Failure detection for segmentation of an image
A mechanism for determining a score indicative of a success of a segmentation of a 3D image, i.e. a success score. The mechanism proposes to obtain one or more 2D images of different target views of a target object in the 3D image, by processing a segmentation result of the 3D image. (A view of) each 2D image is classified using an automated classifier. The classification results are used to determine a success score, which may indicate, for example, whether, or how closely, the 3D segmentation result represents a ground truth segmentation result with sufficient accuracy, e.g. for clinical decision making.
1 . A computer-implemented method of determining a success score of a 3D segmentation result of a 3D image depicting a target object comprising an anatomical feature of a patient subject, the computer-implemented method comprising:
obtaining the 3D segmentation result of the 3D image depicting the target object;
generating a plurality of 2D images of the target object using the 3D segmentation result and the 3D image, each 2D image providing a view of the anatomical feature, wherein each 2D image provides a different view of the anatomical feature;
processing each generated 2D image using an automated classification process, wherein the automated classification process generates a classification result indicating a prediction of whether, or how closely, the view of the anatomical feature provided by the 2D image matches a predetermined target view of the 2D image; and
determining the success score of the 3D segmentation result by processing the classification results of all generated 2D images,
wherein the match is based on a predetermined requirement of containment and orientation of the anatomical feature within the predetermined target view.
2 . The computer-implemented method of claim 1 , wherein the classification result comprises one or more classification scores for a 2D image, each classification score indicating a likelihood that a view provided by a 2D image matches the predetermined target view.
3 . The computer-implemented method of claim 2 , wherein the step of determining the success score of the 3D segmentation result comprises processing the one or more classification scores of each 2D image, and the view of each 2D image, to determine the success score of the 3D segmentation result.
4 . The computer-implemented method of claim 3 , wherein the step of determining the success score of the 3D segmentation result comprises using a machine-learning algorithm to process the one or more classification scores of each 2D image, and the views of each 2D image, to determine the success score of the 3D segmentation result.
5 . The computer-implemented method of claim 1 , wherein the target object is a heart.
6 . The computer-implemented method of claim 5 , wherein the view provided by a at least one of the generated 2D images is one of: a view of 2 chambers of the heart; a view of 3 chambers of the heart; a view of 4 chambers of the heart; or a short axis view.
7 . The computer-implemented method of claim 1 , wherein the 3D segmentation result comprises a 3D mesh.
8 . The computer-implemented method of claim 1 , wherein the a plurality of 2D images comprise at least three 2D images.
9 . The computer-implemented method of claim 1 , wherein the automated classification process uses a machine-learning method to generate the classification result.
10 . A computer-implemented method of generating a 3D segmentation result of a 3D image depicting a target object comprising an anatomical feature of a patient subject, the computer-implemented method comprising:
processing the 3D image using a first segmentation algorithm to generate a 3D segmentation result; and
determining a success score of the 3D segmentation result by performing the method of claim 1 .
11 . The computer-implemented method of claim 10 , further comprising, in response to the success score of the 3D segmentation result indicating that the 3D segmentation result does not meet first criteria, processing the 3D image using a second, different segmentation algorithm to generate a new segmentation result.
12 . The computer-implemented method of claim 10 , further comprising:
in response to the accuracy of the 3D segmentation result indicating that the 3D segmentation result does not meet the first criteria, controlling a user interface to provide a user-perceptible output indicating that the 3D segmentation result is unsuccessful.
13 . The computer-implemented method of claim 10 , wherein the first segmentation algorithm comprises a machine-learning method configured to segment a 3D image.
14 . A computer program product embodied in a computer-readable non-transitory computer memory and comprising computer program code means which, when executed on a computing device having a processing system, cause the processing system to perform all of the steps of the method according to claim 1 .
15 . A processing system for determining a success score of a 3D segmentation result of a 3D image depicting a target object comprising an anatomical feature of a patient subject, the processing system comprising:
an obtaining module configured to obtain the 3D segmentation result of the 3D image depicting the anatomical feature;
a 2D image generating module configured to generate a plurality of 2D images of the anatomical feature using the 3D segmentation result and the 3D image, each 2D image providing a view of the anatomical feature, wherein each 2D image provides a different view of the anatomical feature;
a classifying module configured to process each generated 2D image using an automated classification process, wherein the automated classification process generates a classification result indicating a prediction of whether, or how closely, the view of the anatomical feature provided by the 2D image matches a predetermined target view of the 2D image; and
a success score determining module configured to determine the success score of the 3D segmentation result by processing the classification results of all generated 2D images,
wherein the match is based on a predetermined requirement of containment and orientation of the anatomical feature within the predetermined target view.