JP5397873B2 - Bone axis automatic extraction method of femur and tibia of knee joint, and bone axis automatic extraction program - Google Patents

Description

  The present invention is a technique for automatically extracting the femoral and tibial bone axes of a knee joint by automatically analyzing a lower limb fluoroscopic image on a computer in TKA (Total Knee Arthroplasty) preoperative planning and knee osteoarthritis diagnosis, etc. It is about.

  The medical treatment of the knee joint is generally performed on the basis of an X-ray photograph (X-ray fluoroscopic image). One particularly important medical care item is diagnostic imaging of the femoral and tibia bone axes involved in the knee joint. These bone axis information is also important information in total replacement artificial joint surgery as a treatment for knee joints.

  The conventional method can be divided into the following two methods: a visual / manual method and an interactive method. The former is a method based on radiographic data, and the latter is based on digital data of CT (computed tomography). The X-ray photograph is a photograph of the lower limbs in the diagnosis of the knee joint, which is a photograph in which three X-ray photographs are connected vertically (the right figure in FIG. 5). This device for accurate reading and digitization of lower limb photographs was developed very recently. On the other hand, the CT digital data is the same X-ray image as the X-ray image, but the imaging method is different and provides three-dimensional information with three-dimensional data. Diagnosis of the knee joint may be carried out based on X-ray photographs and using CT data in recent years. Image processing of CT data is performed in an interactive manner. An outline of the conventional method is described below.

[Visual and manual bone axis extraction method]
The visual method is a method in which a doctor visually observes an X-ray and manually extracts a bone axis according to the following procedure (FIG. 1).
A1: Detecting the knee gap line by visually checking the gap line of the knee joint, and setting the bone axis extraction range from 10 cm to 15 cm up and down from the line A2: Detecting the bone area of the femur in the upper bone axis extraction range The left and right bone boundary lines of the bone part are extracted, and the midline of these boundary lines is set as the bone axis of the femur.
A3: In the lower bone axis extraction range, the same bone boundary line detection as in A2 is performed, and the middle line is set as the bone axis of the tibia. As a result, the bone axes of the femur and tibia are extracted, and the femoral tibia lateral angle (FTA), which is an important item for medical care, can be easily calculated.

[Interactive Bone Axis Extraction Method]
The interactive method is a computer processing method using digital data of CT based on an interactive method involving a doctor (Non-Patent Document 1). That is, this is a method in which a doctor performs a difficult process using a computer and extracts a bone axis based on the processing. There are two kinds of information that the doctor is involved in the dialogue method: the knee gap line and the brightness value of the bone to be extracted. That is, as compared with the visual / manual work method described above, the visual method A1 can be performed from a given knee gap line. A2 and A3 can be processed by the following image processing method based on the information on the luminance value of the given bone.
B1: Extraction of bone region by threshold method The luminance value of the X-ray fluoroscopic image of the bone varies depending on the patient and the imaging conditions. Therefore, the luminance value given interactively is required. Based on this value, a bone region having the same luminance value can be extracted.
B2: Edge detection method and contour connection method The bone region obtained in B1 is binarized and the bone boundary is detected by the edge method, and continuous bone boundaries are extracted by the contour connection method, and these boundary lines are extracted. From the midline, the femoral and tibia bone axes can be obtained.

M. Fadda, D. Bertelli, S. Martelli, M. Marcacci, P. Dario, C. Paggetti, D. Caramella, D. Trippi: "Computer Assisted Planning for Total Knee Arthroplasty", Lecturer Notes In Computer Science N. 1205 Editors J. Troccaz, E. Grimson, R. Mosges "CVRMed-MRCAS 1997", Springer Verlag, London England UK, 1997, pp. 619-628. Kenichi Funamizu: "The world of digital medical images starting from now" The Japan Radiologist Association (2005) Fumito Ito, Akio Doi, Hideaki Haneishi, Masahiko Suzuki, Masayuki Tokuda, “Development of TKA Preoperative Planning Support System”, IEICE Technical Report (Medical Imaging Study Group), Vol. 106, No. 509, (pp. 129-132, 2007/1 / 26.3)

  However, the conventional bone axis extraction method increases the burden of preoperative planning due to the intervention of a doctor, and the preoperative planning and the diagnosis of knee osteoarthritis depend on the skill level of the doctor. There is a problem that it is difficult to keep the level of diagnosis such as preoperative planning and osteoarthritis of the knee by more than a certain level.

  In order to solve the above-described problem, the present invention provides an algorithm for automatically analyzing and automatically extracting femoral and tibia bone axes from digital data (Dycom data) of lower limb photographs as two-dimensional data. For the purpose.

In order to achieve the above object, the bone axis automatic extraction method according to claim 1 is a method of automatically analyzing the lower limb fluoroscopic image on a computer to automatically extract the femoral and tibia bone axes of the knee joint,
(A) a process of cutting out an image region of a living tissue from a lower limb fluoroscopic image;
(B) a process of deriving an image (hereinafter referred to as an absorption image) obtained by luminance inversion conversion from the fluoroscopic image cut out in the process (a), and extracting a knee gap line based on the luminance distribution of the absorption image;
(C) The absorption image obtained in the process (b) is converted into an image in which bone tissue is enhanced (hereinafter referred to as an enhanced image), and a bone boundary point is obtained based on the luminance distribution of the enhanced image and the knee gap line. Process,
(D) a process of setting a bone axis analysis range based on the bone boundary point obtained in the process (c) and the knee gap line, and deriving the bone axis from the bone boundary point of the bone axis analysis range;
The femoral and tibial bone axes of the knee joint are automatically extracted and the femoral tibia lateral angle (FTA: Femoro Tibial Angle) is automatically calculated by sequentially executing processes (a) to (d). It is characterized by.

  The method for automatically extracting a bone axis according to claim 2 is characterized in that in process (a), the width direction of the lower limb fluoroscopic image is i-axis and the height direction is j-axis. Using the limb information of the limb or the right limb, a minimum region is cut out from which the noise information other than the biological tissue image is removed from the lower limb fluoroscopic image.

The bone-axis automatic extraction method according to claim 3 is the image-cutting procedure in the i-axis direction in the cutting-out according to claim 2.
(1) The average luminance value IavbM (i) of the luminance distribution is obtained in the j-axis direction at each i-axis coordinate.
(2) In the IavbM (i) distribution, the i-axis coordinates where the luminance value changes from the both ends of the i-axis to a certain luminance value or more are IRngA1 and IRngA2, respectively, and the width (IRng A 2 −IRng A 1) / δ from these points. IRngB1 and IRngB2 are obtained as internal points shifted inward by the length of (δ: preset constant).
(3) The i-axis coordinate point IMinIti of the minimum luminance value in the section (IRngB1, IRngB2) is obtained.
(4) In the case of the left limb, the peak point of the maximum luminance value in the section (IRngB1, IMinIti) is ILine1, the peak point closer to IMinIti in the section (IMinIti, IRngB2) is ILine2, and ILine1 and ILine2 are i-axis coordinate A cut point is used. In the case of the right limb, the peak point of the maximum luminance value in the interval (IMinIti, IRngB2) is ILine2, the peak point closer to IMinIti in the interval (IRngB1, IMinIti) is ILine1, and ILine1 and ILine2 are the cut-out points of the i-axis coordinates. To do.

