Diagnosis support system, diagnosis support method, and storage medium
A diagnosis support system includes a processor. The processor is connected to a plurality of classifiers that are different in performance. The processor displays performance information of each of the classifiers side by side, receives a user's selection of the performance information displayed side by side, and inputs an input image to the classifier associated with the performance information selected by the user.
1 . A diagnosis support system comprising a processor,
the processor being connected to a plurality of classifier candidates that are different in performance;
the processor being configured to:
apply the plurality of classifier candidates to a first input image to obtain performance information corresponding to each of the plurality of classifier candidates, the first input image being an in-vivo image, wherein the performance information includes comparison information indicating differences in performance among the plurality of classifier candidates based on detection of a region of interest in the in-vivo image;
display the performance information of each of the plurality of classifier candidates side by side;
receive a user's selection of the performance information displayed side by side; and
based on the user's selection, input a second input image to a corresponding one of the plurality of classifier candidates, the corresponding one of the plurality of classifier candidates being associated with the performance information selected by the user.
2 . The diagnosis support system as defined in claim 1 , wherein
the first input image is used for acquiring the performance information and is one of a real-time image, a test image, or an image from a previous examination, and
the second input image input after the user's selection is a real-time image.
3 . The diagnosis support system as defined in claim 1 , wherein the plurality of classifier candidates has mutually different detection sensitivities to absence or presence of a region of interest.
4 . The diagnosis support system as defined in claim 3 , wherein the plurality of classifier candidates has at least one of the mutually different detection sensitivities to a shape of the region of interest, the mutually different detection sensitivities to a size of the region of interest, or the mutually different detection sensitivities to a color of the region of interest.
5 . The diagnosis support system as defined in claim 3 , wherein
the region of interest is a region corresponding to a lesion, and
the plurality of classifier candidates has mutually different detection sensitivities to medical classification of the region of interest.
6 . The diagnosis support system as defined in claim 1 , wherein the plurality of classifier candidates output mutually different detection results depending on an imaging state when the first input image is captured.
7 . The diagnosis support system as defined in claim 1 , wherein the plurality of classifier candidates use mutually different processing methods with respect to the first input image.
8 . The diagnosis support system as defined in claim 1 , wherein the plurality of classifier candidates include an original classifier candidate and a customized classifier candidate created by the user.
9 . The diagnosis support system as defined in claim 8 , wherein the customized classifier candidate is created based on machine learning using learning images including an image held by the user.
10 . The diagnosis support system as defined in claim 1 , wherein the processor is configured to perform a process of displaying respective detection results of the plurality of classifier candidates as the performance information.
11 . The diagnosis support system as defined in claim 10 , wherein the processor is configured to perform a process of simultaneously displaying respective detection results of the plurality of classifier candidates with respect to an identical first input image as the performance information.
12 . The diagnosis support system as defined in claim 1 , wherein the processor is configured to perform a process of displaying a list of a plurality of regions of interest that is detectable by each of the plurality of classifier candidates.
13 . The diagnosis support system as defined in claim 12 , wherein
each region of interest is a region corresponding to a lesion, and
the processor is configured to perform a process of displaying a list of a plurality of the lesions that is detectable by each of the plurality of classifier candidates, regarding each of a protruding lesion, a flat lesion, and a depressed lesion.
14 . The diagnosis support system as defined in claim 1 , wherein the processor is configured to perform a process of displaying data indicating a classification performance of the plurality of classifier candidates as the performance information.
15 . The diagnosis support system as defined in claim 14 , wherein the processor is configured to perform a process of displaying data indicating respective detection results of the plurality of classifier candidates with respect to test data as data indicating the classification performance of the plurality of classifier candidates.
16 . The diagnosis support system as defined in claim 14 , wherein the processor is configured to perform a process of displaying learning data used for learning of the plurality of classifier candidates as the data indicating the classification performance of the plurality of classifier candidates.
17 . The diagnosis support system as defined in claim 14 , wherein
the first input image is an in-vivo image in which a living body is captured, and
the processor is configured to perform a process of displaying data indicating appropriateness of a detection result in previous examination using the in-vivo image as the data indicating the classification performance of the plurality of classifier candidates.
18 . The diagnosis support system as defined in claim 1 , wherein
the processor is configured to receive the user's selection for selecting two or more classifier candidates among the plurality of classifier candidates, and
the processor is configured to perform a process of selecting a classifier candidate among the two or more classifier candidates selected by the user.
19 . The diagnosis support system as defined in claim 1 , wherein
the processor is configured to
receive a selection of respective classifier candidates by a plurality of the users, and
output a detection result from each of the classifier candidates in accordance with the user.
20 . The diagnosis support system as defined in claim 1 , wherein
the processor is configured to:
receive the user's first selection of a corresponding one of the classifier candidates before the user's observation of the second input image, and thereafter
receive the user's second selection of a corresponding one of the classifier candidates during the user's observation of the second input image.
21 . The diagnosis support system as defined in claim 1 , wherein the plurality of classifier candidates includes two or more of the classifier candidates generated by making different settings to a common trained model.
22 . The diagnosis support system as defined in claim 1 , wherein the processor is configured to display at least two types of performance that are in a trade-off relationship as the performance information.
23 . The diagnosis support system as defined in claim 22 , wherein the performance in the trade-off relationship is detection accuracy and processing time, or detection accuracy and a frequency of displaying a detection result.
24 . A diagnosis support method comprising:
applying a plurality of classifier candidates to a first input image to obtain performance information corresponding to each of the plurality of classifier candidates, the first input image being an in-vivo image, wherein the performance information includes comparison information indicating differences in performance among the plurality of classifier candidates based on detection of a region of interest in the in-vivo image,
the plurality of classifier candidates outputting mutually different detection results when detecting the region of interest from the first input image;
displaying the performance information of each of the plurality of classifier candidates side by side;
receiving a user's selection for selecting at least one of the plurality of classifier candidates as a classifier serving as an output target;
based on the user's selection, input a second input image to a corresponding one of the plurality of classifier candidates, the corresponding one of the plurality of classifier candidates being associated with the performance information selected by the user; and
outputting a detection result of the classifier selected by the user's selection,
the displaying including presenting at least two types of performance that are in a tradeoff relationship as the performance information.
25 . A non-transitory computer-readable storage medium storing a diagnosis support program that causes a computer to implement:
applying a plurality of classifier candidates to a first input image to obtain performance information corresponding to each of the plurality of classifier candidates, the first input image being an in-vivo image, wherein the performance information includes comparison information indicating differences in performance among the plurality of classifier candidates based on detection of a region of interest in the in-vivo image,
the plurality of classifier candidates outputting mutually different detection results when detecting the region of interest from the first input image;
displaying the performance information of each of the plurality of classifier candidates side by side;
receiving a user's selection for selecting at least one of the plurality of classifier candidates as a classifier serving as an output target;
based on the user's selection, input a second input image to a corresponding one of the plurality of classifier candidates, the corresponding one of the plurality of classifier candidates being associated with the performance information selected by the user; and
outputting a detection result of the classifier candidate selected by the user's selection.