Index calculating apparatus, index calculation method, and recording medium
In order to calculate an index that properly indicates the degree of accuracy of a prediction result outputted by a machine learning model, an index calculating apparatus includes at least one processor, and the at least one processor carries out: a comparing process of comparing a characteristic of a training image used in machine learning of the machine learning model which receives, as an input, an image to be subjected to inference, to classify the image and a characteristic of an input image to be inputted to the machine learning model at a time of inference; and a calculating process of calculating, in accordance with a difference between the characteristic of the training image and the characteristic of the input image which have been compared with each other in the comparing process, an index which indicates a possibility of error in a classification result obtained in a case where the input image is inputted to the machine learning model.
1 . An index calculating apparatus comprising
at least one processor, the at least one processor carrying out:
a comparing process of comparing
a characteristic of a training image used in machine learning of a machine learning model which receives, as an input, an image to be subjected to inference, to classify the image and
a characteristic of an input image to be inputted to the machine learning model at a time of inference; and
a calculating process of calculating, in accordance with a difference between the characteristic of the training image and the characteristic of the input image which have been compared with each other in the comparing process, an index which indicates a possibility of error in a classification result obtained in a case where the input image is inputted to the machine learning model,
wherein the characteristic of the training image and the characteristic of the input image comprises:
a level of noise contained in the input image and the training image; and
a latent vector obtained by inputting the training image and the input image to an autoencoder having been trained with use of the training image,
wherein the image to be subjected to inference is a pathological image in which a cell specimen is contained as a subject, and
the machine learning model is trained so as to output a result of classification regarding whether the cell specimen is a cancer cell.
2 . The index calculating apparatus according to claim 1 , wherein
the at least one processor further carries out:
a judging process of judging whether the index is equal to or greater than a threshold value; and
a notifying process of notifying the classification result and an indication of a possibility of error in the classification result, in a case where in the judging process, the index has been judged to be equal to or greater than the threshold value.
3 . The index calculating apparatus according to claim 2 , wherein
the at least one processor further notifies, in the notifying process, information regarding the index that has been judged to be equal to or greater than the threshold value, to assist a user in decision making, in a case where in the judging process, the index has been judged to be equal to or greater than the threshold value.
4 . An index calculation method comprising:
at least one processor comparing
a characteristic of a training image used in machine learning of a machine learning model which receives, as an input, an image to be subjected to inference, to classify the image and
a characteristic of an input image to be inputted to the machine learning model at a time of inference; and
the at least one processor calculating, in accordance with a difference between the characteristic of the training image and the characteristic of the input image which have been compared with each other in the comparing, an index which indicates a possibility of error in a classification result obtained in a case where the input image is inputted to the machine learning model,
wherein the characteristic of the training image and the characteristic of the input image comprises:
a level of noise contained in the input image and the training image; and
a latent vector obtained by inputting the training image and the input image to an autoencoder having been trained with use of the training image,
wherein the image to be subjected to inference is a pathological image in which a cell specimen is contained as a subject, and
the machine learning model is trained so as to output a result of classification regarding whether the cell specimen is a cancer cell.
5 . A computer-readable, non-transitory recording medium having recorded thereon a program for causing a computer to function as an index calculating apparatus,
the program causing the computer to carry out:
a comparing process of comparing
a characteristic of a training image used in machine learning of a machine learning model which receives, as an input, an image to be subjected to inference, to classify the image and
a characteristic of an input image to be inputted to the machine learning model at a time of inference; and
a calculating process of calculating, in accordance with a difference between the characteristic of the training image and the characteristic of the input image which have been compared with each other in the comparing process, an index which indicates a possibility of error in a classification result obtained in a case where the input image is inputted to the machine learning model,
wherein the characteristic of the training image and the characteristic of the input image comprises:
a level of noise contained in the input image and the training image; and
a latent vector obtained by inputting the training image and the input image to an autoencoder having been trained with use of the training image,
wherein the image to be subjected to inference is a pathological image in which a cell specimen is contained as a subject, and
the machine learning model is trained so as to output a result of classification regarding whether the cell specimen is a cancer cell.