Detecting unacceptable detection and segmentation algorithm output
In an approach for automatically detecting whether an output of a detection and segmentation algorithm is of an acceptable quality, a processor receives an image. A processor applies a detection stage of a detection and segmentation algorithm to the image. A processor computes a set of features from a detection score map output by the detection stage of the detection and segmentation algorithm by analyzing the detection score map at more than one different operating points. A processor inputs the set of features into a classifier that predicts whether a final output of the detection and segmentation algorithm will be of an acceptable quality, wherein the acceptable quality is defined based on whether a detection precision threshold has been reached. A processor receives an output of the classifier.
1 . A computer-implemented method comprising:
receiving, by one or more processors, an image;
applying, by the one or more processors, a detection stage of a detection and segmentation algorithm to the image;
computing, by the one or more processors, a set of features from a detection score map output by the detection stage of the detection and segmentation algorithm by analyzing the detection score map at more than one different operating points, wherein the set of features are computed from the detection score map at different threshold values;
inputting, by the one or more processors, the set of features into a classifier that predicts whether a final output of the detection and segmentation algorithm will be of an acceptable quality, wherein the acceptable quality is defined based on whether a detection precision threshold has been reached;
receiving, by the one or more processors, an output of the classifier;
outputting a secondary image with the same dimensions and structure as the image, in which the intensity of each pixel is replaced with a number between 0 and 1 indicating the algorithm's level of confidence that this pixel belongs to a lesion.
2 . The computer-implemented method of claim 1 , further comprising:
responsive to the detection precision threshold being reached, displaying, by the one or more processors, the final output to a user through a user interface.
3 . The computer-implemented method of claim 1 , further comprising:
responsive to the detection precision threshold not being reached, preventing, by the one or more processors, the final output from being displayed to a user through a user interface.
4 . The computer-implemented method of claim 1 , wherein computing the set of features comprises:
converting, by the one or more processors, the detection score map into a binary map by applying a preset threshold to each score value for each voxel to obtain a corresponding binary value for each voxel;
applying, by the one or more processors, a connected component algorithm to group voxels with a same binary value together;
separating, by the one or more processors, respective grouped voxels with the same binary value; and
computing, by the one or more processors, the set of features based on the respective grouped voxels.
5 . The computer-implemented method of claim 4 , wherein computing the set of features further comprises:
repeating, by the one or more processors, the converting, the applying, the separating, and the computing steps using a different preset threshold.
6 . The computer-implemented method of claim 1 , wherein the classifier is a trained machine learning classifier model based on Random Forest method.
7 . The computer-implemented method of claim 3 , further comprising:
sending, by the one or more processors, a notification to the user that the image could not be reliably analyzed by the detection and segmentation algorithm.
8 . A computer program product comprising:
one or more computer readable storage media, comprising a non-transitory computer readable media, and program instructions collectively stored on the one or more computer readable storage media, the stored program instructions comprising:
program instructions to receive an image;
program instructions to apply a detection stage of a detection and segmentation algorithm to the image;
program instructions to compute a set of features from a detection score map output by the detection stage of the detection and segmentation algorithm by analyzing the detection score map at more than one different operating points, wherein the set of features are computed from the detection score map at different threshold values;
program instructions to input the set of features into a classifier that predicts whether a final output of the detection and segmentation algorithm will be of an acceptable quality, wherein the acceptable quality is defined based on whether a detection precision threshold has been reached;
program instructions to receive an output of the classifier; and
program instructions to output a secondary image with the same dimensions and structure as the image, in which the intensity of each pixel is replaced with a number between 0 and 1 indicating the algorithm's level of confidence that this pixel belongs to a lesion.
9 . The computer program product of claim 8 , further comprising:
responsive to the detection precision threshold being reached, program instructions to display the final output to a user through a user interface.
10 . The computer program product of claim 8 , further comprising:
responsive to the detection precision threshold not being reached, program instructions to prevent the final output from being displayed to a user through a user interface.
11 . The computer program product of claim 8 , wherein the program instructions to compute the set of features comprise:
program instructions to convert the detection score map into a binary map by applying a preset threshold to each score value for each voxel to obtain a corresponding binary value for each voxel;
program instructions to apply a connected component algorithm to group voxels with a same binary value together;
program instructions to separate respective grouped voxels with the same binary value; and
program instructions to compute the set of features based on the respective grouped voxels.
12 . The computer program product of claim 11 , wherein the program instructions to compute the set of features further comprise:
program instructions to repeat the converting, the applying, the separating, and the computing steps using a different preset threshold.
13 . The computer program product of claim 8 , wherein the classifier is a trained machine learning classifier model based on Random Forest method.
14 . The computer program product of claim 10 , further comprising:
program instructions to send a notification to the user that the image could not be reliably analyzed by the detection and segmentation algorithm.
15 . A computer system comprising:
one or more computer processors;
one or more computer readable storage media;
program instructions collectively stored on the one or more computer readable storage media for execution by at least one of the one or more computer processors, the stored program instructions comprising:
program instructions to receive an image;
program instructions to apply a detection stage of a detection and segmentation algorithm to the image;
program instructions to compute a set of features from a detection score map output by the detection stage of the detection and segmentation algorithm by analyzing the detection score map at more than one different operating points, wherein the set of features are computed from the detection score map at different threshold values;
program instructions to input the set of features into a classifier that predicts whether a final output of the detection and segmentation algorithm will be of an acceptable quality, wherein the acceptable quality is defined based on whether a detection precision threshold has been reached;
program instructions to receive an output of the classifier; and
program instructions to output a secondary image with the same dimensions and structure as the image, in which the intensity of each pixel is replaced with a number between 0 and 1 indicating the algorithm's level of confidence that this pixel belongs to a lesion.
16 . The computer system of claim 15 , further comprising:
responsive to the detection precision threshold being reached, program instructions to display the final output to a user through a user interface.
17 . The computer system of claim 15 , further comprising:
responsive to the detection precision threshold not being reached, program instructions to prevent the final output from being displayed to a user through a user interface.
18 . The computer system of claim 15 , wherein the program instructions to compute the set of features comprise:
program instructions to convert the detection score map into a binary map by applying a preset threshold to each score value for each voxel to obtain a corresponding binary value for each voxel;
program instructions to apply a connected component algorithm to group voxels with a same binary value together;
program instructions to separate respective grouped voxels with the same binary value; and
program instructions to compute the set of features based on the respective grouped voxels.
19 . The computer system of claim 18 , wherein the program instructions to compute the set of features further comprise:
program instructions to repeat the converting, the applying, the separating, and the computing steps using a different preset threshold.
20 . The computer system of claim 15 , wherein the classifier is a trained machine learning classifier model based on Random Forest method.