Classification apparatus for pathologic diagnosis of medical image, and pathologic diagnosis system using the same

1. A classification apparatus for pathologic diagnosis of medical image, comprising:

a processor; and

a non-transitory storage medium having program instructions stored thereon, execution of which by the processor causes the classification apparatus to provide functions of:

a feature extraction unit configured to extract feature data for an input image using a feature extraction variable;

a feature vector transformation unit configured to transform the extracted feature data into a feature vector using a vector transform variable; and

a vector classification unit configured to classify the feature vector using a classification variable, and to output results of the classification of pathologic diagnosis for the input image;

wherein the feature extraction unit, the feature vector transformation unit and the vector classification unit are trained based on a first tagged image having a first tag indicative of presence or absence of a lesion and pathologic diagnosis information, a second tagged image having a first tag as well as a second tag indicative of location information of the lesion, and an image having no tag, and

wherein the feature extraction variable is trained based on the second tagged image; the transform variable is trained based on at least one of the first tagged image, the second tagged image and the image having no tag; and the classification variable is trained based on at least one of the first tagged image and the second tagged image.

2. The apparatus of claim 1 , wherein:

the feature extraction unit receives the plurality of patch images obtained by segmenting the input image, and extracts feature data for each of the patch images using the feature extraction variable;

the feature vector transformation unit transforms the feature data for the patch images into a fixed length feature vector using the transform variable; and

the vector classification unit classifies the fixed length feature vector using the classification variable, and outputs results of classification of pathologic diagnosis for the input image.

3. A pathologic diagnosis system, comprising:

the classification apparatus for pathologic diagnosis of a medical image set forth in claim 1 ;

another processor; and

another non-transitory storage medium having program instructions stored thereon, execution of which by the another processor causes the pathologic diagnosis system to provide functions of:

a testing module configured to output final analysis information for an input image based on results of pathologic diagnosis output for the input image by the apparatus for classification of pathologic diagnosis of a medical image;

a training module configured to train the apparatus for classification of pathologic diagnosis of a medical image based on tagged images stored in a tagged image database; and

a tagged image database configured to store tagged images, including first tagged images each having a first tag indicative of presence or absence of a lesion and pathologic diagnosis information, and second tagged images each having a second tag indicative of location information of a lesion as well as a first tag.

4. The system of claim 3 , wherein:

the training module:

generates patch images that are partial images of the input image input to the feature extraction unit, and separates the patch images into data for training and data for verification; and

adjusts the feature extraction variable of the feature extraction unit by repeating a process of adjusting the feature extraction variable and the classification variable so that a cost function is minimized for the data for training until a time at which the cost function for the data for verification is minimized.

5. The system of claim 4 , wherein:

the training module:

sets initial values of the feature extraction variable and the classification variable; and

adjusts the feature extraction variable and the classification variable by repeating i) a first process of calculating feature data and results of the classification based on a current feature extraction variable and a current classification variable, ii) a second process of calculating costs for the feature extraction variable and the classification variable using cost functions, and iii) a third process of calculating variations of the feature extraction variable and the classification variable so that the costs are minimized, calculating differences with respect to an existing feature extraction variable and an existing classification variable, and updating the feature extraction variable and the classification variable.

6. The system of claim 5 , wherein the training module extracts a plurality of pieces of feature data by inputting the plurality of patch images for the input image to the feature extraction unit, and estimates the transform variable adapted to transform the feature data into a fixed length feature vector, thereby enabling the feature vector transformation unit to be trained.

7. The system of claim 2 , wherein:

the training module:

generates patch images that are partial images of the input image, and separates the patch images into data for learning and data for verification; and

adjusts the classification variable of the vector classification unit by repeating a process of adjusting the classification variable so that a cost function is minimized for the data for training until a time at which a cost function for the data for verification is minimized.

8. The system of claim 7 , wherein:

the training module:

sets an initial value of the classification variable; and

adjusts the classification variable by repeating i) a first process of calculating results of the classification based on a current classification variable, ii) a second process of calculating a cost for the classification variable using a cost function, and iii) a third process of calculating a variation of the classification variable so that the cost is minimized, calculating a difference with respect to an existing classification variable, and updating the classification variable.