JP4639338B2 - Rib cage boundary detection method from chest X-ray image - Google Patents
Rib cage boundary detection method from chest X-ray image Download PDFInfo
- Publication number
- JP4639338B2 JP4639338B2 JP2007081989A JP2007081989A JP4639338B2 JP 4639338 B2 JP4639338 B2 JP 4639338B2 JP 2007081989 A JP2007081989 A JP 2007081989A JP 2007081989 A JP2007081989 A JP 2007081989A JP 4639338 B2 JP4639338 B2 JP 4639338B2
- Authority
- JP
- Japan
- Prior art keywords
- rib cage
- point sequence
- image
- boundary
- cage boundary
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 210000000038 chest Anatomy 0.000 title claims description 188
- 238000001514 detection method Methods 0.000 title claims description 36
- 238000011976 chest X-ray Methods 0.000 title claims description 20
- 210000004072 lung Anatomy 0.000 claims description 79
- 238000000034 method Methods 0.000 claims description 56
- 238000007796 conventional method Methods 0.000 description 23
- 230000002123 temporal effect Effects 0.000 description 10
- 206010058467 Lung neoplasm malignant Diseases 0.000 description 4
- 238000004195 computer-aided diagnosis Methods 0.000 description 4
- 201000005202 lung cancer Diseases 0.000 description 4
- 208000020816 lung neoplasm Diseases 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 3
- 210000003484 anatomy Anatomy 0.000 description 2
- 238000003745 diagnosis Methods 0.000 description 2
- 230000002708 enhancing effect Effects 0.000 description 2
- 210000001370 mediastinum Anatomy 0.000 description 2
- 238000011410 subtraction method Methods 0.000 description 2
- 239000002131 composite material Substances 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000003601 intercostal effect Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
Images
Landscapes
- Apparatus For Radiation Diagnosis (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Description
本発明は、胸部Ｘ線像からリブケイジ（ribcage）境界を精度よく自動検出するための画像処理方法に関するものである。 The present invention relates to an image processing method for automatically detecting a ribcage boundary from a chest X-ray image with high accuracy.
胸部Ｘ線像からリブケイジ境界を自動検出する方法は、X．W．Xu and K．Doi, Image feature analysis for computer-aided diagnosis : Accurate determination of ribcage boundary in chest radiographs, Medical Physics, Vol．22, No．5, pp．617-626, 1995や特開平7-37074号公報に記載されているように、コンピュータ支援診断の分野で広く利用されている。
従来のリブケイジ境界検出法（X．W．Xu and K．Doi, Image feature analysis for computer-aided diagnosis : Accurate determination of ribcage boundary in chest radiographs, Medical Physics, Vol．22, No．5, pp．617-626, 1995）は、画像の垂直プロファイル、水平プロファイルを用いて、リブケイジ境界候補点を検出する。右側、左側リブケイジ境界は、行方向に急激に変化するのに対して、上肺リブケイジ境界は、列方向に急激に変化する。そこで従来法は、リブケイジ境界を上肺、右側、左側リブケイジ境界の三つの部分に分け、右側および左側リブケイジ境界候補点は水平プロファイルを用いて検出し、上肺リブケイジ境界候補点は垂直プロファイルを用いて検出する。
また、従来法は、上肺、右側、左側リブケイジ境界ごとに境界候補点を求めるとき、まず開始点を求め、前に検出された境界候補点の近傍で次の境界候補点を求めるという処理を繰り返して、境界候補点をたどって行く。そして、境界候補点を検出した後、上肺、右側、左側リブケイジ境界ごとに、カーブフィッティングを適用して、これの境界を曲線で与える。
従来法の問題点の一つは、この方法がリブケイジ境界を上肺、右側、左側リブケイジ境界の三つの部分に分けて検出するため、三つの境界を求めた後、これらを一つの連続する曲線に結合する必要があることである。
従来法の２番目の問題点は、従来法は上肺リブケイジ境界の検出に失敗することが多いことである。従来法は、上肺リブケイジ境界を求めるために、まず、肺野上端線を求め、肺野上端線上の一つの点を境界候補点列の開始点とする。しかし、従来法は肺野上端線を求めるために、画像上部の垂直プロファイルの極大点という局所的情報のみを用いるので、肺野上端線を正しく求めることができないことが多い。肺野上端線が正しく求まらないと、境界候補点列の開始点が求まらないので、上肺リブケイジ境界は求まらない。また、従来法は、前に検出された境界候補点の近傍で、垂直プロファイルの２次微分の極小点を求め、値が小さいほうから二つの極小点のうちの一方を次の境界候補点とするが、上肺部は複雑な合成像となっているため、開始点が正しく求まった場合でも、従来法が用いるような方法では上肺リブケイジ境界候補点を正しくたどれないことも多い。
従来法の３番目の問題点は、従来法ではリブケイジ境界の検出精度が、境界候補点列の開始点の検出精度に大きく依存するにもかかわらず、従来法の開始点の検出精度は低いことである。従来法は、上肺リブケイジ境界と比較すると、右側、左側リブケイジ境界を画像によらず比較的安定して検出できる。しかし、従来法は、肺の中央（行方向に関する中央）で境界候補点列の開始点を求めた後、上下の方向に境界候補点をたどることによって右側、左側リブケイジ境界を求めるが、開始点を求めるために、水平プロファイルの２次微分の最小点という局所的情報のみを用いるため、開始点の検出に失敗することが多い。上肺リブケイジ境界検出の場合と同様、開始点の検出に失敗すると、従来法は右側、左側リブケイジ境界を正しく検出できない。
特開2002-177249号公報には、Xuらの方法を改良したリブケイジ境界検出法が示されているが、この方法の基本的アルゴリズムはXuらの方法と同じであり、この方法もXuらの方法と同様な問題点をもつ。
X-W. Xu and K. Doi, Image feature analysis for computer-aided diagnosis: Accurate determination of ribcage boundary in chest radiographs, Medical Physics, Vol. 22, No. 5, pp. 617- 626, 1995) detect rib cage boundary candidate points using the vertical profile and horizontal profile of the image. The right and left rib cage boundaries change abruptly in the row direction, while the upper lung rib cage boundary changes abruptly in the column direction. Therefore, the conventional method divides the rib cage boundary into three parts: the upper lung, the right side, and the left rib cage boundary, detects the right and left rib cage boundary candidate points using the horizontal profile, and uses the vertical profile for the upper lung rib cage boundary candidate points. To detect.