The bone axis automatic extraction method according to claim 4 is an image extraction procedure in the j-axis direction in the extraction according to claim 2.
(1) The average luminance value Javb [j] of the luminance distribution is obtained in the i-axis direction at each j-axis coordinate.
(2) In the Javb [j] distribution, coordinates whose luminance values have changed from the both ends of the j-axis to a predetermined set value or more are JLine1 and JLine2, respectively, and these coordinates are cut-out points in the j-axis direction.
It is characterized by that.

  The bone axis automatic extraction method according to claim 5 is a method of smoothing the distribution of Javb [j] in the j-axis direction in the cut-out section (JLine1, JLine2) in the image cut-out procedure in the j-axis direction according to claim 4. Distribution JavbM [j] is determined, and the section of the X-ray photograph seam line (hereinafter referred to as line noise) is specified using the fluctuation distribution JavbD [j] = Javb [j] −JavbM [j], and the line noise is determined. Line noise is eliminated by correcting the luminance distribution in the section by linear interpolation from the luminance distribution outside the section.

ここでIBmaxとIBminはそれぞれ透視画像の最大輝度値と最小輝度値とする。
（２）吸収画像の輝度分布IBrt1(j,i)を用いて式(2)により輝度差分分布DBrt(j)を求める。

（３）DBrt(j)のピークのj軸座標を膝間隙ラインとして抽出する。
ことを特徴とする。 The bone axis automatic extraction method according to claim 6, wherein in the process (b), the width direction of the image is the i axis and the height direction is the j axis.
(1) The luminance distribution IBrt1 (j, i) of the absorption image obtained by performing luminance inversion conversion on the fluoroscopic image IBrt0 (j, i) cut out in the process (a) is obtained by the equation (1).

Here, IBmax and IBmin are the maximum luminance value and the minimum luminance value of the fluoroscopic image, respectively.
(2) Using the luminance distribution IBrt1 (j, i) of the absorption image, the luminance difference distribution DBrt (j) is obtained by Equation (2).

(3) The j-axis coordinate of the peak of DBrt (j) is extracted as a knee gap line.
It is characterized by that.

（３）プロセス(a)で切り出された透視画像の輝度範囲が(IBmin,IBmax)になるようにIBrt2(j,i)を線形補間し、γピクセル（γ：予め設定された定数）のスケールの移動平均による平滑化を行い、j軸方向の圧縮画像を得る。
ことを特徴とする。ここでβは比較的大きいスケール（例えばβ=50ピクセル）であり、γは比較的小さいスケール（例えばγ=3ピクセル）とする。 In the bone axis automatic extraction method according to claim 7, in the extraction of the knee gap line according to claim 6, dynamic range compression (hereinafter referred to as DR compression) is performed in the j-axis direction as preprocessing, and the luminance fluctuation amount is calculated. Deriving the amplified luminance distribution, and DR compression filtering,
(1) Smoothing is performed by moving average of β pixels (β: a preset constant) only in the j-axis direction to create a blurred image Boke (j, i).
(2) Using the blurred image Boke (j, i) and the absorption image IBrt1 (j, i), create a luminance variation image IBrt2 (j, i) according to Equation (3).

(3) IBrt2 (j, i) is linearly interpolated so that the brightness range of the fluoroscopic image cut out in process (a) becomes (IBmin, IBmax), and the scale of γ pixel (γ: preset constant) To obtain a compressed image in the j-axis direction.
It is characterized by that. Here, β is a relatively large scale (for example, β = 50 pixels), and γ is a relatively small scale (for example, γ = 3 pixels).

The bone axis automatic extraction method according to claim 8, wherein in the process (c),
(1) An enhanced image that enhances the luminance level of bone tissue is derived from the absorption image obtained in process (b) using dynamic range compression image processing.
(2) A cluster of distributions (hereinafter referred to as islands) having a luminance level higher than a predetermined luminance level with respect to the luminance distribution in the i-axis direction, where the width direction of the enhanced image is the i-axis and the height direction is the j-axis. ) Is detected.
(3) If there are two islands detected in (2), the two islands are selected. If there are three or more islands detected in (2), determine whether the femur or tibia based on the knee gap line;
(3-1) For the femur, select two islands from the higher island.
(3-2) In the case of the tibia of the left limb, select the two islands on the left side with priority.
(3-3) In the case of the tibia of the right limb, select the two islands on the right side preferentially.
(4) Using the edge image from the two islands selected in the above (3), a position where the brightness value of the island becomes a peak is obtained, and the position is extracted as a bone boundary point.
It is characterized by that.

（３）圧縮画像IBrt2(j,i)から、式(4)〜式(6)に示す強調係数fact(j,i)を導入し、強調画像IBrt3(j,i)を作成する。ここで平均的輝度レベルIBrtLVL(j)はボケ画像Boke(j,i)において、各j軸座標に対して、i軸方向の全区間の輝度分布についてその平均値をMean、その標準偏差値をStd、αを予め設定した定数で求める。
fact(j,i)=Boke(j,i)/IBrtLVL(j) ・・・式(4)
IBrt3(j,i)= fact(j,i)＊IBrt2(j,i) ・・・式(5)
IBrtLVL(j)=Mean+α×Std ・・・式(6)
ことを特徴とする。 In the bone axis automatic extraction method according to claim 9, the derivation of the emphasized image according to claim 8
(1) on both the direction of j axis and i axis beta pixels: performs smoothing by moving average of (beta preset constant), creating a blurred image Boke (j, i).
(2) A compressed image IBrt2 (j, i) is created by Expression (3) using the blurred image Boke (j, i) and the absorption image IBrt1 (j, i) .

(3) From the compressed image IBrt2 (j, i), the enhancement coefficient fact (j, i) shown in the equations (4) to (6) is introduced to create the enhanced image IBrt3 (j, i). Here, the average luminance level IBrtLVL (j) is the average value of the luminance distribution of all sections in the i-axis direction and the standard deviation value thereof for each j-axis coordinate in the blurred image Boke (j, i). Std and α are obtained by preset constants.
fact (j, i) = Boke (j, i) / IBrtLVL (j) (4)
IBrt3 (j, i) = fact (j, i) * IBrt2 (j, i) (5)
IBrtLVL (j) = Mean + α × Std (6)
It is characterized by that.

  The bone axis automatic extraction method according to claim 10 is the bone boundary point extraction according to claim 8, wherein an edge image is derived by applying a Mexican hat filter to the emphasized image, and the edge image is used. A bone boundary point is obtained based on a luminance distribution of the enhanced image.

