a plurality of intervals each corresponding to a respective phase of the different phases of the periodic motion.

5 . The system of claim 4 , wherein a first interval of the plurality of intervals of the first PCA component corresponds to an end-exhalation phase, a second interval of the plurality of intervals corresponds to a mid-ventilation phase, and a third interval of the plurality of intervals corresponds to an end-inhalation phase.

6 . The system of claim 1 , wherein applying a model comprises applying a neural network autoencoder to the training data.

7 . The system of claim 6 , wherein the operations further comprise training the neural network autoencoder by:
obtaining a training data set; applying an encoder portion of the neural network autoencoder to the training data set to generate a plurality of feature vectors; applying a decoder portion of the neural network autoencoder to the plurality of feature vectors to generate reconstructed training data; computing a deviation based on a comparison of the training data and the reconstructed training data; and updating parameters of the neural network autoencoder based on the computed deviation.

8 . The system of claim 1 , wherein applying the model comprises applying a pre-trained feature extractor to the plurality of training data to generate a set of features corresponding to the lower-dimensional feature space representation of the plurality of training data.

9 . The system of claim 1 , wherein the operations further comprise:
receiving the new data sample of the target region; generating a new lower-dimensional feature representation for the new data sample; and classifying the new lower-dimensional feature based on its proximity to cluster centers to determine the motion phase associated with the new data sample.

10 . The system of claim 1 , wherein the operations further comprise obtaining motion direction information for the periodic motion.

11 . The system of claim 10 , wherein the operations further comprise:
applying an ordering function to the clustered lower-dimensional feature space representation based on the motion direction information to generate a direction-ordered clustered lower-dimensional feature space representation.

12 . The system of claim 1 , wherein the operations further comprise:
selecting a pair of data samples from the plurality of training data; and registering the pair of data samples to determine motion direction between the plurality of groups.

13 . The system of claim 12 , wherein the operations further comprise:
obtaining a first component of a Principal Component Analysis (PCA) representation of the plurality of training data; computing a difference between the first component of each of the plurality of training data; and selecting the pair of data samples from the plurality of training data in response to determining that the difference between the first component of the pair of data samples exceeds a threshold value.

14 . The system of claim 1 , wherein the operations further comprise accessing breathing information from an external device to determine motion direction information or accessing metadata associated with at least one of the plurality of training data to determine the motion direction information.

15 . The system of claim 1 , wherein the training data comprises a plurality of images collected from a magnetic resonance (MR) scanner, an ultrasound scanner, a kV image scanner, or an X-ray imaging device.

16 . The system of claim 1 , wherein applying a model comprises applying a machine learning model to the training data.

17 . A method comprising:
receiving a plurality of training data representing different phases of a periodic motion of a target region in a patient; applying a model to the plurality of training data to generate a lower-dimensional feature space representation of the plurality of training data; clustering the lower-dimensional feature space representation of the plurality of training data into a plurality of groups corresponding to the different phases of the periodic motion; and classifying a motion phase associated with a new data sample of the target region in the patient based on the plurality of groups of the clustered lower-dimensional feature space representation of the plurality of training data.

18 . The method of claim 17 , wherein applying the model comprises:
generating a Principal Component Analysis (PCA) representation of the plurality of training data; obtaining motion direction information for the periodic motion; and applying a correction factor to the PCA representation based on the motion direction information to generate a scaled PCA representation.

19 . A non-transitory computer readable medium comprising non-transitory computer readable instructions that, when executed by a processor, cause the processor to perform operations comprising:
receiving a plurality of training data representing different phases of a periodic motion of a target region in a patient; applying a model to the plurality of training data to generate a lower-dimensional feature space representation of the plurality of training data; clustering the lower-dimensional feature space representation of the plurality of training data into a plurality of groups corresponding to the different phases of the periodic motion; and classifying a motion phase associated with a new data sample of the target region in the patient based on the plurality of groups of the clustered lower-dimensional feature space representation of the plurality of training data.

20 . The non-transitory computer readable medium of claim 19 , wherein applying the model comprises:
generating a Principal Component Analysis (PCA) representation of the plurality of trainin

Assignees

• Elekta Inc

Inventors

• Hébert François
• Tremblay Sebastien
• Novosad Philip P

Classifications

• G06F18/214

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

• G06N3/045

  Combinations of networks · CPC title

• G06F18/2135

  based on approximation criteria, e.g. principal component analysis · CPC title

• G06F18/23

  Clustering techniques · CPC title

• G06F18/217

  Validation; Performance evaluation; Active pattern learning techniques · CPC title