Further, in the conventional method, when obtaining the boundary candidate point for each upper lung, right side, and left side rib cage boundary, first, the start point is obtained, and the next boundary candidate point is obtained in the vicinity of the previously detected boundary candidate point. Repeat and follow the boundary candidate points. Then, after detecting the boundary candidate points, curve fitting is applied to each of the upper lung, right side, and left side rib cage boundaries, and the boundary is given by a curve.
One of the problems with the conventional method is that the rib cage boundary is detected by dividing it into three parts: the upper lung, the right side, and the left rib cage boundary. It is necessary to combine with.
The second problem with the conventional method is that the conventional method often fails to detect the upper lung rib cage boundary. In the conventional method, in order to obtain the upper lung rib cage boundary, first, the lung field upper end line is obtained, and one point on the lung field upper end line is set as the start point of the boundary candidate point sequence. However, since the conventional method uses only local information such as the maximum point of the vertical profile at the top of the image in order to obtain the lung field top line, the lung field top line cannot often be obtained correctly. If the upper end line of the lung field is not obtained correctly, the starting point of the boundary candidate point sequence is not obtained, and therefore the upper lung rib cage boundary is not obtained. In addition, the conventional method obtains the minimum point of the second derivative of the vertical profile in the vicinity of the previously detected boundary candidate point, and sets one of the two minimum points from the smaller value as the next boundary candidate point. However, since the upper lung portion is a complex composite image, even when the start point is obtained correctly, the upper lung rib cage boundary candidate points are often not correctly traced by the method used by the conventional method.
The third problem of the conventional method is that the detection accuracy of the rib cage boundary in the conventional method is largely dependent on the detection accuracy of the starting point of the boundary candidate point sequence, but the detection accuracy of the starting point of the conventional method is low. It is. Compared with the upper lung rib cage boundary, the conventional method can detect the right and left rib cage boundaries relatively stably regardless of the image. However, in the conventional method, the start point of the boundary candidate point sequence is obtained at the center of the lung (the center in the row direction), and then the right and left rib cage boundaries are obtained by tracing the boundary candidate points in the vertical direction. Therefore, since only local information, ie, the minimum point of the second derivative of the horizontal profile is used, the detection of the start point often fails. As in the case of detecting the upper lung rib cage boundary, if the detection of the starting point fails, the conventional method cannot correctly detect the right and left rib cage boundaries.
Japanese Patent Laid-Open No. 2002-177249 discloses a rib cage boundary detection method improved from the method of Xu et al., But the basic algorithm of this method is the same as that of Xu et al. It has the same problems as the method.
前述したように、従来のリブケイジ境界検出法は、上肺部のリブケイジ境界の検出に失敗することが多い。それ故、肺野上端線から肺野下端線までのすべての行に対して、リブケイジ境界の位置を正しく求めることができるリブケイジ境界検出法が望まれる。
また、上肺部以外のリブケイジ境界についても、従来法のリブケイジ境界の検出精度は、境界候補点列の開始点の検出精度に大きく依存するため、あまり高くない。
従来法が上肺部のリブケイジ境界の検出に失敗したり、境界候補点列の開始点の検出に失敗する主な理由は、従来法がノイズに弱いことによる。従来法は、画像の垂直プロファイルの極大点、垂直プロファイルの２次微分の極小点、水平プロファイルの最小点、水平プロファイルの２次微分の極小点という局所的情報を用いて境界候補点を検出するため、ノイズに対して弱い。それ故、画像の品質に依らず安定してリブケイジ境界を検出できる方法が望まれる。
そこで、本発明では、画像の品質に依らず安定して肺野上端線から肺野下端線までのすべての行に対して、リブケイジ境界の位置を正しく検出できるリブケイジ検出方法を提供する。
As described above, the conventional rib cage boundary detection method often fails to detect the rib cage boundary of the upper lung portion. Therefore, a rib cage boundary detection method that can correctly determine the position of the rib cage boundary for all the rows from the lung field upper end line to the lung field lower end line is desired.
Also, for rib cage boundaries other than the upper lung part, the detection accuracy of the rib cage boundary according to the conventional method is not so high because it depends greatly on the detection accuracy of the start point of the boundary candidate point sequence.
The main reason why the conventional method fails to detect the rib cage boundary of the upper lung part or fails to detect the start point of the boundary candidate point sequence is that the conventional method is vulnerable to noise. In the conventional method, boundary candidate points are detected using local information such as the maximum point of the vertical profile of the image, the minimum point of the second derivative of the vertical profile, the minimum point of the horizontal profile, and the minimum point of the second derivative of the horizontal profile. Therefore, it is weak against noise. Therefore, a method that can stably detect rib cage boundaries regardless of image quality is desired.
Therefore, the present invention provides a rib cage detection method capable of correctly detecting the position of the rib cage boundary for all the rows from the lung field upper end line to the lung field lower end line stably regardless of the image quality.