The bone axis automatic extraction method according to claim 11 is the bone boundary point extraction according to claim 8, wherein the bone boundary point is extracted from the bone boundary point to the bone boundary point by using a regression line approximation method in the j-axis direction in the height direction. It is characterized by detecting and correcting mixed noise.

The bone axis automatic extraction method according to claim 12, wherein the procedure for setting the bone axis analysis range in the process (d) includes:
(1)
A j-axis coordinate range (1 to Jmask1) in the height direction of the enhanced image IBrt3 (j, i) for obtaining the bone boundary point is determined. (Jmask1: j-axis coordinate whose distance from the knee gap is 8cm)
(2)
Temporary left and right bone boundary lines and midlines are determined by the regression line method from the left and right bone boundary points and their midpoints in the temporary analysis range (10 cm to 15 cm from the knee gap line (J15 to J10)).
(3)
Intersection C between the midline and the knee gap line. On the middle line, find point B where BC = 10 cm and point A where AC = 15 cm.
(Four)
Find a straight line perpendicular to the midline at points A and B, calculate the intersection (A1, A2, B1, B2) with the temporary left and right bone boundary lines obtained in (2) above, and the right bone boundary line The range (A2 to B2) for the left bone boundary line and the range (A1 to B1) for the left bone boundary line are the bone boundary analysis range.
It is characterized by that.

  In the bone axis automatic extraction method according to claim 13, the bone axis derivation procedure in process (d) derives the left and right bone boundary lines from the bone boundary points in the bone axis analysis range using a regression line method. A straight line that is equidistant from the left and right bone boundary lines is derived as a straight line of the bone axis.

  The bone axis automatic extraction program according to claim 14 causes a computer to execute the bone axis automatic extraction method according to any one of claims 1 to 13. If this program is executed by a computer, the same operation and effect can be realized in accordance with basically the same principle as in the method according to any one of claims 1 to 13.

According to the first aspect of the invention, the extraction of the knee gap and the bone axes of the femur and tibia is automatically analyzed on a computer to reduce the burden on the doctor's preoperative plan and to improve the skill so far. It has the effect of maintaining the level of diagnosis such as preoperative planning and osteoarthritis of the knee by a surgeon who has relied on.
In addition, by using the bone boundary point extraction analysis method based on the luminance distribution reflected in the bone tissue as the bone boundary detection algorithm, the effect of increasing the reliability of accuracy compared to the conventional method based on the simple threshold method of luminance values There is.

According to the second aspect of the present invention, it is possible to automatically analyze on a computer to cut out the minimum region including only the image region of the living tissue using the luminance distribution characteristic of the lower limb fluoroscopic image. Furthermore, it becomes possible to remove artificial noise. Here, for example, artificial noise includes the following.
(1) Outer frame of radiograph
(2) Seam line and clasp of three photos
(3) Scale display (4) Mark

According to the invention of claim 3, by performing the image cutout procedure in the i-axis direction, the analysis efficiency can be improved and artificial noise (for example, non-biological images at both ends, scale display) is removed. It becomes possible.

According to the invention of claim 4, by performing the image cut-out procedure in the j-axis direction, it is possible to improve the analysis efficiency and to remove artificial noise (for example, non-biological images at both ends). It becomes.

  According to the fifth aspect of the invention, in addition to the effect of the fourth aspect of the invention, it is possible to remove artificial noise (for example, a seam line of an X-ray photograph).

  According to the invention according to claim 6, the knee gap line is obtained by automatically analyzing on the computer using the luminance distribution characteristic based on the image (absorption image) obtained by reversing the luminance of the lower limb fluoroscopic image. Thus, the accuracy of extracting the knee gap line is ensured to a certain level or more, and the burden on the doctor is reduced.

  According to the invention of claim 7, in addition to the effect of the invention of claim 6, by deriving a luminance distribution in which the dynamic range is compressed in the j-axis direction and the luminance variation amount is amplified as preprocessing, This has the effect of improving the accuracy of line extraction.

  According to the eighth aspect of the present invention, the extraction of the bone boundary point can be performed by automatic analysis on the computer, and the accuracy of the extraction of the bone boundary point is ensured to a certain level or more, and the burden on the doctor is reduced. There is an effect to reduce.

  According to the ninth aspect of the invention, in addition to the effect of the eighth aspect of the invention, an enhanced image that enhances the luminance level of the bone part (femur and tibia) to be analyzed is derived using dynamic range compression image processing. By doing so, there is an effect of improving the accuracy of extracting the bone boundary point.

  According to the invention of claim 10, in addition to the effect of the invention of claim 8, an effect of further improving the accuracy of extracting the bone boundary point by using an edge image obtained by applying a Mexican hat filter to the enhanced image There is.

  According to the invention of claim 11, in addition to the effect of the invention of claim 8, there is an effect of further improving the accuracy of bone boundary point extraction by detecting and correcting noise mixed in the bone boundary point. is there.

  According to the twelfth aspect of the present invention, the setting of the bone axis analysis range can be performed by automatic analysis on the computer, and the setting of the bone axis analysis range becomes accurate, and the effect of reducing the burden on the doctor is achieved. is there.

  According to the invention of claim 13, the bone axis can be derived by automatic analysis on a computer, the accuracy of bone axis derivation is ensured to a certain level or more, and the burden on the doctor is reduced. effective.

  According to the invention of claim 14, the same function and effect as those of the inventions of claims 1 to 13 can be realized.

  Next, the bone axis automatic extraction method according to the embodiment of the present invention will be described with reference to the drawings. In addition, this invention is not limited by this embodiment.

FIG. 2 is a flowchart showing the entire bone axis automatic extraction method according to the embodiment of the present invention. The bone axis automatic extraction method of the present invention automatically extracts the femoral and tibial bone axes when X-ray fluoroscopic limb images (Dycom data) are input to a computer and information on the right limb or the left limb is given. It is a method. As shown in FIG. 2, the bone axis automatic extraction method is:
C1: a process of cutting out an image region of a living tissue from a lower limb fluoroscopic image;
C2: a process of deriving an image (hereinafter referred to as an absorption image) obtained by luminance inversion conversion from the fluoroscopic image cut out in the process C1, and extracting a knee gap line based on the luminance distribution of the absorption image;
C3: converting the absorption image obtained in the process C2 into an image in which bone tissue is enhanced (hereinafter referred to as an enhanced image), and obtaining a bone boundary point based on the luminance distribution of the enhanced image and the knee gap line;
C4: a process of setting a bone axis analysis range based on the bone boundary point obtained in process C3 and the knee gap line, and deriving a bone axis from the bone boundary point of the bone axis analysis range;
And the bone axes of the femur and tibia of the knee joint are automatically extracted by sequentially executing processes C1 to C4 on the computer, and the femoral tibia lateral angle (FTA) is automatically calculated. Below, each process of C1 to C4 is demonstrated.

