Apparatus for lesion diagnosis and method thereof

1 . An apparatus for lesion diagnosis, the apparatus comprising:

a processor; and

a memory configured to store one or more instructions,

wherein the processor configured to, by executing the one or more stored instructions, perform:

acquiring a medical image of a subject;

extracting a blood vessel region from the acquired medical image;

measuring a distance between blood vessel bifurcation points in the extracted blood vessel region; and

predicting a size of a lesion based on the measured distance,

wherein the medical image comprises a first image including the blood vessel region and a second image including both the blood vessel region and the lesion region,

the extracting a blood vessel region comprises:

extracting the blood vessel region from the first image, and

the predicting a size of the lesion comprises:

predicting the size of the lesion based on a relative size of a region including the blood vessel bifurcation points and the lesion region on the second image.

2 . The apparatus of claim 1 , wherein the extracting a blood vessel region comprises extracting the blood vessel region using a deep learning model configured to perform semantic segmentation.

3 . The apparatus of claim 2 , wherein the deep learning model comprises an encoder configured to perform a down-sampling process on the input image and a decoder configure to perform an up-sampling process on a feature map extracted during the down-sampling process.

4 . The apparatus of claim 2 , wherein the deep learning model comprises a first neural network and a second neural network having a structure corresponding to the first neural network,

the first neural network comprises a first encoder configured to perform a first down-sampling process on the input image and a first decoder configured to perform a first up-sampling process on a feature map extracted during the first down-sampling process, and

the second neural network comprises a second encoder configured to perform a second down-sampling process on a feature map output from the first decoder and a second decoder configured to perform a second up-sampling process on a feature map extracted during the second down-sampling process.

5 . The apparatus of claim 2 , wherein the medical image is a colonoscopy image,

the deep learning model is trained using a first training image set with correct answer label information and a second training image set without the correct answer label information,

the first training image set comprises a plurality of eye fundus images, and

the second training image set comprises a plurality of colonoscopy images.

6 . The apparatus of claim 1 , wherein the extracted blood vessel region is a blood vessel region located inside the lesion or to be adjacent to the lesion.

7 . The apparatus of claim 1 , wherein the measuring a distance between the blood vessel bifurcation points comprises:

performing image processing comprising a thresholding operation and a morphology operation on the extracted blood vessel region; and

measuring a distance between the blood vessel bifurcation points in the blood vessel region in which the image processing is performed.

8 . The apparatus of claim 1 , wherein the measuring a distance between the blood vessel bifurcation points comprises:

detecting a dense region from the extracted blood vessel region based on a density of the blood vessel; and

measuring the distance between the blood vessel bifurcation points located in the dense region.

9 . The apparatus of claim 1 , wherein the measuring a distance between the blood vessel bifurcation points comprises:

detecting a main blood vessel from the extracted blood vessel region based on a thickness of the blood vessel; and

measuring the distance between the bifurcation points formed in the main blood vessel.

10 . The apparatus of claim 1 , wherein the first image is an image which satisfies a predetermined capturing condition, and

the predetermined capturing condition comprises a condition that the blood vessel is located within a predetermined distance from a camera or a condition that the blood vessel is located in a predetermined range from a center of a viewing angle of the camera.

11 . The apparatus of claim 1 , wherein the medical image is a colonoscopy image, and the lesion is a polyp.

12 . A method for lesion diagnosis performed by a computing device, the method comprising:

acquiring a first medical image of a subject including a blood vessel region;

extracting the blood vessel region from the first medical image;

measuring a distance between blood vessel bifurcation points in the extracted blood vessel region,

acquiring a second medical image including both the blood vessel region and a lesion region; and

predicting a size of the lesion based on a relative size relationship between (i) a region including the blood vessel bifurcation points and (ii) the lesion region in the second medical image.