本発明は、上記問題を解決するためになされたものであり、その特徴とするところは次の通りである。
（１）胸部Ｘ線像に１次微分オペレータを適用して各画素の勾配方向を求めた後、画像の左半分（右胸部）においては、右肺の肋骨側面部下縁の画素と同じ勾配方向をもつ画素（右エッジ要素と呼ぶ）の値を１とし、その他の画素の値を０として、また画像の右半分（左胸部）においては、左肺の肋骨側面部下縁の画素と同じ勾配方向をもつ画素（左エッジ要素と呼ぶ）の値を１とし、その他の画素の値を０として、２値画像Ｂを作成し、
この後、画像Ｂの左半分において、右エッジ要素の連結成分を求め、面積最大の連結成分の左境界点列（各行において連結成分の最も左側にある画素をつないで得られる道）を右側リブケイジ境界候補点列とし、画像Ｂの右半分において、左エッジ要素の連結成分を求め、面積最大の連結成分の右境界点列（各行において連結成分の最も右側にある画素をつないで得られる道）を左側リブケイジ境界候補点列とし、
次に、右側リブケイジ境界候補点列をLとして、まず、y方向に関してLに包含される右エッジ要素の連結成分を削除した後、左境界点列がLとの間でマージの条件を満たす連結成分があれば、Lとその左境界点列をマージして得られる点列をあらためて右側リブケイジ境界候補点列Lとし、この処理をLに新たな点列がマージされなくなるまで繰り返すことによって最終的な右側リブケイジ境界候補点列を求め、
また、同様な方法で左側リブケイジ境界候補点列を求める。
（２）前記（１）の方法で求まる右側リブケイジ境界候補点列の最上点の行番号と左側リブケイジ境界候補点列の最上点の行番号の平均によって、肺野上端線の位置を与える。
（３）前記（１）の方法で求まる右側リブケイジ境界候補点列と左側リブケイジ境界候補点列のそれぞれに対して、点列上の点の列番号をx、行番号をyとするとき、xがyの4次多項式で与えられると仮定して、点列にカーブフィッティングを適用して得られる曲線を右側リブケイジ境界、左側リブケイジ境界とし、これらと前記（２）の方法で求まる肺野上端線、および、肺野下端線によってリブケイジ境界を与える。
ことを特徴とするリブケイジ境界検出方法。
The present invention has been made to solve the above problems, and the features thereof are as follows.
(1) After applying the first-order differential operator to the chest X-ray image to determine the gradient direction of each pixel, in the left half of the image (right chest), the same gradient direction as the pixel at the lower edge of the rib side surface of the right lung The value of the pixel with the value (referred to as the right edge element) is 1, the value of the other pixels is 0, and in the right half of the image (left chest), the gradient direction is the same as the pixel at the lower edge of the rib side surface of the left lung A binary image B is created by setting the value of a pixel (referred to as a left edge element) having 1 to 1 and the values of other pixels to 0,
Thereafter, in the left half of the image B, the connected component of the right edge element is obtained, and the left boundary point sequence of the connected component with the largest area (the path obtained by connecting the pixels on the leftmost side of the connected component in each row) is the right rib cage. As a boundary candidate point sequence, the connected component of the left edge element is obtained in the right half of the image B, and the right boundary point sequence of the connected component having the largest area (the path obtained by connecting the pixels on the rightmost side of the connected component in each row) Is the left rib cage boundary candidate point sequence,
Next, assuming that the right rib cage boundary candidate point sequence is L, first, after removing the connected component of the right edge element included in L with respect to the y direction, the left boundary point sequence is connected to L satisfying the merge condition If there are components, the point sequence obtained by merging L and its left boundary point sequence is changed to the right rib cage boundary candidate point sequence L, and this process is repeated until no new point sequence is merged into L. A right-side rib cage boundary candidate point sequence,
Further, a left rib cage boundary candidate point sequence is obtained in the same manner.
(2) The position of the upper end line of the lung field is given by the average of the row number of the uppermost point of the right rib cage boundary candidate point sequence obtained by the method of (1) and the row number of the uppermost point of the left rib cage boundary candidate point sequence.
(3) For each of the right rib cage boundary candidate point sequence and the left rib cage boundary candidate point sequence obtained by the method of (1), when the column number of the point on the point sequence is x and the row number is y, x Is the right rib cage boundary and the left rib cage boundary obtained by applying curve fitting to the point sequence, and the upper line of the lung field obtained by the above method (2). And the rib cage boundary is given by the lower lung line.
The rib cage boundary detection method characterized by the above-mentioned.
医師による胸部単純Ｘ線写真からの肺癌診断を支援するための画像処理手法として、経時的差分（temporal subtraction）と呼ばれる方法がある。この方法は、同一被検者に対して撮影された過去画像と現在画像の２枚の画像の差をとることにより、２枚の画像間に存在する経時変化を強調する。過去画像に結節がなく、現在画像に結節があるとき、経時的差分によって現在画像中の結節が強調される。
過去画像と現在画像の間には、撮影体位やＸ線入射方向の差異に起因する位置ずれが存在するので、差分処理に先立ち、両者の間で位置合わせを行う必要があるが、胸部Ｘ線像における胸郭外の領域は再現性がない領域であることから、過去画像と現在画像間の位置合わせは、胸郭内部の解剖学的構造に関する情報を用いて行う必要がある。それ故、位置合わせに先立ち、過去画像と現在画像のそれぞれから、リブケイジ境界を検出し、胸郭内部領域を抽出する必要がある。本発明のリブケイジ境界検出方法は、画像に依らず安定して正確なリブケイジ境界を検出できることから、この方法を経時的差分における過去画像と現在画像の位置合わせに利用したとき、結節は強調され、かつ、偽陽性の数（差分画像の中で黒く強調された結節候補領域の中で、結節でないものの数）が少ない差分画像を得ることができ、この結果として、医師による肺癌の検出率を向上させることができる。
経時的差分が、過去画像と現在画像の２枚の画像の差分をとることによって結節を強調するのに対して、１枚の胸部Ｘ線像を胸郭中心軸で折り返して左右反転像をつくり、原画像と左右反転像の間で差分をとり結節を強調する方法があり、この方法は対側差分法（contralateral subtraction）と呼ばれている。対側差分法における原画像と左右反転像の位置合わせのためにも、胸郭内部領域を抽出することが重要であり、この目的のためにも、本発明のリブケイジ境界検出方法は利用できる。
There is a method called temporal subtraction as an image processing technique for supporting a lung cancer diagnosis from a chest simple radiograph by a doctor. In this method, a temporal change existing between two images is emphasized by taking a difference between two images of a past image and a current image taken for the same subject. When there is no nodule in the past image and there is a nodule in the current image, the nodule in the current image is emphasized by the temporal difference.