[C1: Lower-limb analysis image region extraction process]
In order to remove noise information other than the biological tissue image, the process C1 uses the lower limb perspective image around the knee joint as the i-axis and the height direction as the j-axis (the image coordinate system is shown in FIG. 4). Using the luminance distribution characteristics in the i-axis direction and the j-axis direction and the limb information on the left limb or the right limb, the minimum region including only the image region of the living tissue is cut out. FIG. 3 is a flowchart showing the extraction algorithm. This cutout method not only minimizes the size of an analysis image for improving processing efficiency but also removes artificial noise that causes trouble in the bone boundary extraction method described later. As can be seen from FIG. 5 (right figure), the artificial noise includes, for example, the following.
(1) Outer frame of radiograph
(2) Seam line and clasp of three photos
(3) Scale display (4) Mark

[C11: How to cut out images in the i-axis direction]
Next, the image cutout method (C11) in the i-axis direction will be described. As can be seen from the lower limb fluoroscopic image in FIG. 5, a portion having a high X-ray absorption rate, such as a bone, has a low luminance value, resulting in a black image. In order to separate a living tissue as one lump, it can be said that it is reasonable to first cut out an image from the i-axis direction. Therefore, the average distribution of the luminance used is a distribution obtained by calculating the average value of the luminance distribution in the j-axis direction at each i-axis coordinate and smoothing in the i-axis direction (FIG. 6). Moreover, in order to ensure the cut-out in the i-axis direction, information on the right limb or the left limb is required. The analysis procedure is as follows.
1) At each i-axis coordinate, the average luminance value IavbM (i) of the luminance distribution is obtained in the j-axis direction.
2) In the IavbM (i) distribution, the i-axis coordinates where the luminance value has changed beyond a certain luminance value from both ends of the i-axis are IRngA1 and IRngA2, respectively, and from these points, the width is removed to remove the outer frame of the photograph. IRngB1 and IRngB2 are obtained as internal points shifted inward by the length of (IRng A 2 −IRng A 1) / δ (for example, δ = 5).
3)
Section (IRngB1,
The i-axis coordinate point (IMinIti) of the minimum luminance value in IRngB2) is obtained.
Four)
In the case of the left limb, the peak point of the maximum luminance value in the interval (IRngB1, IMinIti) is ILine1, the peak point closer to IMinIti in the interval (IMinIti, IRngB2) is ILine2, and ILine1 and ILine2 are the cut-out points of the i-axis coordinates. To do. In the case of the right limb, the peak point of the maximum luminance value in the interval (IMinIti, IRngB2) is ILine2, the peak point closer to IMinIti in the interval (IRngB1, IMinIti) is ILine1, and ILine1 and ILine2 are the cut-out points of the i-axis coordinates. To do. Therefore, the size of the image in the i-axis direction is WidthI = Iline2-ILine1.
In 4) above, information on the right limb or left limb was introduced to remove the scale display from the artificial noise. That is, the scale display is usually placed outside the vicinity of the limb to be examined. From the information on the right limb or the left limb, it is possible to identify the section with the scale among the two sections in 4). In that section, the peak to be selected is closer to the IMinIti point than the peak of the maximum luminance value. Here, the definition of the peak point is a point having a luminance value larger than two points on both sides of the point, that is, the surrounding four points.

[C12: Image cutout method in the j-axis direction]
Next, the image cutout method (C12) in the j-axis direction will be described. The clipping in the j-axis direction is performed for the following two purposes based on the image processed by the image clipping method (C11) in the i-axis direction.
(1) Removal of non-biological image areas at both ends (2) Luminance correction of X-ray photograph seam line (this is called line noise)

The above (1) is performed by the following procedure in the same manner as C11 in order to improve the efficiency of image processing.
1) At each j-axis coordinate, an average luminance value Javb [j] (FIG. 7) of the luminance distribution is obtained in the i-axis direction.
2) In the Javb [j] distribution, coordinates whose luminance values have changed from the both ends of the j-axis to a preset value or more are JLine1 and JLine2, respectively, and these coordinates are cut-out points in the j-axis direction. Therefore, the size of the image in the j-axis direction is HeightJ = JLine2-JLine1.

  Above (2) is the cut-out section (JLine1, JLine2), and the distribution JavbM [j] obtained by smoothing the distribution of Javb [j] in the j-axis direction is obtained, and the fluctuation distribution JavbD [j] = Javb [j] −JavbM [ j] for analysis. As shown in FIG. 8, JavbD [j] takes the maximum value at two line noises (points a and b in FIG. 8). Line noise, which is a problem in bone axis analysis, crosses the tibia in the latter half of the j-axis coordinates (point b in FIG. 8) and needs to be eliminated. The elimination method was as follows. The j-axis coordinates Jpst and Jped having a luminance value that is 1/10 of the peak luminance value are obtained on both sides from the peak coordinate (point b). The luminance distribution in the section (Jpst, Jped) that is line noise is corrected by linear interpolation from the luminance distribution outside the section.

  As shown in FIG. 5 (left), the lower limb fluoroscopic image (Height: 2140 pixels, Width: 1760 pixels) to the lower limb analysis image (HeightJ: 1935 pixels, WidthI: 351 pixels) Will be obtained.

j=1,…,M i=1,…,N
ここでIBmaxとIBminはそれぞれ透視画像の最大輝度値と最小輝度値である。ここで求める膝間隙ラインは間隙の平均的なj軸座標Jrgkとする。Jrgkの抽出法は次の通りである。
(2) 式(2)により吸収画像の輝度分布IBrt1(j,i)からj軸方向に１階差分の絶対値を求め、かつi軸方向に平均化を行い、輝度差分分布DBrt(j)を求める。

(3) DBrt(j)のピークのj軸座標が求める膝間隙ラインとなる。 [C2: Knee gap line extraction process]
Next, the process C2 for extracting the knee gap line will be described. FIG. 9 is a flowchart showing a knee gap line extraction algorithm. In process C2, the knee gap line is extracted based on the fluoroscopic image cut out in C1.
(1) The luminance inversion conversion is performed on the fluoroscopic image (luminance value IBrt0 (j, i)) by the equation (1). The converted image (luminance value IBrt1 (j, i)) is called an absorption image. By this conversion, a bone portion having a high X-ray absorption rate has high luminance.

j = 1,…, M i = 1,…, N
Here, IBmax and IBmin are the maximum luminance value and the minimum luminance value of the fluoroscopic image, respectively. The knee gap line obtained here is an average j-axis coordinate Jrgk of the gap. The extraction method of Jrgk is as follows.
(2) The absolute value of the first-order difference is obtained in the j-axis direction from the luminance distribution IBrt1 (j, i) of the absorption image by equation (2), and averaged in the i-axis direction to obtain the luminance difference distribution DBrt (j) Ask for.

(3) DBrt (j) is the knee gap line determined by the j-axis coordinate of the peak.

