Medical image analysis method and device using same
Provided is a computing device for supporting medical image analysis. The computing device may include: at least one or more memories for storing a program composed of one or more modules for performing medical image analysis; a communication unit for acquiring a medical image and a facial image for a patient's head; and at least one processor for detecting a plurality of feature points from the facial image through a feature point detection module performing artificial neural network-based machine learning to thus superimpose the medical image and the facial image on top of each other with respect to a plurality of first feature points displayed on the medical image and a plurality of second feature points as at least some of the feature points detected from the facial image, wherein the first feature points are anatomical reference points indicating relative positions of at least one of facial skeleton, teeth, and facial contour.
1 . A computing device for supporting medical image analysis, the computing device comprising:
at least one memory storing a computer program composed of at least one module for performing the medical image analysis;
a communication device receiving a medical image and a facial image of a patient's head; and
at least one processor;
detecting a plurality of feature points from the facial image using a feature point detection module, which performs machine learning based on artificial neural networks, and
controlling an image superimposition module to superimpose the medical image and the facial image on each other, using a plurality of first feature points displayed in the medical image and a plurality of second feature points, which is a subset of the plurality of feature points from the facial image, as a reference,
wherein the image superimposition module aligns the first feature points with the second feature points by utilizing Mean Squared Error (MSE), and Weighted Mean Squared Error (WMSE) is applied to the first feature points and the second feature points within a predetermined area around a mouth, and
wherein the first feature points are anatomical reference points indicating relative positions of at least one of facial skeleton, teeth, and facial contour.
2 . The computing device according to claim 1 , wherein the at least one processor controls the feature point detection module, which learns from training data accumulated with a plurality of different images, from which a specialist reads the second feature points.
3 . The computing device according to claim 1 , wherein the image superimposition module learns from accumulated training data, the accumulated training data including the medical image and the facial image that are superimposed on each other based on the first feature points and the second feature points by a specialist.
4 . The computing device according to claim 1 , wherein the image superimposition module calculates an offset value from a specific reference point, enlargement or reduction rates, and rotation angle, which minimizes distances between the first feature points and the second feature points, and based on the calculated values, the image superimposition module superimposes the medical image and the facial image on each other.
5 . The computing device according to claim 1 , wherein the feature point detection module detects a plurality of boundary boxes from the medical image, which is predicted to contain at least some of an individual anatomical features corresponding to the first feature points and determines given points within at least some of the detected boundary boxes as the first feature points.
6 . The computing device according to claim 1 , wherein the feature point detection module uses a landmark detection technique to detect a plurality of landmarks from the facial image, and wherein the landmark detection technique allows at least some of the points corresponding to the detected landmarks to be determined as the plurality of feature points.
7 . The computing device according to claim 1 , wherein the feature point detection module detects a plurality of boundary boxes which is expected that at least some of individual facial features corresponding to the plurality of feature points exist from the facial image and detects the plurality of feature points through an object detection technique in which given points included in at least some of the detected boundary boxes are determined as the feature points.
8 . A medical image analysis method comprising:
receiving a medical image and a facial image of a patient's head;
detecting, by at least one processor, a plurality of feature points from the facial image using a feature point detection module, which performs machine learning based on artificial neural networks; and
controlling, by the at least one processor, an image superimposition module to superimpose the medical image and the facial image on each other, using a plurality of first feature points displayed in the medical image and a plurality of second feature points, which is a subset of the plurality of feature points from the facial image, as a reference,
wherein the image superimposition module aligns the first feature points with the second feature points by utilizing Mean Squared Error (MSE), and Weighted Mean Squared Error (WMSE) is applied to the first feature points and the second feature points within a predetermined area around a mouth, and
wherein the first feature points are anatomical reference points indicating relative positions of at least one of facial skeleton, teeth, and facial contour.
9 . The method according to claim 8 , further comprising controlling the feature point detection module, which learns from training data accumulated with a plurality of different images, from which a specialist reads the second feature points.
10 . The method according to claim 8 , wherein the image superimposition module learns from accumulated training data, the accumulated training data including the medical image and the facial image that are superimposed on each other based on the first feature points and the second feature points by a specialist.
11 . The method according to claim 8 , wherein the image superimposition module calculates an offset value from a specific reference point, enlargement or reduction rates, and rotation angle, which minimizes distances between the first feature points and the second feature points, and based on the calculated values, the image superimposition module superimposes the medical image and the facial image on each other.
12 . The method according to claim 8 , wherein the feature point detection module detects a plurality of boundary boxes from the medical image, which predicted to contain at least some of an individual anatomical features corresponding to the first feature points and determines given points within at least some of the detected boundary boxes as the first feature points.
13 . The method according to claim 8 , wherein the feature point detection module uses a landmark detection technique to detect a plurality of landmarks from the facial image, and wherein the landmark detection technique allows at least some of the points corresponding to the detected landmarks to be determined as the plurality of feature points.
14 . The method according to claim 8 , wherein the feature point detection module detects a plurality of boundary boxes which is expected that at least some of individual facial features corresponding to the plurality of feature points exist from the facial image and detects the plurality of feature points through an object detection technique in which given points included in at least some of the detected boundary boxes are determined as the feature points.