Since there is a positional shift between the past image and the current image due to a difference in the photographing position and the X-ray incident direction, it is necessary to perform alignment between the two before the difference processing. Since the region outside the rib cage in the image is a region having no reproducibility, alignment between the past image and the current image needs to be performed using information regarding the anatomical structure inside the rib cage. Therefore, prior to positioning, it is necessary to detect rib cage boundaries from each of the past image and the current image and extract the internal area of the rib cage. Since the rib cage boundary detection method of the present invention can detect the rib cage boundary stably and accurately regardless of the image, the nodule is emphasized when this method is used for registration of the past image and the current image in the temporal difference, In addition, it is possible to obtain a differential image with a small number of false positives (the number of non-nodule nodule candidate regions highlighted in black in the differential image). As a result, the detection rate of lung cancer by doctors is improved. Can be made.
While the temporal difference emphasizes the nodule by taking the difference between the two images of the past image and the current image, one chest X-ray image is folded back on the central axis of the rib cage to create a horizontally reversed image, There is a method of enhancing the nodule by taking a difference between the original image and the horizontally reversed image, and this method is called a contralateral subtraction method. The rib cage boundary detection method of the present invention can be used for the purpose of extracting the inner region of the rib cage also for the alignment of the original image and the horizontally reversed image in the contralateral difference method.
本発明の胸部Ｘ線像からのリブケイジ境界検出方法は、基本的に次の３ステップにより、リブケイジ境界を得るものである。
ステップ１：胸部X線像の左半分（右胸部）から右肺の肋骨側面部下縁の画素と同じ勾配方向をもつ画素を抽出し、これらの連結成分の左境界点列から右側リブケイジ境界候補点列をつくる。また、胸部X線像の右半分（左胸部）から左肺の側肋骨側面部下縁の画素と同じ勾配方向をもつ画素を抽出し、これらの連結成分の右境界点列から左側リブケイジ境界候補点列をつくる。
ステップ２：ステップ１で求まる右側リブケイジ境界候補点列の最上点の行番号と左側リブケイジ境界候補点列の最上点の行番号の平均によって、肺野上端線を与える。
ステップ３：ステップ１で得られた右側リブケイジ境界候補点列、左側リブケイジ境界候補点列にカーブフィッティングを適用して得られる曲線と、肺野上端線、肺野下端線によってリブケイジ境界を与える。
そこで、発明を実施するための具体的な最良の形態については、後述の実施例１により詳細に紹介する。
The rib cage boundary detection method from the chest X-ray image of the present invention basically obtains the rib cage boundary by the following three steps.
Step 1: Extract a pixel having the same gradient direction as the pixel at the lower edge of the rib side surface of the right lung from the left half (right chest) of the chest X-ray image, and right rib cage boundary candidate points from the left boundary point sequence of these connected components Create a line. Also, from the right half of the chest X-ray image (left chest), a pixel having the same gradient direction as the pixel at the lower edge of the lateral side wall of the left lung is extracted, and the left rib cage boundary candidate point from the right boundary point sequence of these connected components Create a line.
Step 2: A lung field upper end line is given by the average of the row number of the uppermost point of the right rib cage boundary candidate point sequence obtained in
Step 3: A rib cage boundary is given by a curve obtained by applying curve fitting to the right rib cage boundary candidate point sequence and the left rib cage boundary candidate point sequence obtained in
Therefore, a specific best mode for carrying out the invention will be introduced in detail in Example 1 described later.
本発明の胸部Ｘ線像からのリブケイジ境界検出方法の実施例を具体的な処理ステップ順で説明する。
以下の記述では、画像の左上隅を原点とし、画像の列、行をそれぞれx軸、y軸とする座標系を用いる（図１参照）。また、画像の列数、行数をM、Nで表す。さらに、入力として与えられる胸部Ｘ線像では、Ｘ線の透過量が多い領域ほど黒く、画素値は小さいものとする。従って、縦隔は白く（画素値が大きく）、肺野は黒い（画素値が小さい）。また、胸郭の中心軸は、ほぼ画像の中心軸近くにあり、画像の中心軸より左側に右肺、右側に左肺があると仮定する。
＜本発明のリブケイジ境界検出方法＞
１．肋骨側面部下縁エッジの検出
まず、肋骨側面部のコントラストを強調するため、原画像にヒストグラム平坦化を適用する。図１の胸部Ｘ線像にヒストグラム平坦化を適用して得られる画像を図２に示す。ヒストグラム平坦化によって胸部Ｘ線像全体のコントラストが改善されるわけではない。しかし、実験の結果、肺野中央部のようなＸ線の透過量が多い領域では、ヒストグラム平坦化によってコントラストは低下するが、縦隔やリブケイジ境界付近のようなＸ線の透過量が少ない領域では、ヒストグラム平坦化によりコントラストが改善されることが分った。
原画像にヒストグラム平坦化を適用した後、ヒストグラム平坦化適用後の画像に、１次微分オペレータを適用して、各画素(x, y)のx方向、y方向の一次微分Dx、Dyを求める。なお、１次微分オペレータとしては図３のマスクを用いる。そして、Dx、Dyから各画素の勾配（gradient）の大きさeと勾配方向φを次式で計算する。
An embodiment of a rib cage boundary detection method from a chest X-ray image according to the present invention will be described in the order of specific processing steps.
In the following description, a coordinate system is used in which the upper left corner of the image is the origin and the columns and rows of the image are the x axis and the y axis, respectively (see FIG. 1). Also, the number of columns and the number of rows of the image are represented by M and N. Further, in the chest X-ray image given as an input, it is assumed that the region where the amount of X-ray transmission is larger is blacker and the pixel value is smaller. Therefore, the mediastinum is white (pixel value is large) and the lung field is black (pixel value is small). Further, it is assumed that the central axis of the rib cage is almost near the central axis of the image, and the right lung is on the left side and the left lung is on the right side of the central axis of the image.