(3)
透視画像の輝度範囲が(IBmin,IBmax)になるようにIBrt2(j,i)を線形補間し、γピクセル（例えばγ=3ピクセル）のスケールの移動平均による平滑化を行い、j軸方向の圧縮画像を得る。
図10に｜勾配｜分布、DBrt(j)に対してDRPressYフィルター有無での効果を示した。この図から、次のようなDRPressYフィルターの効果が示された。このフィルターは平均値を含む比較的大きなスケールの変動を除去するために、得られた勾配分布は大きさと形において、より安定した分布となり、ピーク位置の抽出が容易となることである。 Next, in order to ensure the extraction of the knee gap line, the dynamic range compression (referred to as DR compression, here referred to as the DRPressY filter ) in the j-axis direction is performed on the absorption image IBrt1 (j, i ) as preprocessing. Introduced. The DRPressY filter process is as follows.
(1)
Smoothing is performed by moving average of β pixels (for example, β = 50 pixels, about 2.5 cm) in the j-axis direction to create a blurred image Boke (j, i).
(2)
Using the blurred image Boke (j, i), a brightness variation image IBrt2 (j, i) according to Equation (3) is created.

(3)
IBrt2 (j, i) is linearly interpolated so that the brightness range of the fluoroscopic image is (IBmin, IBmax), smoothed by a moving average of the scale of γ pixels (for example, γ = 3 pixels), and in the j-axis direction Get a compressed image.
FIG. 10 shows the effect of the presence or absence of the DRPressY filter on the | gradient | distribution and DBrt (j). From this figure, the effect of the following DRPressY filter was shown. Since this filter removes a relatively large scale variation including an average value, the obtained gradient distribution is more stable in size and shape, and the peak position can be easily extracted.

  The bone axis analysis image (j axis: HeightJJ = 801 pixel i axis: WidthI = 351 pixel) shown in FIG. 11 can be cut out by the process of C2 described above. The width direction of this image is the same as C1, but the height direction is a range of 20 cm in the vertical direction with the knee gap Jrgk as the center. This range in the height direction is a range determined because the bone axis analysis range normally covers a range of 10 cm to 15 cm vertically from the knee gap.

[C3: Bone boundary point extraction process]
Next, the process C3 (bone boundary extraction method) for extracting bone boundary points will be described. FIG. 12 is a flowchart showing an algorithm for extracting the bone boundary point and analyzing the bone axis. Process C3
(1) From the absorption image obtained in the process (b) (the left diagram in FIG. 14), an enhanced image that enhances the luminance level of the bone tissue is derived using dynamic range compression image processing (C31).
(2) A cluster of distributions (hereinafter referred to as islands) having a luminance level higher than a predetermined luminance level with respect to the luminance distribution in the i-axis direction, where the width direction of the enhanced image is the i-axis and the height direction is the j-axis. ) Is detected.
(3) If two islands are detected, select the two islands. If there are 3 or more islands detected, determine whether the femur or tibia based on the knee gap line,
(3-1) For the femur, select two islands from the higher island.
(3-2) In the case of the tibia of the left limb, select the two islands on the left side with priority.
(3-3) In the case of the tibia of the right limb, select the two islands on the right side preferentially.
(4) Using the edge image from the two islands selected in the above (3), a position where the brightness value of the island becomes a peak is obtained, and the position is extracted as a bone boundary point (C32: transverse brightness distribution method).

得られた圧縮画像は図13との比較から、骨組織を反映した画像と言える。すなわち、緻密質が最も高い輝度値で、髄腔が低い輝度値の画像である。
（３）強調画像IBrt3(j,i)（図15の左図）の作成
上記の圧縮画像は図14の右図から分かるように、解析する大腿骨と脛骨以外にも輝度の高い生体組織があるために、骨境界点の抽出に対して不十分と言える。よって、解析する骨部を増幅させ、反対にそれ以外では減衰させる強調係数を各j軸座標に設定することにした。このために式(4)〜式(6)に示す強調係数fact(j,i)を導入する。
fact(j,i)=Boke(j,i)/IBrtLVL(j) ・・・式(4)
IBrt3(j,i)= fact(j,i)*IBrt2(j,i) ・・・式(5)
IBrtLVL(j)=Mean+α×Std ・・・式(6)
強調係数fact(j,i)は、ボケ画像Boke(j,i)に対して、各j軸座標の平均的輝度レベルIBrtLVL(j)で規格化することで求めた。ここでIBrtLVL(j)はボケ画像の、各j軸座標に対して、i軸方向の全区間（1, WidthI）の輝度分布の平均値をMean、その標準偏差をStd、αを予め設定された定数で式(6)で与えられる。
式(4)より、強調係数fact(j,i)は、高い輝度骨部に対して１以上で増幅され、逆に低い輝度の生体組織では１以下で減衰される。従って、強調画像は式(5)で得られ、圧縮画像に比べ、解析する大腿骨と脛骨のみがより高い輝度の画像となる（図15の左図）。このことは代表的な横断輝度分布（図16と図17）からも明白に示される。 [C31: Dynamic range compression weighted image processing]
Next, dynamic range compression enhanced image processing (C31) will be described. The desired bone axis is the bone axis of the femoral and tibia shafts in the bone axis analysis range up and down from the knee gap. As shown in FIG. 13, the diaphysis to be analyzed has a cavity called a medullary cavity at the center, and has a hard dense mass (bone) as the trunk wall. The developed dynamic range compression weighted image processing is image processing that reflects such bone tissue. The specific analysis procedure is as follows. The images shown below are normalized so that the absorption image and the luminance range are the same.
(1) Creation of blurred image Boke (j, i) (middle figure of Fig. 14)
Smoothing is performed by moving average of β pixels (β: a preset constant) in both the j-axis and i-axis directions, and a blurred image Boke (j, i) is created.
(2) Expression using the generated blurred image Boke (j, i) and the absorption image IBrt1 (j, i) of the compressed image IBrt2 (j, i) (right diagram in FIG. 14) in the j-axis and i-axis directions The compressed image IBrt2 (j, i) is created by (3).

From the comparison with FIG. 13, the obtained compressed image can be said to be an image reflecting bone tissue. That is, it is an image having a luminance value with the highest density and a luminance value with a low medullary cavity.
(3) Creation of emphasized image IBrt3 (j, i) (left figure in FIG. 15) As can be seen from the right figure in FIG. 14, the above compressed image shows a living tissue with high brightness in addition to the femur and tibia to be analyzed. Therefore, it can be said that it is insufficient for extracting the bone boundary point. Therefore, the emphasis coefficient that amplifies the bone part to be analyzed and vice versa is set for each j-axis coordinate. For this purpose, the enhancement coefficient fact (j, i) shown in the equations (4) to (6) is introduced.
fact (j, i) = Boke (j, i) / IBrtLVL (j) (4)
IBrt3 (j, i) = fact (j, i) * IBrt2 (j, i) (5)
IBrtLVL (j) = Mean + α × Std (6)
The enhancement coefficient fact (j, i) was obtained by normalizing the blurred image Boke (j, i) with the average luminance level IBrtLVL (j) of each j-axis coordinate. Here, IBrtLVL (j) is preset for each j-axis coordinate of the blurred image, with the mean value of the luminance distribution of all sections in the i-axis direction (1, WidthI) being Mean, its standard deviation being Std, and α. Is given by equation (6).
From equation (4), the enhancement factor fact (j, i) is amplified by 1 or more for high-luminance bones, and is attenuated by 1 or less for biological tissues with low luminance. Therefore, the enhanced image is obtained by Expression (5), and only the femur and tibia to be analyzed are images with higher luminance than the compressed image (the left diagram in FIG. 15). This is also clearly shown from typical transverse luminance distributions (FIGS. 16 and 17).

