Predicting correctness of algorithmic segmentation
A prediction model is provided which is capable of predicting a correctness of a segmentation by a segmentation algorithm. The prediction model may be trained using a machine learning technique, and after training used to predict the correctness of a segmentation of a boundary in respective image portions of an image by the segmentation algorithm. The predicted correctness may then be visualized, for example as an overlay of the segmentation.
1. A system for generating a prediction model for predicting a correctness of segmentation by a segmentation algorithm, the system comprising:
a data interface for accessing:
training image data; and
labels for respective image portions of the training image data, wherein the labels are obtained by applying the segmentation algorithm to the training image data, wherein the labels are indicative of a correctness of the segmentation of a boundary in the respective image portion by the segmentation algorithm; and
a processor subsystem configured to:
train the prediction model using a machine learning technique and using said labelled image portions as input; and
output model data representing the trained prediction model, wherein the processor subsystem is further configured to generate the labels for the respective image portions by:
via the data interface, accessing:
first segmentation data representing the segmentation of the boundary in the respective image portions by the segmentation algorithm;
second segmentation data representing a corrected segmentation of the boundary in the respective image portions by a user;
determining the correctness of segmentation by comparing the first segmentation data with the second segmentation data; and
assigning the labels which are indicative of the correctness of segmentation to the respective image portions.
2. The system according to claim 1 , wherein the processor subsystem is configured to determine the correctness of the segmentation by the segmentation algorithm for an image portion based on a distance between corresponding parts of the respective segmentations of the boundary.
3. A workstation or imaging apparatus comprising the system according to claim 1 .
4. A system for predicting a correctness of segmentation by a segmentation algorithm, the system comprising:
a processor subsystem configured to:
identify image portions to which a prediction model is to be applied on the basis of the image portions containing part of a boundary;
apply the prediction model to respective image portions of an input image data to predict the correctness of the segmentation of the boundary in the respective image portions by the segmentation algorithm; and
generate output data indicative of the correctness of the segmentation to enable an incorrect segmentation of the boundary by the segmentation algorithm to be identified on the basis of the output data.
5. The system according to claim 4 , wherein the processor subsystem is further configured to:
apply the segmentation algorithm to the input image data, thereby obtaining the segmentation of the boundary in the input image data.
6. The system according to claim 5 , wherein:
the input image data is 3D image data;
the segmentation algorithm provides a 3D segmentation of the boundary in the input image data; and
the processor subsystem is configured to identify the image portions to which the prediction model is to be applied as sub-volumes of the 3D image data which are centred around respective parts of the 3D segmentation.
7. The system according to claim 4 , wherein the processor subsystem is configured to generate the output data as output image data containing a visualization of the correctness of the segmentation.
8. The system according to claim 7 , wherein the output image data contains the visualization of the correctness of the segmentation as an overlay over a visualization of the segmentation of the boundary.
9. The system according to claim 4 , wherein the processor subsystem is configured to adjust a visualization of the segmentation of the boundary to visually highlight one or more image portions in which the boundary is considered to be incorrectly segmented in accordance with a segmentation quality criterion.
10. The system according to claim 4 , wherein the processor subsystem is configured to:
execute a clinical application which uses the segmentation of the boundary by the segmentation algorithm; and
adjust a workflow of the clinical application on the basis of the correctness of the segmentation.
11. A non-transitory computer-readable medium comprising data representing a prediction model which is trained to predict a correctness of segmentation by a segmentation algorithm, wherein the prediction model is trained on labelled image portions of training image data, wherein labels of the labelled image portions are indicative of the correctness of the segmentation by the segmentation algorithm of a boundary in the respective image portion, wherein the labels comprise a distance value or color coding that is indicative of the correctness.
12. A computer-implemented method of predicting a correctness of segmentation by a segmentation algorithm, the method comprising:
accessing:
model data representing a prediction model which is trained on labelled image portions of training image data, wherein labels of the labelled image portions are indicative of a correctness of segmentation of a boundary in the respective image portion by the segmentation algorithm; and
input image data containing a boundary to be segmented by the segmentation algorithm;
applying the prediction model to respective image portions of the input image data to predict the correctness of the segmentation of the boundary in the respective image portions by the segmentation algorithm;
generating output data indicative of the correctness of segmentation to enable an incorrect segmentation of the boundary by the segmentation algorithm to be identified on the basis of the output data; and
adjusting a visualization of the segmentation of the boundary to visually highlight one or more image portions in which the boundary is considered to be incorrectly segmented in accordance with a segmentation quality criterion.
13. A computer-implemented method of generating a prediction model for predicting a correctness of segmentation by a segmentation algorithm, the method comprising:
accessing:
training image data;
labels for respective image portions of the training image data, wherein the labels are obtained by applying the segmentation algorithm to the training image data, wherein the labels are indicative of a correctness of the segmentation of a boundary in the respective image portion by the segmentation algorithm;
training the prediction model using a machine learning technique and using said labelled image portions as input; and
outputting model data representing the trained prediction model.
14. A non-transitory computer-readable medium comprising data representing instructions arranged to cause a processor system to perform the computer-implemented method according to claim 13 .