<The rib cage boundary detection method of this invention>
1. Detection of the lower edge of the rib side surface First, in order to enhance the contrast of the rib side surface portion, histogram flattening is applied to the original image. FIG. 2 shows an image obtained by applying histogram flattening to the chest X-ray image of FIG. Histogram flattening does not improve the overall contrast of the chest X-ray image. However, as a result of the experiment, in a region where the amount of X-ray transmission is large, such as the central part of the lung field, the contrast is reduced due to the flattening of the histogram, but the region where the amount of X-ray transmission is small such as the mediastinum or rib cage boundary. Then, it was found that the contrast is improved by the histogram flattening.
After applying histogram flattening to the original image, the first differential operator is applied to the image after applying histogram flattening to obtain the primary differentials Dx and Dy of each pixel (x, y) in the x and y directions. . Note that the mask shown in FIG. 3 is used as the primary differential operator. Then, the gradient magnitude e and the gradient direction φ of each pixel are calculated from Dx and Dy by the following equations.
右胸部においては、肋骨側面部は図６(a)に示すような形をしており、また、左胸部においては、肋骨側面部は図６(b)に示すような形をしている。しかも、肋骨は肋間部よりも白い（画素値が大きい）。それ故、右胸部においては、肋骨側面部下縁の画素のほとんどが135°〜270°の範囲の勾配方向をもち、左胸部においては、肋骨側面部下縁の画素のほとんどが270°〜405°の範囲の勾配方向をもつ。従って、画像Ｂで値1をもつ画素は、画像の左半分（右胸部）においては、右肺の肋骨側面部下縁の画素と同じ勾配方向をもつ画素であり、画像の右半分（左胸部）においては、左肺の肋骨側面部下縁の画素と同じ勾配方向をもつ画素である。
以後、記述を容易にするため、画像Ｂの左半分（右胸部）で値1をもつ画素を右エッジ要素と呼び、画像Ｂの右半分（左胸部）で値1をもつ画素を左エッジ要素と呼ぶ。
２．右側リブケイジ境界候補点列と左側リブケイジ境界候補点列の検出
図５の画像から得られる面積最大の右エッジ要素の連結成分と、面積最大の左エッジ要素の連結成分を、元の胸部Ｘ線像の上に重ねて示した図を、図７に示す（図７では、連結成分を見やすくするため、原画像のコントラストを低下させて示している）。図７の例では、面積最大の右エッジ要素の連結成分の左境界点列（各行において連結成分の最も左側にある画素をつないで得られる道）がほぼ右側リブケイジ境界に一致し、面積最大の左エッジ要素の連結成分の右境界点列（各行において連結成分の最も右側にある画素をつないで得られる道）がほぼ左側リブケイジ境界に一致していることが分かる。実験結果から、上に述べたようなことが、ほとんどの胸部Ｘ線像において成り立つことが分った。
そこで、本発明のリブケイジ境界検出法は、右側リブケイジ境界を求めるために、右エッジ要素からなる面積最大の連結成分の左境界点列を利用し、左側リブケイジ境界を求めるために、左エッジ要素からなる面積最大の連結成分の右境界点列を利用する。
ただ、ほとんどの画像では、図７の例のように、右リブケイジ境界のほぼ全体が、右エッジ要素からなる面積最大の連結成分の左境界点列によって与えられ、左側リブケイジ境界のほぼ全体が、左エッジ要素からなる面積最大の連結成分の右境界点列によって与えられるが、画像によっては、図８の例のように、面積最大の連結成分の境界点列に他の連結成分の境界点列をマージしなければ、右側リブケイジ境界（または左側リブケイジ境界）の全体が求まらない場合がある。（図８には、3個の連結成分A、B、Cが示されている。ここでAが面積最大の連結成分である。この例では、Aの右境界点列とBの右境界点列をマージして得られる点列が左側リブケイジ境界全体を与える。）
そこで、右側および左側リブケイジ境界候補点列を求めるために図９の処理を用いる。（まず、右エッジ要素に対して図９の処理を適用して右側リブケイジ境界候補点列を求め、次に、左エッジ要素に対して図９の処理を適用して左側リブケイジ境界候補点列を求める。）図９は、この発明の実施形態に係る胸部Ｘ線像からのリブケイジ境界検出方法を示したものであって、右側リブケイジ境界候補点列および左側リブケイジ境界候補点列の決定に関するフローチャートである。
以下に図９の処理を説明する。図９では、右エッジ要素の連結成分に対しては、その左境界点列を外側境界点列と呼び、左エッジ要素の連結成分に対しては、その右境界点列を外側境界点列と呼ぶ。さらに、右側リブケイジ境界候補点列（または左側リブケイジ境界候補点列）を、単にリブケイジ境界候補点列と呼ぶ。図９では、リブケイジ境界候補点列をＬで表している。図９の処理が終了した時点でのＬが、最終的な右側および左側リブケイジ境界候補点列を与える。
最初に、右側リブケイジ境界候補点列検出の場合は右エッジ要素の連結成分を求め、左側リブケイジ境界候補点列検出の場合は左エッジ要素の連結成分を求める（ステップＳ１）。次に、面積が閾値以下の連結成分を削除する（ステップＳ２）。実施例では、閾値としてM/10を用いた。そして、面積が最大の連結成分をR1とする（ステップＳ３）。次に、R1の外側境界点列を求め、これをLとし、またRをR1のみからなる集合とする（ステップＳ４）。ここまでの処理は初期設定のための処理であり、以後の処理はＬが更新されるたびに繰り返される。
まず、y方向に関して点列Lに包含される連結成分を削除する（ステップＳ５）。ただし、Lの最上点、最下点のy座標をY1、Y2とするとき、連結成分Rk中のすべての点が行Y1〜Y2の範囲に属すとき、Rkはy方向に関して点列Lに包含されるという。
次に、Ｌと重なった部分の長さの、Ｌの長さに対する比率が閾値以上の連結成分を削除する（ステップＳ６）。このステップの詳細は次の通りである。まず、点列Lの最上点、最下点のy座標をY1、Y2とし、両者の差をCで表す。そして、連結成分Rkの中で、次式で定義されるGの値のCに対する比率が閾値以上であるRkを削除する。実施例では閾値として0.7を用いた。ただし、次式におけるZ1、Z2は、Rkの最上点、最下点のy座標を表す。
In the right chest, the rib side surface has a shape as shown in FIG. 6 (a), and in the left chest, the rib side surface has a shape as shown in FIG. 6 (b). Moreover, the ribs are whiter than the intercostal space (the pixel value is larger). Therefore, in the right chest, most of the pixels at the lower edge of the rib side have a gradient direction ranging from 135 ° to 270 °, and in the left chest, most of the pixels at the lower edge of the rib side are from 270 ° to 405 °. Has a range gradient direction. Therefore, the pixel having the
Hereinafter, for ease of description, a pixel having a value of 1 in the left half (right chest) of the image B is referred to as a right edge element, and a pixel having a value of 1 in the right half (left chest) of the image B is referred to as a left edge element. Call it.