The characteristics of the enhanced image are also examined from the luminance frequency distribution. FIG. 18 shows luminance frequency distributions of the absorption image, the compressed image, and the enhanced image. This figure shows the following.
I. The DR compression process used concentrates the frequency of luminance near the luminance level IBrtLVL (j).
II. The enhancement process further promotes the concentration of the luminance of the bone by multiplying the enhancement coefficient created by the blurred image standardized at the above luminance level .
The above II is clearly shown in the next peak-enhanced image (middle of FIG. 15). That is, the peak-enhanced image is a binary image created with the luminance values of only the peak frequency in FIG. 18, and is indicated by the clear presence of the bone part. The peak-enhanced image is created by converting the brightness of the enhanced image in FIG. 18 so that the peak luminance value range is 440 to 450, the luminance value below 440 is 440, and the luminance value above 450 is 450. It is.

[C32: Transverse luminance distribution method]
Next, the transverse luminance distribution method (C32) will be described. This method obtains a bone boundary point using an edge image based on the transverse luminance distribution of the emphasized image in the bone axis analysis range. The edge image is created by applying the Mexican hat filter (non-patent document (2)) represented by the 9 × 9 matrix shown in FIG. 19 to the emphasized image , and the fact (j, i) in equation (4) (Equation (5)) was performed (the right figure of FIG. 15). Then, noise mixed in the bone boundary point is detected and corrected by the regression line correction of the bone boundary point.

The upper diagram of FIG. 20 shows the transverse luminance distribution of the emphasized image and the edge image at an arbitrary j-axis coordinate (j = 160 in the figure) in the bone axis analysis range in the femur. The transverse luminance distribution is a variable luminance distribution from the reference luminance level Level (j). Level (j) is calculated by equation (6). The analysis procedure is as follows.
1) In the enhanced image, a cluster of distributions with positive luminance values (hereinafter called islands) is detected, and the islands are parameterized by height and position (island peak luminance points). Select an island. These islands correspond to the fineness of Figure 13. According to FIG. 20, island information (i-axis coordinates A _k , B _k , C _k : k = 1, 2) is determined. For the central medullary cavity, subscript 1 indicates the left side and subscript 2 indicates the right side.
2) The edge image gives the island shown in the enhanced image, that is, the dense boundary point. From FIG. 20, the peak point D _{k corresponds} to the outer bone boundary point, and the zero point E _k corresponds to the bone boundary point.
3) The tibia is different from the femur as shown in the lower figure of FIG. For this reason, only the island of the tibia is selected using the position information of the island based on the information of the right limb or the left limb. Specifically, in the emphasized image, when there are three or more islands, the rib on the right limb is always on the left side of the tibia, so the island on the right side is preferentially selected. In the case of the left limb, the rib is on the right side, so the left island should be selected with priority.

Next, the regression line correction of the bone boundary point will be described. The bone axis has been regarded as a straight line in the range of bone axis analysis, and the bone boundary line has been regarded as a straight line as well. Therefore, the bone boundary line is accurately determined within the range by the straight line obtained by the regression line approximation method in the longitudinal direction (j-axis direction) of the bone boundary points (E1 and E2 in FIG. 20). Using this property, we detected and corrected noise mixed in the bone boundary point. The analysis procedure is as follows.
1) Perform regression line approximation on bone boundary points, remove data that deviates more than the standard deviation each time, and perform regression line approximation until it disappears.
2) Correct the removed data using the convergent regression line.

From the above, the transverse luminance distribution method mainly consists of the island analysis method and the regression line method. These methods are superior to the conventional interactive binarization method and connection method in the following points, and finally enable the automatic analysis of the bone axis without requiring the intervention of a doctor.
1) The analysis by island uses the information of brightness value (island height) more than the binarization method.
2) The regression line method is more efficient than the connection method by detecting and correcting noise using the fact that the bone axis is linearly approximated.

[C4 : Bone axis analysis from bone boundary points ]
Next, the process C4 for analyzing the bone axis from the bone boundary point will be described. FIG. 12 is a flowchart showing an algorithm for extracting the bone boundary point and analyzing the bone axis. In the bone axis analysis, the bone axes of the femur and tibia are obtained from the left and right bone boundary points of the shaft. In this analysis, the following two analyzes are important.
(1) Setting the exact range for bone axis analysis
(2) How to find the bone axis from the left and right bone boundary points

Regarding (1) above, in the X-ray photograph, when there is an angle between the femur and the tibia, this is an accurate determination method in the range of 10 cm to 15 cm above and below the knee gap. Since the bone axis is determined by the left and right bone boundary lines, the bone axis analysis range is the bone boundary line analysis range. Therefore, it is possible to determine the analysis range by the midline by the midpoint obtained from the bone boundary point. The analysis procedure is shown in FIG. 21 for the case of the femur.
1) The range for obtaining the bone boundary point was set to be sufficiently wider than the analysis range of the bone boundary line. Actually, the j-axis coordinates (1 to Jmask1) were used. Jmask1 is a j-axis coordinate whose distance from the knee gap is 8 cm.
2) First, the left and right bone boundary points in the temporary analysis range (10 cm to 15 cm from the knee gap line (J15 to J10)) and the midpoint between the left and right bone boundary points, the temporary left and right bone boundary lines and the middle line Ask.
3) Intersection C between the midline and the knee gap line. On the middle line, find point B where BC = 10 cm and point A where AC = 15 cm.
4) A straight line perpendicular to the center line is obtained at points A and B, and intersection points (A ₁ , A ₂ , B ₁ , B ₂ ) with the temporary left and right bone boundary lines obtained in ₂ ) are calculated. From these, the analysis range of the accurate bone boundary line is determined. That is, the range is (A _{2 to} B ₂ ) for the right bone boundary line and the range (A _{1 to} B ₁ ) for the left bone boundary line. Therefore, the bone axis analysis range of the present invention is defined by the range from point A to point B on the middle line (range of j-axis coordinates J15m to J10m).
Although the above has described the range of femoral bone axis analysis, it can also be obtained for the tibia.

