A method for extracting key points from a lateral cephalogram

Abstract

The present invention discloses a method for extracting key points from a lateral cephalogram, including: S1, collecting a plurality of lateral cephalograms; S2, performing conditional screening on the collected plurality of lateral cephalograms; S3, preprocessing the plurality of lateral cephalograms to obtain a grayscale image of the lateral cephalogram; S4, calculating the probability density of the image using a Gaussian probability density function to obtain the foreground region and the background region in the image; S5, using a corner feature method to extract the dynamic points among the key points of the head in the foreground region; S6, taking the FH plane on the head in the foreground region as the abscissa and the plane perpendicular to the FH plane as the ordinate to construct a two-dimensional rectangular coordinate system; S7, extracting the coordinate information of the dynamic points among the multiple key points of the head obtained in step S5. Compared with the traditional manual positioning, the present invention is faster and more efficient. Compared with the traditional software-assisted positioning, the present invention has high accuracy and is more convenient, fast and easy to operate.

Description

Technical Field

[0001] The present invention belongs to the technical field of cephalometric radiograph recognition, and particularly relates to a method for extracting key points of cephalometric radiographs. Background Art

[0002] Cephalometric radiographs are one of the criteria for doctors to judge whether a patient has dental protrusion or skeletal protrusion. During orthodontic treatment, doctors need to calibrate some key points based on the patient's cephalometric radiographs, and use these key points to calculate some medical index information, so as to use these index information for diagnosis and orthodontic treatment plan formulation.

[0003] The automation of X-ray projection measurement analysis makes the diagnosis and treatment design of dentofacial deformities more accurate, and greatly reduces the burden on operators.

[0004] For existing key point localization, there is a method of direct manual judgment, which has high requirements for the experience of specialist doctors, large variability, and significant differences in opinions among different doctors;

[0005] There is also a method of software-assisted determination, but it has high requirements for doctors' fixed-point experience, is time-consuming, has low efficiency and low accuracy. It has requirements for software accuracy, requires training to master the operation methods of relevant software, and the operation logic of some software is complex and difficult to get started.

[0006] The accuracy of key point localization will directly affect the reliability of measurement results. If the input of key points still remains at the level of manual fixed-pointing, it not only requires a lot of time, but also has strong subjective factors, inevitably generating human errors, endangering the accuracy of measurement results and thus affecting the orthodontic treatment effect. Summary of the Invention

[0007] The purpose of the present invention is to provide a method for extracting key points of cephalometric radiographs to solve the problem of inaccurate key point localization in existing cephalometric radiographs in view of the above deficiencies in the prior art.

[0008] To achieve the above object, the technical solution adopted by the present invention is:

[0009] A method for extracting key points of cephalometric radiographs, which includes the following steps:

[0010] S1. Collect a number of cephalometric radiographs;

[0011] S2. Perform condition screening on the collected number of cephalometric radiographs, and select multiple cephalometric radiographs that meet the conditions;

[0012] S3. Preprocess the multiple cephalometric radiographs in step S2 to obtain a grayscale image of the cephalometric radiograph;

[0013] S4. Calculate the probability density of the image in step S3 using the Gaussian probability density function. Based on the calculated probability density, obtain the foreground region and the background region in the image. The foreground region is the head contour in the lateral cephalogram.

[0014] S5. Use the corner feature method to extract the dynamic points among the key points of the head in the foreground region.

[0015] S6. Take the FH plane on the head in the foreground region as the abscissa, and the plane perpendicular to the FH plane as the ordinate to construct a two-dimensional rectangular coordinate system.

[0016] S7. Extract the coordinate information of the dynamic points among the multiple key points of the head obtained in step S5.

[0017] The present invention uses the corner feature extraction method and the Gaussian density function to extract the key points of the lateral cephalogram, and places the extracted key points in the constructed rectangular coordinate system, which is convenient for the accurate positioning of each key point.

[0018] Preferably, the lateral cephalograms collected in step S1 include lateral cephalograms of various bony types, different genders, different age stages, and different head pitch degrees.

[0019] The present invention selects lateral cephalograms of various different types for the extraction of key points, which is more accurate.

[0020] Preferably, in step S3, the preprocessing of the lateral cephalogram includes:

[0021] S3.1. Convert all the lateral cephalograms into images of the same size.