2. Detection of Right Rib Cage Boundary Candidate Point Sequence and Left Rib Cage Boundary Candidate Point Sequence The original chest X-ray image shows the connected component of the right edge element with the largest area and the connected component of the left edge element with the largest area obtained from the image of FIG. FIG. 7 shows a diagram superimposed on (in FIG. 7, the contrast of the original image is reduced in order to make the connected components easier to see). In the example of FIG. 7, the left boundary point sequence of the connected component of the right edge element with the largest area (the path obtained by connecting the pixels on the leftmost side of the connected component in each row) substantially coincides with the right rib cage boundary, It can be seen that the right boundary point sequence of the connected component of the left edge element (the path obtained by connecting the pixels on the rightmost side of the connected component in each row) substantially coincides with the left rib cage boundary. From the experimental results, it has been found that what has been described above holds in most chest X-ray images.
Therefore, the rib cage boundary detection method of the present invention uses the left boundary point sequence of the largest connected component consisting of the right edge element to obtain the right rib cage boundary, and uses the left edge element to obtain the left rib cage boundary. The right boundary point sequence of the connected component with the largest area is used.
However, in most images, as shown in the example of FIG. 7, almost the entire right rib cage boundary is given by the left boundary point sequence of the largest connected component consisting of the right edge elements, and almost the entire left rib cage boundary is Although given by the right boundary point sequence of the largest connected component composed of the left edge elements, depending on the image, as shown in the example of FIG. Otherwise, the entire right rib cage boundary (or left rib cage boundary) may not be obtained. (FIG. 8 shows three connected components A, B, and C. Here, A is the connected component with the largest area. In this example, the right boundary point sequence of A and the right boundary point of B (The sequence of points obtained by merging the columns gives the entire left rib cage boundary.)
9 is used to obtain the right and left rib cage boundary candidate point sequences. (First, the processing in FIG. 9 is applied to the right edge element to obtain the right rib cage boundary candidate point sequence, and then the processing in FIG. 9 is applied to the left edge element to obtain the left rib cage boundary candidate point sequence. FIG. 9 shows a rib cage boundary detection method from a chest X-ray image according to an embodiment of the present invention, and is a flowchart relating to determination of a right rib cage boundary candidate point sequence and a left rib cage boundary candidate point sequence. is there.
The process of FIG. 9 will be described below. In FIG. 9, for the connected component of the right edge element, the left boundary point sequence is referred to as the outer boundary point sequence, and for the connected component of the left edge element, the right boundary point sequence is referred to as the outer boundary point sequence. Call. Further, the right rib cage boundary candidate point sequence (or the left rib cage boundary candidate point sequence) is simply referred to as a rib cage boundary candidate point sequence. In FIG. 9, the rib cage boundary candidate point sequence is represented by L. L at the time when the processing of FIG. 9 ends gives the final right and left rib cage boundary candidate point sequences.
First, in the case of detecting the right rib cage boundary candidate point sequence, the connected component of the right edge element is obtained, and in the case of detecting the left rib cage boundary candidate point sequence, the connected component of the left edge element is obtained (step S1). Next, connected components whose area is equal to or smaller than the threshold are deleted (step S2). In the example, M / 10 was used as the threshold value. The connected component having the largest area is defined as R1 (step S3). Next, an outer boundary point sequence of R1 is obtained, and this is set as L, and R is set as a set consisting only of R1 (step S4). The processing so far is processing for initial setting, and the subsequent processing is repeated every time L is updated.
First, connected components included in the point sequence L with respect to the y direction are deleted (step S5). However, when the y-coordinates of the top and bottom points of L are Y1 and Y2, Rk is included in the point sequence L with respect to the y direction when all the points in the connected component Rk belong to the range of rows Y1 to Y2. It is said.
Next, a connected component whose ratio of the length of the portion overlapping L to the length of L is equal to or greater than a threshold is deleted (step S6). The details of this step are as follows. First, the y-coordinates of the uppermost point and lowermost point of the point sequence L are Y1 and Y2, and the difference between them is represented by C. Then, Rk in which the ratio of the G value defined by the following equation to C is greater than or equal to the threshold is deleted from the connected components Rk. In the embodiment, 0.7 was used as the threshold value. However, Z1 and Z2 in the following expression represent the y-coordinates of the highest point and the lowest point of Rk.