上記の２つの直線を左右の骨境界線とすれば、式(7)より骨軸の直線が得られることになる。
解析された膝間隙と、大腿骨と脛骨の骨軸をグラフ（図22）と画像（図23）で示した。
これらの２つの骨軸の直線情報(y=b₃x+a₃とy=b₄x+a₄とする)から、大腿脛骨外側角(FTA)は容易に次式から計算される。

For (2), the bone boundary line is calculated by the regression line method from the bone boundary point in the above accurate analysis range. The calculation method for obtaining the bone axis from the left and right bone boundary lines uses the following mathematical formula. A straight line that is equidistant from two straight lines (y = b ₁ x + a ₁ and y = b ₂ x + a ₂ ) is given by the following equation.

If the above two straight lines are the left and right bone boundary lines, the straight line of the bone axis can be obtained from the equation (7).
The knee gap analyzed and the bone axes of the femur and tibia are shown in a graph (FIG. 22) and an image (FIG. 23).
From the linear information of these two bone axes (assuming y = b ₃ x + a ₃ and y = b ₄ x + a ₄ ), the femoral tibia lateral angle (FTA) is easily calculated from the following equation.

  Next, a program executed on the computer in order to implement the above-described bone axis automatic extraction method will be described. If this program is executed by a computer, the same action and effect can be realized according to basically the same principle as in the bone axis automatic extraction method described above.

  FIG. 24 shows an embodiment in which the bone axis automatic extraction program of the present invention is incorporated in a system for simulating a preoperative plan for total replacement artificial knee joint surgery using a computer (Non-patent Document 3). From this embodiment, the knee gap line and the femoral / tibia bone axes are automatically calculated, and the template is efficiently arranged on the screen, making it easy to install the implant. Furthermore, it is possible to adjust the bone reference point and template position by manual operation to reflect the doctor's rich experience.

  Also from this example, the extraction of the knee gap and the bone axes of the femur and tibia is automatically analyzed on a computer to reduce the burden on the preoperative planning of the doctor and the surgeon who has relied on the skill so far It can be seen that there is an effect of maintaining the level of diagnosis such as preoperative planning and osteoarthritis of the knee by more than a certain level.

It is a figure which shows the visual and manual work method of the conventional method. It is a flowchart figure which shows the whole algorithm based on this invention. It is a flowchart figure which shows the cutting-out algorithm of the leg analysis image which concerns on this invention. It is a figure which shows the coordinate system of the image which concerns on this invention. It is a figure which shows the leg image (right figure, j-axis 2140 pixel i-axis 1760 pixel) and leg analysis image (left figure, j-axis 1935 pixel i-axis 351 pixel) after image cutting according to the present invention. It is a figure which shows the area (ILine1, ILine2) of the image cut-out of the i-axis direction which concerns on this invention. The i-axis direction width 1760 pixels is reduced to the interval (ILine2-ILine1) = 351 pixels. It is a figure which shows the area (Jpst, Jped) of the image cut-out of the j-axis direction which concerns on this invention. The j-axis height of 2140 pixels is reduced to the number of sections (Jped-Jpst) = 1935 pixels. It is a figure which shows the brightness | luminance fluctuation distribution Javbd [j] in the area (JLine1, JLine2) based on this invention. It is a flowchart figure which shows the extraction algorithm of the knee gap line which concerns on this invention. It is a figure which shows the gradient distribution of the j-axis direction which calculates | requires the knee gap (Jrgk) based on this invention. (With no DRPressY, the level is adjusted to match the peak value) It is the figure which extracted the knee gap line (Jrgk) based on this invention. It is a flowchart figure which shows the analysis algorithm of the bone boundary point and bone axis based on this invention. It is a figure which shows the external appearance (left figure) and longitudinal cross-section (right figure) of a long bone. It is a figure which shows the absorption image (left figure), blurred image (middle figure), and compressed image (right figure) which concern on this invention. It is a figure which shows the emphasized image (left figure), peak emphasized image (middle figure), and edge image (right figure) which concern on this invention. It is a figure which shows the comparison of the transverse luminance distribution of the original image which concerns on this invention, a blurred image, a compression image, and an emphasis image. It is a figure which shows the comparison of the original image which concerns on this invention, a blurred image, a compression image, and an emphasized image. It is a figure which shows the frequency distribution of the brightness | luminance of the absorption image which concerns on this invention, a compression image, and an emphasis image. It is a figure which shows the Mexican hat filter (9x9) used when creating an edge image from the emphasized image which concerns on this invention. It is a figure which shows the crossing luminance distribution of the emphasis image and edge image which concern on this invention. It is a figure explaining how to obtain the bone axis analysis range (in the case of a femur) according to the present invention. It is a figure which shows the bone axis analysis range and bone axis of the femur and tibia which concern on this invention. It is a figure which shows the leg leg image of the knee gap | interval which concerns on this invention, the femur, and the tibia. It is a figure which shows the Example of the preoperative plan assistance system of the total replacement type artificial knee joint operation which concerns on this invention.

Claims (14)

A method of automatically analyzing a lower limb fluoroscopic image on a computer to automatically extract the femoral and tibia bone axes of the knee joint,
(A) a process of cutting out an image region of a living tissue from a lower limb fluoroscopic image;
(B) a process of deriving an image (hereinafter referred to as an absorption image) obtained by luminance inversion conversion from the fluoroscopic image cut out in the process (a), and extracting a knee gap line based on the luminance distribution of the absorption image;
(C) The absorption image obtained in the process (b) is converted into an image in which bone tissue is enhanced (hereinafter referred to as an enhanced image), and a bone boundary point is obtained based on the luminance distribution of the enhanced image and the knee gap line. Process,
(D) a process of setting a bone axis analysis range based on the bone boundary point obtained in the process (c) and the knee gap line, and deriving the bone axis from the bone boundary point of the bone axis analysis range;
The bone axis of the femur and tibia of the knee joint is automatically extracted by sequentially executing processes (a) to (d), and the femoral tibia lateral angle is automatically calculated. Extraction method.   In the process (a), with the width direction of the lower limb fluoroscopic image as the i axis and the height direction as the j axis, using the luminance distribution characteristics in the i axis direction and the j axis direction and the limb information on the left limb or the right limb, 2. The method for automatically extracting a bone axis according to claim 1, wherein a minimum region obtained by removing noise information other than a biological tissue image is cut out from the lower limb fluoroscopic image. In the cutout according to claim 2, the image cutout procedure in the i-axis direction is:
(1) The average luminance value IavbM (i) of the luminance distribution is obtained in the j-axis direction at each i-axis coordinate.
(2) In the IavbM (i) distribution, the i-axis coordinates where the luminance value changes from the both ends of the i-axis to a certain luminance value or more are IRngA1 and IRngA2, respectively, and the width (IRng A 2 −IRng A 1) / δ from these points. IRngB1 and IRngB2 are obtained as internal points shifted inward by the length of (δ: preset constant).
(3) The i-axis coordinate point IMinIti of the minimum luminance value in the section (IRngB1, IRngB2) is obtained.
(4) In the case of the left limb, the peak point of the maximum luminance value in the section (IRngB1, IMinIti) is ILine1, the peak point closer to IMinIti in the section (IMinIti, IRngB2) is ILine2, and ILine1 and ILine2 are i-axis coordinate A cut point is used. In the case of the right limb, the peak point of the maximum luminance value in the interval (IMinIti, IRngB2) is ILine2, the peak point closer to IMinIti in the interval (IRngB1, IMinIti) is ILine1, and ILine1 and ILine2 are the cut-out points of the i-axis coordinates. To do.
The method for automatically extracting a bone axis according to claim 2. In the cutout according to claim 2, the image cutout procedure in the j-axis direction is:
(1) The average luminance value Javb [j] of the luminance distribution is obtained in the i-axis direction at each j-axis coordinate.
(2) In the Javb [j] distribution, coordinates whose luminance values have changed from the both ends of the j-axis to a predetermined set value or more are JLine1 and JLine2, respectively, and these coordinates are cut-out points in the j-axis direction.
The method for automatically extracting a bone axis according to claim 2.   In the image cutout procedure in the j-axis direction according to claim 4, a distribution JavbM [j] obtained by smoothing the distribution of Javb [j] in the j-axis direction in the cutout section (JLine1, JLine2) is obtained, and the fluctuation distribution JavbD [ j] = Javb [j] −JavbM [j] is used to identify the section of the X-ray line (hereinafter referred to as line noise), and the luminance distribution in the line noise section is linearly interpolated from the luminance distribution outside the section. 5. The method for automatically extracting a bone axis according to claim 4, wherein the line noise is eliminated by performing correction. In process (b), the width direction of the image is the i-axis and the height direction is the j-axis.
(1) The luminance distribution IBrt1 (j, i) of the absorption image obtained by performing luminance inversion conversion on the fluoroscopic image IBrt0 (j, i) cut out in the process (a) is obtained by the equation (1).