[0022] S3.2. Perform grayscale processing on the images in step S3.1.

[0023] S3.3. Perform Gaussian filtering on the images in step S3.2.

[0024] The present invention needs to convert all the images into images of the same size to facilitate the distinction between the foreground region and the background region in the later calculation and the unity of the later coordinate system construction; adopting grayscale processing and Gaussian processing can reduce image noise and reduce the level of details.

[0025] Preferably, in step S4, the Gaussian probability density function is used to calculate the probability density of the image in step S3. Based on the calculated probability density, the foreground region and the background region in the image are obtained. The foreground region is the head contour in the lateral cephalogram, and it includes:

[0026] S4.1. Calculate the probability density of the image in step S3 using the Gaussian probability density function:

[0027]

[0028] Among them, p(x) is the probability density of the current pixel point, and N(x|μ k , ∑k) is the k-th component of the Gaussian probability density function, and π k is the weight of N(x|μ k , ∑k), and μ k is the mean of the k-th component;

[0029] S4.2. Based on the probability density p(x) of the current pixel point, draw a grayscale histogram of the image. The grayscale histogram has double peaks and double valleys. One peak corresponds to the background area, and the other peak corresponds to the foreground area. The foreground area is the head contour area in the lateral head film.

[0030] Preferably, in step S5, the corner feature method is used to extract the dynamic points among the head key points in the foreground area, which includes:

[0031] S5.1. Extract corner features from the head contour in the foreground area:

[0032]

[0033] Among them, A(x) is the corner feature matrix, w(x, y) is the weight of the corresponding pixel point position, is the second derivative in the x direction, is the second derivative in the y direction, C x C y (x) is the first derivative in the x direction, and a, b, and c are respectively C x C y (x), corresponding values;

[0034] S5.2. Calculate the corner response value of the corner feature matrix:

[0035] D = detA - m(traceA) 2 = (ac - b) 2 - m(a + c) 2

[0036] Among them, D is the corner response value, det and trace are the determinant and trace operators, and m is a constant;

[0037] S5.3. When the corner response value is greater than the set threshold and is the local maximum in the neighborhood of this point, then this point is determined to be a corner point, and this corner point is the point with the largest pixel change in this neighborhood, which is the dynamic point among the head key points;

[0038] S5.4. Repeat steps S5.1 - S5.3 until all the dynamic points among the key points of the skull are found.

[0039] Preferably, the key points at least include: sella turcica point, nasion point, orbitale point, auriculare point, subspinale point, supramentale point, pogonion point, mental point, gnathion point, gonion point, incisal edge point of mandibular central incisor, incisal edge point of maxillary central incisor, labrale superius point, labrale inferius point, subnasale point, soft tissue pogonion point, posterior nasal spine, anterior nasal spine, articulare point, vertex point, center point of condyle, pterion point, nasion peak point, lowest point of temporomandibular joint, internal gonion point, apical point of maxillary central incisor, mesiobuccal cusp point of maxillary first permanent molar, mesiobuccal cusp point of mandibular first permanent molar, marginal ridge point of mandibular first permanent molar, apical point of mandibular central incisor, posterior border of ramus of mandible and posterior border of mandibular angle.

[0040] The method for extracting key points of a lateral cephalogram provided by the present invention has the following beneficial effects:

[0041] The present invention uses the Gaussian probability density function to calculate the region where the skull is located in the lateral cephalogram, combines the corner feature extraction method to extract the dynamic points of the key points within the region where the skull is located, and constructs a two - dimensional rectangular coordinate system based on the FH plane of the skull itself, and maps the dynamic points of the key points onto this two - dimensional rectangular coordinate system to accurately locate the dynamic points among the key points; compared with the traditional manual point - setting, the present invention is faster and more efficient, and compared with the traditional software - assisted point - setting, the present invention has high accuracy and is more convenient, fast and easy to operate. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 is a flowchart of the method for extracting key points of a lateral cephalogram. DETAILED DESCRIPTION OF THE INVENTION

[0043] The following describes the specific embodiments of the present invention to facilitate those skilled in the art of the present technology to understand the present invention, but it should be clear that the present invention is not limited to the scope of the specific embodiments. For those of ordinary skill in the art of the present technology, as long as various changes are within the spirit and scope of the present invention defined and determined by the appended claims, these changes are obvious, and all inventions created using the concept of the present invention are within the scope of protection.