そして、Sが空ならば、現在のLを最終的なリブケイジ境界候補点列として処理を終了する（ステップＳ８）。Sが空でないなら、Sから1個の連結成分Rkを取り出し、Rkの外側境界点列をLkとする（ステップＳ９）。そして、後述する方法を用いて、LkとLがマージの条件を満たすか否かを判定する（ステップＳ１０）。そして、LkがLとの間でマージの条件を満たすときは、後述する方法を用いてLとLkをマージした後、マージ後の点列をあらためてLとして、さらにRにRkを挿入して、ステップＳ５へ戻る（ステップＳ１１）。Lkがマージの条件を満たさないときは、RkをSから削除した後、ステップＳ９を繰り返す。
以下に、点列LkとLをマージする方法、および、LkとLがマージの条件を満たすか否かを判定する方法を示す。
まず、点列LkとLをマージする方法を示す。Lは行Y1〜Y2に存在し、Lkは行Z1〜Z2の範囲に存在するものとする。ｙ方向に関してLに包含される点列はステップＳ５で削除されているので、Lkはｙ方向に関してLに包含されない点列である。それ故、Y1，Y2，Z1，Z2の間の大小関係としては、次の（A）〜（D）のいずれかが成り立つ。
（A）Z1 ＜Z2 ＜ Y1 ＜ Y2
（B）Z1 ＜ Y1 ≦ Z2 ＜ Y2
（C）Y1 ＜ Z1 ≦ Y2 ＜ Z2
（D）Y1 ＜ Y2 ＜ Z1 ＜ Z2
以下に、LとLkをマージする方法を上記の(A)〜(D)の場合ごとに示す。なお、マージ後の点列は（A）、（B）の場合は、行Z1〜Y2の範囲に存在し、（C）、（D）の場合は行Y1〜Z2の範囲に存在する。
（A）の場合、マージ後の点列の行y（y = Z1〜Y2）における点は、行Z1〜Z2に対してはLkの点で与え、行Y1〜Y2に対してはLの点で与える。そして、行Z2＋1〜Y1−1に対しては、行Z2におけるLkの点と行Y1におけるLの点を結ぶ線分上の点によって与える。
（B）の場合、マージ後の点列の行y（y = Z1〜Y2）における点は、行Z1〜Y1−1に対してはLkの点で与え、行Z2＋1〜Y2に対してはLの点で与える。そして、行Y1〜Z2に対しては、右側リブケイジ境界候補点列検出の場合は、Lの点とLkの点のうち左側にある点によって与え、左側リブケイジ境界候補点列検出の場合は、Lの点とLkの点のうち右側にある点によって与える。
（C）の場合、マージ後の点列の行y（y = Y1〜Z2）における点は、行Y1〜Z1−1に対してはLの点で与え、行Y2＋1〜Z2に対してはLkの点で与える。そして行Z1〜Y2に対しては、右側リブケイジ境界候補点列検出の場合は、Lの点とLkの点のうち左側にある点によって与え、左側リブケイジ境界候補点列検出の場合は、Lの点とLkの点のうち右側にある点によって与える。
（D）の場合、マージ後の点列の行y（y = Y1〜Z2）における点は、行Y1〜Y2に対してはLの点で与え、行Z1〜Z2に対してはLkの点で与える。そして行Y2＋1〜Z1−1に対しては、行Y2におけるLの点と行Z1におけるLkの点を結ぶ線分上の点によって与える。
次に、点列LとLkをマージすべきか否かを判定するために用いた方法を示す。右側リブケイジ境界を下から上へたどったとき、左方向に大きく変化することはない。また左側リブケイジ境界を下から上へたどったとき、右方向に大きく変化することはない。そこで、このことを用いて、LとLkをマージすべきか否かを判定する。LとLk間のマージの条件は、前述した（A）、（B）の場合と、（C）、（D）の場合で異なる。
（A）、（B）の場合は、Zを次式で与える。
If S is empty, the process ends with the current L as the final rib cage boundary candidate point sequence (step S8). If S is not empty, one connected component Rk is extracted from S, and the outer boundary point sequence of Rk is set to Lk (step S9). Then, using a method to be described later, it is determined whether Lk and L satisfy the merging condition (step S10). And when Lk satisfies the merge condition with L, after merging L and Lk using the method described later, the point sequence after merging is changed to L, and Rk is further inserted into R, The process returns to step S5 (step S11). If Lk does not satisfy the merging condition, Rk is deleted from S and then step S9 is repeated.
Hereinafter, a method for merging point sequences Lk and L and a method for determining whether or not Lk and L satisfy the merging condition will be described.
First, a method of merging the point sequences Lk and L is shown. L is present in the rows Y1 to Y2, and Lk is present in the range of the rows Z1 to Z2. Since the point sequence included in L with respect to the y direction is deleted in step S5, Lk is a point sequence not included in L with respect to the y direction. Therefore, any one of the following (A) to (D) holds as the magnitude relationship between Y1, Y2, Z1, and Z2.
(A) Z1 <Z2 <Y1 <Y2
(B) Z1 <Y1 ≤ Z2 <Y2
(C) Y1 <Z1 ≤ Y2 <Z2
(D) Y1 <Y2 <Z1 <Z2
A method for merging L and Lk will be described below for each of the cases (A) to (D). The point sequence after merging exists in the range of rows Z1 to Y2 in the case of (A) and (B), and exists in the range of rows Y1 to Z2 in the cases of (C) and (D).
For (A), the points in row y (y = Z1 to Y2) of the merged point sequence are given as Lk points for rows Z1 to Z2 and L points for rows Y1 to Y2. Give in. For the rows Z2 + 1 to Y1-1, the points on the line segment connecting the Lk point in the row Z2 and the L point in the row Y1 are given.
In the case of (B), the points in the row y (y = Z1 to Y2) of the point sequence after merging are given as Lk points for the rows Z1 to Y1-1, and L for the rows Z2 + 1 to Y2. Give in terms of. And for rows Y1 to Z2, in the case of right rib cage boundary candidate point sequence detection, it is given by the point on the left side of the L point and Lk point, and in the case of left rib cage boundary candidate point sequence detection, L Is given by the point on the right side of the points and Lk.
In the case of (C), the points in the row y (y = Y1 to Z2) of the point sequence after merging are given as L points for the rows Y1 to Z1-1, and Lk for the rows Y2 + 1 to Z2 Give in terms of. And for rows Z1 to Y2, in the case of right rib cage boundary candidate point sequence detection, it is given by the point on the left side of the points of L and Lk, and in the case of left rib cage boundary candidate point sequence detection, It is given by the point on the right side of the points and Lk points.