Here, IBmax and IBmin are the maximum luminance value and the minimum luminance value of the fluoroscopic image, respectively.
(2) Using the luminance distribution IBrt1 (j, i) of the absorption image, the luminance difference distribution DBrt (j) is obtained by Equation (2).

(3) The j-axis coordinate of the peak of DBrt (j) is extracted as a knee gap line.
The method for automatically extracting a bone axis according to claim 1. The knee gap line extraction according to claim 6, wherein as a pre-processing, dynamic range compression (hereinafter referred to as DR compression) in the j-axis direction is performed to derive a luminance distribution in which a luminance variation amount is amplified, and DR compression The filtering process is
(1) Smoothing is performed by moving average of β pixels (β: a preset constant) only in the j-axis direction to create a blurred image Boke (j, i).
(2) Using the blurred image Boke (j, i) and the absorption image IBrt1 (j, i), create a luminance variation image IBrt2 (j, i) according to Equation (3).

(3) IBrt2 (j, i) is linearly interpolated so that the brightness range of the fluoroscopic image cut out in process (a) becomes (IBmin, IBmax), and the scale of γ pixel (γ: preset constant) To obtain a compressed image in the j-axis direction.
The bone axis automatic extraction method according to claim 6. In process (c)
(1)
An enhanced image that enhances the luminance level of the bone tissue is derived from the absorption image obtained in the process (b) by using dynamic range compression image processing.
(2)
Detects a cluster of distributions (hereinafter referred to as islands) having a luminance level higher than a predetermined luminance level with respect to the luminance distribution in the i-axis direction, where the width direction of the enhanced image is the i-axis and the height direction is the j-axis. To do.
(3)
When there are two islands detected in (2), the two islands are selected. If there are three or more islands detected in (2), determine whether the femur or tibia based on the knee gap line;
(3-1)
In the case of the femur, two islands are selected from the higher island height.
(3-2)
In the case of the left limb tibia, the two islands on the left are preferentially selected.
(3-3)
In the case of the right limb tibia, the two islands on the right side are preferentially selected.
(Four)
From the two islands selected in the above (3), a position where the brightness value of the island becomes a peak is obtained using the edge image, and the position is extracted as a bone boundary point.
The method for automatically extracting a bone axis according to claim 1. The derivation of the enhanced image according to claim 8 is as follows .
(1) on both the direction of j axis and i axis beta pixels: performs smoothing by moving average of (beta preset constant), creating a blurred image Boke (j, i).
(2) A compressed image IBrt2 (j, i) is created by Expression (3) using the blurred image Boke (j, i) and the absorption image IBrt1 (j, i) .

(3) From the compressed image IBrt2 (j, i), the enhancement coefficient fact (j, i) shown in the equations (4) to (6) is introduced to create the enhanced image IBrt3 (j, i). Here, the average luminance level IBrtLVL (j) is the average value of the luminance distribution of all sections in the i-axis direction and the standard deviation value thereof for each j-axis coordinate in the blurred image Boke (j, i). Std and α are obtained by preset constants.
fact (j, i) = Boke (j, i) / IBrtLVL (j) (4)
IBrt3 (j, i) = fact (j, i) * IBrt2 (j, i) (5)
IBrtLVL (j) = Mean + α × Std (6)
The method for automatically extracting a bone axis according to claim 8.   The bone boundary point extraction according to claim 8, wherein an edge image is derived by applying a Mexican hat filter to the enhanced image, and the bone boundary point is determined based on a luminance distribution of the enhanced image using the edge image. The bone axis automatic extraction method according to claim 8, wherein the bone axis automatic extraction method is obtained. The extraction of the bone boundary point according to claim 8, wherein the detection and correction of noise mixed in the bone boundary point is performed using a regression linear approximation method in the j-axis direction in the height direction from the bone boundary point. The bone axis automatic extraction method according to claim 8. The procedure for setting the bone axis analysis range in process (d) is as follows:
(1)
A j-axis coordinate range (1 to Jmask1) in the height direction of the enhanced image IBrt3 (j, i) for obtaining the bone boundary point is determined. (Jmask1: j-axis coordinate whose distance from the knee gap is 8cm)
(2)
Temporary left and right bone boundary lines and midlines are determined by the regression line method from the left and right bone boundary points and their midpoints in the temporary analysis range (10 cm to 15 cm from the knee gap line (J15 to J10)).
(3)
Intersection C between the midline and the knee gap line. On the middle line, find point B where BC = 10 cm and point A where AC = 15 cm.
(Four)
Find a straight line perpendicular to the midline at points A and B, calculate the intersection (A1, A2, B1, B2) with the temporary left and right bone boundary lines obtained in (2) above, and the right bone boundary line The range (A2 to B2) for the left bone boundary line and the range (A1 to B1) for the left bone boundary line are the bone boundary analysis range.
The method for automatically extracting a bone axis according to claim 1.   In the derivation procedure of the bone axis in the process (d), the left and right bone boundary lines are derived from the bone boundary points in the bone axis analysis range by using the regression line method, and straight lines that are equidistant from the left and right bone boundary lines. The bone axis automatic extraction method according to claim 1, wherein: is derived as a straight line of the bone axis.   A bone axis automatic extraction program that causes a computer to execute the bone axis automatic extraction method according to any one of claims 1 to 13.

Images