[0044] According to an embodiment of the present application, referring to Figure 1 , for the method for extracting key points of a lateral cephalogram in this solution, this solution can achieve the extraction of key points of the lateral cephalogram, specifically extract the dynamic points among the key points. The dynamic points are the points with the largest position change among the key points. By extracting these dynamic points, it can assist doctors in accurately and quickly judging the lateral cephalogram. The specific steps are as follows:

[0045] Step S1. Collect several lateral cephalograms;

[0046] Step S2: Conditionally screen a number of lateral cephalograms collected, and select multiple lateral cephalograms that meet the conditions;

[0047] Step S3: Preprocess the multiple lateral cephalograms in Step S2 to obtain a grayscale image of the lateral cephalogram;

[0048] Step S4: Calculate the probability density of the image in Step S3 using the Gaussian probability density function, and obtain the foreground region and the background region in the image according to the calculated probability density. The foreground region is the head contour in the lateral cephalogram;

[0049] Step S5: Use the corner feature method to extract the dynamic points among the key points of the head in the foreground region;

[0050] Step S6: Use the FH plane on the head in the foreground region as the abscissa, and the plane perpendicular to the FH plane as the ordinate to construct a two-dimensional rectangular coordinate system;

[0051] Step S7: Extract the coordinate information of the dynamic points among the multiple key points of the head obtained in Step S5.

[0052] The present invention uses the corner feature extraction method and the Gaussian density function to extract the key points of the lateral cephalogram, and places the extracted key points in the constructed rectangular coordinate system, which is convenient for accurately positioning each key point.

[0053] Specifically, the following will describe each of the above embodiments in detail;

[0054] Step S1: Collect a number of lateral cephalograms;

[0055] The collected lateral cephalograms include lateral cephalograms of various bony types, different genders, different age stages, and different head pitch degrees. That is, multiple different types of lateral cephalograms are selected for key point extraction, which is more accurate and universal.

[0056] Step S2: Conditionally screen a number of lateral cephalograms collected, and select multiple lateral cephalograms that meet the conditions;

[0057] This step is mainly used for images that meet the clarity and quality standards, and removes images with obvious defects, such as missing parts, low clarity, and overly small images, etc.

[0058] Step S3: Preprocess the multiple lateral cephalograms in Step S2 to obtain a grayscale image of the lateral cephalogram, which specifically includes:

[0059] Step S3.1: Convert all the lateral cephalograms into images of the same size;

[0060] Convert all images into images of the same size to facilitate the distinction between the foreground area and the background area in subsequent calculations and the unity of the subsequent coordinate system construction;

[0061] Step S3.2: Perform grayscale processing on the images in Step S3.1;

[0062] Step S3.3: Perform Gaussian filtering on the images in Step S3.2.

[0063] By performing grayscale processing and Gaussian processing, image noise can be reduced and the level of details can be lowered to improve the accuracy of subsequent calculations.

[0064] Step S4: Use the Gaussian probability density function to calculate the probability density of the images in Step S3. According to the calculated probability density, obtain the foreground area and the background area in the images. The foreground area is the head contour in the lateral cephalogram, which includes:

[0065] Step S4.1: Use the Gaussian probability density function to calculate the probability density of the images in Step S3:

[0066]

[0067] where p(x) is the probability density of the current pixel point, N(x|μ k , ∑k) is the k-th component of the Gaussian probability density function, π k is the weight of N(x|μ k , ∑k), and μ k is the mean of the k-th component;

[0068] Step S4.2: Based on the probability density p(x) of the current pixel point, draw a grayscale histogram of the image. The grayscale histogram has double peaks and double valleys. One peak corresponds to the background area, and the other peak corresponds to the foreground area. The foreground area is the head contour area in the lateral cephalogram.

[0069] Through the calculation of this step, the original overall image can be converted into an image with only the head contour area, which can greatly reduce the amount of calculation and have higher accuracy for subsequent corner feature extraction.

