Image processing device, image processing method, and storage medium
The image processing device 1 X includes a first acquisition means 30 X, a second acquisition means 31 X, and an inference means 33 X. The first acquisition means 30 X acquires a set value of a first index indicating an accuracy relating to a lesion analysis. The second acquisition means 31 X acquires, for each of plural models which make inference regarding a lesion, a predicted value of a second index, which is an index of the accuracy other than the first index, on an assumption that the set value of the first index is satisfied. The inference means 33 X makes inference regarding the lesion included in an endoscopic image of an examination target, based on the predicted value of the second index and the plural models.
1 . An image processing device comprising:
at least one memory configured to store instructions; and
at least one processor configured to execute the instructions to:
acquire a set value of a first index indicating an accuracy relating to a lesion analysis;
acquire, for each of plural models which infer a lesion, a predicted value of a second index, which is an index of the accuracy other than the first index, on an assumption that the set value of the first index is satisfied; and
infer the lesion included in an endoscopic image of an examination target, based on the predicted value of the second index, the plural models, and a threshold value for the lesion associated with a combination of the set value of the first index and the predicted value of the second index.
2 . The image processing device according to claim 1 ,
is wherein the second index has a trade-off relation with the first index.
3 . The image processing device according to claim 1 ,
wherein a combination of the first index and the second index is a combination of sensitivity and a false positive rate or a combination of sensitivity and the precision.
4 . The image processing device according to claim 1 ,
wherein, for each of plural models, the at least one processor is further configured to execute the instructions to:
acquire correspondence information indicative of a correspondence relation between the first index and the second index, and
acquire, as the predicted value, a value of the second index corresponding to the set value according to the correspondence information.
5 . The image processing device according to claim 4 ,
wherein the correspondence information indicates an ROC curve, an LROC curve, an FROC curve, or a PR curve.
6 . The image processing device according to claim 4 ,
wherein curves of the plural models on two-dimensional coordinates with the first index and the second index as coordinate axes include intersecting points at which the curves intersect each other.
7 . The image processing device according to claim 1 ,
wherein the at least one processor is further configured to execute the instructions to:
select a model with a best accuracy indicated by the predicted value from the plural models, and
infer the lesion based on the selected model and the endoscopic image.
8 . The image processing device according to claim 1 ,
wherein the at least one processor is further configured to execute the instructions to:
generate an inference result obtained by weighting, based on the predicted value, each inference result based on the endoscopic image by the plural models.
9 . The image processing device according to claim 1 ,
wherein the at least one processor is further configured to execute the instructions to:
execute a first mode or a second mode while switching between the first mode and the second mode based on a predetermined condition,
wherein, in the first mode, the at least one processor is further configured to execute the instructions to:
select a model with a best accuracy indicated by the predicted value from the plural models, and
infer the lesion based on the selected model and the endoscopic image, and
wherein, in the second mode, the at least one processor is configured to execute the instructions to generate an inference result obtained by weighting, based on the predicted value, each inference result based on the endoscopic image by the plural models.
10 . The image processing device according to claim 9 ,
wherein the at least one processor is further configured to execute the instructions to:
execute the first mode or the second mode while switching between the first mode and the second mode based on whether or not a presence or absence of detection of a disorder other than a target disorder of the lesion analysis.
11 . The image processing device according to claim 9 ,
wherein the at least one processor is further configured to execute the instructions to:
execute the first mode or the second mode while switching between the first mode and the second mode based on an external input.
12 . The image processing device according to claim 1 ,
wherein the at least one processor is further configured to execute the instructions to:
select a predetermined number of models with top accuracies indicated by the predicted value from the plural models, and
generates an inference result obtained by weighting, based on the predicted value, each inference result by the predetermined number of models.
13 . The image processing device according to claim 1 ,
wherein each of the plural model is a model obtained through a machine learning in which sets of the endoscopic image and correct answer data are used as training data,
the correct answer data indicating an inference result to be outputted by the model when the endoscopic image is inputted to the model.
14 . The image processing device according to claim 1 ,
wherein the at least one processor is further configured to further execute the instructions to:
output an inference result by a display device or an audio output device to support examiner's decision making.
15 . An image processing method executed by a computer, the image processing method comprising:
acquiring a set value of a first index indicating an accuracy relating to a lesion analysis;
acquiring, for each of plural models which infer a lesion, a predicted value of a second index, which is an index of the accuracy other than the first index, on an assumption that the set value of the first index is satisfied; and
inferring the lesion included in an endoscopic image of an examination target, based on the predicted value of the second index, the plural models, and a threshold value for the lesion associated with a combination of the set value of the first index and the predicted value of the second index.
16 . A non-transitory computer readable storage medium storing a program executed by a computer, the program causing the computer to:
acquire a set value of a first index indicating an accuracy relating to a lesion analysis;
acquire, for each of plural models which infer a lesion, a predicted value of a second index, which is an index of the accuracy other than the first index, on an assumption that the set value of the first index is satisfied; and
infer the lesion included in an endoscopic image of an examination target, based on the predicted value of the second index, the plural models, and a threshold value for the lesion associated with a combination of the set value of the first index and the predicted value of the second index.