Machine learning model for automatic image registration quality assessment and correction

Patent metadata

Field	Value
Publication number	US-10424045-B2
Application number	US-201715629254-A
Country	US
Kind code	B2
Filing date	Jun 21, 2017
Priority date	Jun 21, 2017
Publication date	Sep 24, 2019
Grant date	Sep 24, 2019

Abstract

A medical registration training component executing within a medical registration system performs a training medical registration operation on a pair of medical studies. Responsive to the medical registration training system determining that the training medical registration operation succeeds, the medical registration training system records a medical registration instance for the pair of medical studies in a medical registration history and marks the medical registration instance as a positive instance in the medical registration history. Responsive to the medical registration training system determining that the training medical registration operation requires correction, the medical registration training system records a medical registration instance for the pair of medical studies in the medical registration history and marks the medical registration instance as a negative instance in the medical registration history. The medical registration training system trains a failure prediction machine learning model based on the medical registration history using machine learning such that the failure prediction machine learning model predicts whether a new medical registration operation will require correction. Responsive to the failure prediction machine learning model predicting that the new medical registration operation will require correction, the mechanism takes steps to automatically correct the new medical registration operation.

First claim

What is claimed is:

1. A computer program product comprising a non-transitory computer readable storage medium having a computer readable program stored therein, wherein the computer readable program, when executed on a computing device, causes the computing device to implement a medical registration system, wherein:
a medical registration training component executing within the medical registration system performs a training medical registration operation on a pair of medical studies, wherein each medical study within the pair of medical studies comprises at least one image and information about the medical study, wherein the training medical registration operation attempts to align the at least one image in a first medical study within the pair of medical studies with the at least one image in a second medical study within the pair of medical studies;
the medical registration training system records a medical registration instance for the pair of medical studies in a medical registration history and marks the medical registration instance as a positive instance in the medical registration history responsive to the medical registration training system determining that the training medical registration operation succeeds, wherein the medical registration history comprises a plurality of medical registration instances;
the medical registration training system records a medical registration instance for the pair of medical studies in the medical registration history and marks the medical registration instance as a negative instance in the medical registration history responsive to the medical registration training system determining that the training medical registration operation fails to converge;
the medical registration training system trains a failure prediction machine learning model based on the medical registration history using machine learning such that the failure prediction machine learning model predicts whether a new medical registration operation will fail to converge;
a registration component executing within the medical registration system performs the new medical registration operation and outputs metrics of the new medical registration operation;
a metric convergence analysis component executing within the medical registration system performs metric convergence analysis on the metrics output by the registration component;
the medical registration system applies the failure prediction machine learning model to the new medical registration operation to predict whether the new medical registration operation will fail to converge based on results of the metric convergence analysis; and
the medical registration system takes steps to automatically correct the new medical registration operation responsive to the failure prediction machine learning model predicting that the new automated medical registration operation will fail to converge.

2. The computer program product of claim 1 , wherein the information about the medical study includes information about a modality, a lesion size, a lesion type, a slice overlap, a body part, demographic information about the patient, disease history, use of contrast agents, and imaging protocols.

3. The computer program product of claim 1 , wherein the failure prediction machine learning model comprises a classifier model, wherein predicting whether the new medical registration operation will require correction comprises:
generating a feature vector for the new medical registration operation based on results of the medical registration operation and information about a pair of medical studies on which the new medical registration operation is being performed; and
applying the classifier model to the feature vector to classify the new medical registration operation as a positive instance or a negative instance.

4. The computer program product of claim 1 , wherein the failure prediction machine learning model comprises a regression model wherein predicting whether the new medical registration operation will require correction comprises:
generating a feature vector for the new medical registration operation based on results of the medical registration operation and information about a pair of medical studies on which the new medical registration operation is being performed;
applying the regression model to the feature vector to generate a failure probability value;
comparing the failure probability value to a threshold;
responsive to the failure probability value not being greater than the threshold, predicting that the new medical registration operation will not require correction; and
responsive to the failure probability value being greater than the threshold, predicting that the new medical registration operation will require correction.

5. The computer program product of claim 1 , wherein taking steps to automatically correct the new medical registration operation comprises at least one of performing region of interest (ROI) analysis to register two volumes, performing symmetric registration in both directions and checking convergence to a same point, or performing key organ segmentation and comparing a location of registered volumes.

6. The computer program product of claim 5 , wherein:
the medical registration training system records actions required to manually correct the training medical registration operation responsive to determining that the training medical registration operation requires correction; and
the medical registration training system trains a re-registration machine learning model based on the medical registration history using machine learning such that the re-registration machine learning model identifies a plurality of initialization points for the new medical registration operation.

7. The computer program product of claim 6 , wherein:
the medical registration system applies the re-registration machine learning model to the new medical registration operation to identify a cluster of medical registration instances in the medical registration history that are similar to the new medical registration operation responsive to the failure prediction machine learning model predicting that the new medical registration operation will fail to converge and identifies actions required to correct the medical registration operation for negative instances within the cluster of medical registration instances,
wherein taking steps to automatically correct the new medical registration operation comprises determining the plurality of initialization points for the new medical registration operation based on the identified actions.

8. The computer program product of claim 1 , wherein taking steps to automatically correct the new medical registration operation comprises:
determining a plurality of initialization points for the new medical registration operation;
performing re-registration of the new medical registration operation from the plurality of initialization points; and
selecting a best-converging result of the re-registration from the plurality of initialization points.

9. An apparatus comprising:
at least one processor; and
a memory coupled to the at least one processor, wherein the memory comprises instructions which, when executed by the at least one processor, cause the at least one processor to implement a medical registration system, wherein:
a medical registration training component executing within the medical registration system performs a training medical registration operation on a pair of medical studies, wherein each medical study within the pair of medical studies comprises at least one image and information about the medical study, wherein the training medical registration operation attempts to align the at least one image in a first medical s

Assignees

• IBM

Inventors

• Park Sun Young
• Sargent Dustin M

Classifications

• G06T2207/10081

  Computed x-ray tomography [CT] · CPC title

• G06T7/30 Primary

  Determination of transform parameters for the alignment of images, i.e. image registration · CPC title

• G06T2207/30004

  Biomedical image processing · CPC title

• G06T7/0002 Primary

  Inspection of images, e.g. flaw detection · CPC title

• G06T2207/20076

  Probabilistic image processing · CPC title