[0070] Step S5: Use the corner feature method to extract the dynamic points among the head key points in the foreground area, which includes:

[0071] Step S5.1: Perform corner feature extraction on the head contour in the foreground area:

[0072]

[0073] where \(A(x)\) is the corner feature matrix, and \(w(x, y)\) is the weight of the corresponding pixel position. is the second derivative in the \(x\) direction. is the second derivative in the \(y\) direction, \(C\) x C y (x) is the first derivative in the \(x\) direction, and \(a\), \(b\), and \(c\) are respectively C x C y (x), the corresponding values.

[0074] Step S5.2: Calculate the corner response value of the corner feature matrix:

[0075] \(D = \det A - m(\text{trace}A)\) 2 \(=(ac - b)\) 2 \(-m(a + c)\) 2

[0076] where \(D\) is the corner response value, \(\det\) and \(\text{trace}\) are the determinant and trace operators, and \(m\) is a constant.

[0077] Step S5.3: When the corner response value is greater than the set threshold and is the local maximum within the neighborhood of this point, then determine that this point is a corner. This corner is the point with the largest pixel change within the neighborhood, which is the dynamic point among the key points of the skull.

[0078] Move the block area in any direction within the neighborhood of the pixel point. If there is a drastic change in intensity, then the pixel point at the change is a corner, and the threshold is set based on this. For the key points, if the skull in the lateral skull radiograph is normal, then the position of each key point is normal, that is, the intensity of this pixel point will not change strongly (as long as the change range is within the normal range, it is considered normal). On the contrary, if this key point has a lesion or is different from most people, then it will be detected as a corner, which is the dynamic point among the key points of the skull. If a corner is detected, it proves that the key points of this patient are significantly different from those of ordinary people. During the orthodontic treatment process, a targeted plan will be given, which can assist the doctor to perform orthodontic treatment more accurately. In addition, it should be noted that the present invention is only used to identify the dynamic points of the key positions in the lateral skull radiograph and does not involve treatment and diagnosis.

[0079] The key points in this embodiment at least include: sella turcica point, nasion point, orbitale point, auriculare point, supradentale point, sub-dentale point, pogonion point, menton point, gnathion point, gonion point, incisal point of mandibular central incisor, incisal point of maxillary central incisor, labrale superius point, labrale inferius point, subnasale point, soft tissue pogonion point, posterior nasal spine, anterior nasal spine, articulare point, vertex point, center point of condyle, pterion point, nasion peak point, lowest point of temporomandibular joint, internal gonion point, apical point of upper central incisor, mesiobuccal cusp point of upper first permanent molar, mesiobuccal cusp point of lower first permanent molar, marginal ridge point of lower first permanent molar, apical point of lower central incisor, posterior border of ramus of mandible and posterior border of lower mandibular angle;

[0080] Step S5.4: Repeat steps S5.1 to S5.3 until all the dynamic points among the key head points are found.

[0081] Step S6: Using the FH plane on the head in the foreground area as the abscissa and the plane perpendicular to the FH plane as the ordinate, construct a two-dimensional rectangular coordinate system;

[0082] This embodiment constructs a coordinate system based on the FH plane, which can more accurately reflect the dynamic information of the key points. And the coordinate system of this embodiment is not a two-dimensional rectangular coordinate system in the conventional sense, because there is often a certain angle between the FH plane and the horizontal plane, and this angle is an acute angle. Therefore, the rectangular coordinate system of this embodiment is an inclined rectangular coordinate system. And choosing the FH plane as the abscissa reference is more comparable for the positions of the key points and can better reflect the accuracy of their positions.

[0083] Step S7: Extract the coordinate information of the dynamic points among the multiple key head points obtained in step S5.

[0084] The present invention uses the Gaussian probability density function to calculate the area of the head in the lateral cephalogram, combines the corner feature extraction method to extract the dynamic points of the key points in the head area, and constructs a two-dimensional rectangular coordinate system based on the FH plane of the head itself, and maps the dynamic points of the key points onto this two-dimensional rectangular coordinate system to accurately locate the dynamic points among the key points; compared with the traditional manual point determination, the present invention is faster and more efficient, and compared with the traditional software-assisted point determination, the present invention has high accuracy and is more convenient, fast and easy to operate.

[0085] Although the specific implementation manners of the invention have been described in detail with reference to the accompanying drawings, it should not be construed as a limitation on the protection scope of this patent. Within the scope described in the claims, various modifications and deformations that can be made by those skilled in the art without creative labor still fall within the protection scope of this patent.

