Training apparatus, training method, recording medium, and classifying apparatus

A training method carried out by at least one processor included in a training apparatus for training a machine learning model which generates features of an input image to carry out an image recognition task, the training method comprising:

• acquiring a plurality of training images;
• and
• training the machine learning model with use of the plurality of training images,
• each of the plurality of training images belonging to one of a plurality of classes which are classifications appropriate to features of a subject contained in that training image, and belonging to one of a plurality of subclasses obtained by further dividing each of the plurality of classes into a plurality of classifications appropriate to the features of the subject,
• in the training,
• with respect to an image of interest included in the plurality of training images,
• a training image belonging to a subclass the same as a subclass to which the image of interest belongs being treated as a positive sample,
• a training image belonging to a subclass different from the subclass to which the image of interest belongs and belonging to a class the same as a class to which the image of interest belongs being treated as a quasi-positive sample,
• a training image belonging to a class different from the class to which the image of interest belongs being treated as a negative sample, and
• the machine learning model being trained such that
• features of the image of interest and features of the positive sample have a high degree of similarity to each other,
• the features of the image of interest and features of the quasi-positive sample have a low degree of similarity to each other,
• the features of the image of interest and features of the negative sample have a low degree of similarity to each other, and
• the degree of similarity between the features of the image of interest and the features of the quasi-positive sample is higher than the degree of similarity between the features of the image of interest and the features of the negative sample,
• the subject contained in each of the plurality of training images being a cell specimen,
• classification into the plurality of classes being made according to whether the cell specimen is benign or malignant, and
• classification into the plurality of subclasses being made according to type of the cell specimen, wherein:
• each of the training images further belongs to one of a plurality of middle classes which are obtained by dividing each of the plurality of classes into a plurality of classifications and each of which has one or more of the subclasses grouped together;
• and
• in the training, the at least one processor further treats,
• a training image which is the quasi-positive sample and which belongs to a middle class the same as a middle class to which the image of interest belongs, as a first quasi-positive sample, and
• a training image which is the quasi-positive sample and which belongs to a middle class different from the middle class to which the image of interest belongs, as a second quasi-positive sample,
• to train the machine learning model such that
• the features of the image of interest and features of the first quasi-positive sample have a higher degree of similarity to each other than the features of the image of interest and features of the second quasi-positive sample have.

A computer-readable, non-transitory recording medium having recorded thereon a program for causing a computer to function as a training apparatus for training a machine learning model which generates features of an input image to carry out an image recognition task, the program causing the computer to carry out:

• acquiring process of acquiring a plurality of training images;
• and
• training process of training the machine learning model with use of the plurality of training images,
• each of the plurality of training images belonging to one of a plurality of classes which are classifications appropriate to features of a subject contained in that training image, and belonging to one of a plurality of subclasses obtained by further dividing each of the plurality of classes into a plurality of classifications appropriate to the features of the subject,
• in the training,
• with respect to an image of interest included in the plurality of training images,
• a training image belonging to a subclass the same as a subclass to which the image of interest belongs, being treated as a positive sample,
• a training image belonging to a subclass different from the subclass to which the image of interest belongs and belonging to a class the same as a class to which the image of interest belongs, being treated as a quasi-positive sample,
• a training image belonging to a class different from the class to which the image of interest belongs, being treated as a negative sample, and
• the machine learning model being trained such that
• features of the image of interest and features of the positive sample have a high degree of similarity to each other,
• the features of the image of interest and features of the quasi-positive sample have a low degree of similarity to each other,
• the features of the image of interest and features of the negative sample have a low degree of similarity to each other, and
• the degree of similarity between the features of the image of interest and the features of the quasi-positive sample is higher than the degree of similarity between the features of the image of interest and the features of the negative sample,
• the subject contained in each of the plurality of training images being a cell specimen,
• classification into the plurality of classes being made according to whether the cell specimen is benign or malignant, and
• classification into the plurality of subclasses being made according to type of the cell specimen, wherein:
• each of the training images further belongs to one of a plurality of middle classes which are obtained by dividing each of the plurality of classes into a plurality of classifications and each of which has one or more of the subclasses grouped together;
• and
• in the training, the at least one processor further treats,
• a training image which is the quasi-positive sample and which belongs to a middle class the same as a middle class to which the image of interest belongs, as a first quasi-positive sample, and
• a training image which is the quasi-positive sample and which belongs to a middle class different from the middle class to which the image of interest belongs, as a second quasi-positive sample,
• to train the machine learning model such that
• the features of the image of interest and features of the first quasi-positive sample have a higher degree of similarity to each other than the features of the image of interest and features of the second quasi-positive sample have.