Claims

1. A method for extracting key points from a lateral cephalogram, characterized in that, It includes the following steps: S1. Collect several lateral cephalograms; S2. Screen the conditions of the several collected lateral cephalograms, and select multiple lateral cephalograms that meet the conditions; S3. Preprocess the multiple lateral cephalograms in step S2 to obtain a grayscale image of the lateral cephalogram; S4. Use the Gaussian probability density function to calculate the probability density of the image in step S3. According to the calculated probability density, obtain the foreground region and the background region in the image. The foreground region is the head contour in the lateral cephalogram; S5. Use the corner feature method to extract the dynamic points among the head key points in the foreground region; S6. Use the FH plane on the head in the foreground region as the abscissa and the plane perpendicular to the FH plane as the ordinate to construct a two-dimensional rectangular coordinate system; S7. Extract the coordinate information of the dynamic points among the multiple head key points obtained in step S5; In step S4, the Gaussian probability density function is used to calculate the probability density of the image in step S3. According to the calculated probability density, the foreground region and the background region in the image are obtained. The foreground region is the head contour in the lateral cephalogram, which includes: S4.

1. Use the Gaussian probability density function to calculate the probability density of the image in step S3; where p(x) is the probability density of the current pixel point, N(x|μ k , ∑k) is the k-th component of the Gaussian probability density function, and πk is the weight of N(x|μ k , ∑k), and μ k is the mean of the k-th component; S4.

2. Based on the probability density p(x) of the current pixel point, draw a grayscale histogram of the image. The grayscale histogram has double peaks and double valleys. One peak corresponds to the background region, and the other peak corresponds to the foreground region. The foreground region is the head contour region in the lateral cephalogram; In step S5, the corner feature method is used to extract the dynamic points among the head key points in the foreground region, which includes: S5.

1. Extract the corner features of the head contour in the foreground region; Among them, A(x) is the corner feature matrix, and ω(x, y) is the weight of the corresponding pixel position. is the second-order derivative in the x direction. is the second-order derivative in the y direction, C x C y (x) is the first-order derivative in the x direction, and a, b, and c are respectively C x C y (x), the corresponding values. S5.

2. Calculate the corner response value of the corner feature matrix; D = detA - m(traceA) 2 = (ac - b) 2 - m(a + c) 2 where D is the corner response value, det and trace are the operators of the determinant and the trace, and m is a constant; S5.

3. When the corner response value is greater than the set threshold and is the local maximum in the neighborhood of this point, then this point is determined to be a corner, and this corner is the point with the largest pixel change in the neighborhood, that is, the dynamic point among the head key points; S5.

4. Repeat steps S5.1 to S5.3 until all the dynamic points among the head key points are found.

2. The method for extracting key points from a lateral cephalogram according to claim 1, characterized in that, The lateral cephalograms collected in step S1 include lateral cephalograms of various bone types, different genders, different age stages, and different head pitch degrees.

3. The method for extracting key points from a lateral cephalogram according to claim 1, characterized in that, In step S3, the preprocessing of the lateral cephalogram includes: S3.

1. Convert all lateral cephalograms into images of the same size; S3.

2. Perform grayscale processing on the image in step S3.1; S3.

3. Perform Gaussian filtering on the image in step S3.

2.

4. The method for extracting key points from a lateral cephalogram according to claim 1, characterized in that, The key points at least include: sella turcica point, nasion point, orbitale point, auriculare point, supradentale point, sub-dentale point, pogonion point, menton point, gnathion point, gonion point, incisal point of mandibular central incisor, incisal point of maxillary central incisor, labrale superius point, labrale inferius point, subnasale point, pogonion point of soft tissue, posterior nasal spine, anterior nasal spine, articulare point, vertex point, center point of condyle, pterion point, nasion summit point, lowest point of temporomandibular joint, internal gonion point, apical point of upper central incisor, mesiobuccal cusp point of upper first permanent molar, mesiobuccal cusp point of lower first permanent molar, marginal ridge point of lower first permanent molar, apical point of lower central incisor, posterior border of ascending ramus of mandible and posterior border of lower mandibular angle.

Images