For (D), the points in row y (y = Y1 to Z2) of the merged point sequence are given as L points for rows Y1 to Y2 and Lk points for rows Z1 to Z2 Give in. For the rows Y2 + 1 to Z1-1, the points on the line segment connecting the L point in the row Y2 and the Lk point in the row Z1 are given.
Next, the method used to determine whether or not the point sequences L and Lk should be merged is shown. When the right rib cage boundary is traced from bottom to top, it does not change significantly to the left. When the left rib cage boundary is traced from bottom to top, there is no significant change in the right direction. Therefore, this is used to determine whether L and Lk should be merged. The conditions for merging between L and Lk differ between the cases (A) and (B) described above and the cases (C) and (D).
For (A) and (B), Z is given by the following equation.
（C）、（D）の場合は、Zを次式で与える。
For (C) and (D), Z is given by the following equation.
（A）〜（D）のいずれの場合においても、Zが閾値以下のとき、LとLkはマージの条件を満たすと判定する。なお実施例では、閾値としてM/10を用いた。
３．肺野上端線の求め方
右肺と左肺の肺尖位置（肺尖のy方向の位置）は、一般には異なるので、通常は、右肺と左肺の肺尖位置の平均として肺野上端線が定義される。しかし、右肺と左肺の肺尖位置を画像によらず安定して検出できる方法は、現在までのところ知られていない。
本発明の方法は、２．で述べた方法によって求まる右側リブケイジ境界候補点列の最上点のy座標YR、および、左側リブケイジ境界候補点列の最上点のy座標YLによって、右肺の肺尖位置と左肺の肺尖位置を与え、肺野上端線の位置ytを次式で与える。
In any of the cases (A) to (D), when Z is equal to or smaller than the threshold value, L and Lk are determined to satisfy the merge condition. In the embodiment, M / 10 is used as the threshold value.
3. How to find the top of the lung field Since the apex position of the right lung and the left lung (the position of the apex in the y direction) is generally different, usually the top of the lung field is the average of the apex positions of the right and left lungs. A line is defined. However, a method that can stably detect the apex positions of the right lung and the left lung irrespective of an image has not been known so far.
The method of the present invention comprises: The apex position of the right lung and the apex of the left lung by the y coordinate YR of the top point of the right rib cage boundary candidate point sequence and the y coordinate YL of the top point of the left rib cage boundary candidate point sequence obtained by the method described in And the position yt of the upper lung line is given by
４．右側リブケイジ境界と左側リブケイジ境界の求め方
本発明の方法は、２．で述べた方法で求まる右側または左側リブケイジ境界候補点列上の点を(x, y)とするとき、xとyの間に数７の関係が成り立つと仮定して、
4). How to find the right rib cage boundary and the left rib cage boundary Assuming that a point on the right or left rib cage boundary candidate point sequence obtained by the method described in (x) is (x, y), the relationship of Equation 7 is established between x and y.
そして、このようにして求まる曲線の行yt〜ybの部分を、右側リブケイジ境界（または左側リブケイジ境界）とする。ただし、ytは３．で述べた方法で求まる肺野上端線の位置を表し、ybは５．で述べる方法によって求まる肺野下端線の位置を表す。
図７の画像において、２．で述べた方法で求まるリブケイジ境界候補点列上の点(x, y)に、数７の形の曲線を適合させて得られる曲線を図１０に示す。図１０には、また、３．で述べた方法によって求まる肺野上端線の位置ytと、５．で述べる方法によって求まる肺野下端線の位置ybを示している。
本発明の方法が、上述したように、リブケイジ境界候補点列上の点を(x, y)とするとき、xをyの4次多項式で与えるのに対して、従来法は、yをxの4次多項式または3次多項式で与える。つまり、リブケイジ境界を次の数８や数９で与える。
The portion of the curve line yt to yb obtained in this way is defined as the right rib cage boundary (or left rib cage boundary). However, yt is 3. Represents the position of the upper end line of the lung field obtained by the method described in
In the image of FIG. FIG. 10 shows a curve obtained by fitting the curve of the formula 7 to the point (x, y) on the rib cage boundary candidate point sequence obtained by the method described in the above. In FIG. 4. The position yt of the lung field upper end line obtained by the method described in 1. The position yb of the lower end line of the lung field obtained by the method described in Fig. 4 is shown.
As described above, in the method of the present invention, when a point on the rib cage boundary candidate point sequence is (x, y), x is given by a fourth-order polynomial of y, whereas in the conventional method, y is set to x Is given as a 4th or 3rd order polynomial. That is, the rib cage boundary is given by the following
２．で述べた方法によって求まるリブケイジ境界候補点列は、行yt〜ybの範囲でほぼ正しいリブケイジ境界の位置を与えるので、この点列に数７を適合させて得られる曲線は、行yt〜ybの範囲でほぼ正しいリブケイジ境界を与える。一方、従来法で求まるリブケイジ境界候補点列は、特に上肺部において、誤差が大きな点を含むことが多い。それ故、従来法から求まるリブケイジ境界候補点列に数７の形の曲線を適合させても、正しいリブケイジ境界が求まらないことが多い。
５．肺野下端線の検出
前述したように、２．で述べた方法から求まる右側リブケイジ境界候補点列、左側リブケイジ境界候補点列の最上点は、ほぼ肺野上端線の位置を与える。しかし、これらの点列の最下点は、必ずしも、肺野下端線の位置を与えない。（肋骨は肺野下端線より下にも存在するので、これらの点列の最下点の位置が、肺野下端線より下になることもある。）それ故、肺野下端線を求めるためには別の方法を用いる必要がある。
本発明では従来法と同様に、右肺と横隔膜の境界のy方向の位置を検出し、これを肺野下端線とする。
まず、４．の方法で得られた右側リブケイジ境界と左側リブケイジ境界のy＝N／2との交点をＰ，Ｑとし、線分ＰＱの中点をＣとする。さらに、線分ＰＣの中点のx座標をX0で表す。このとき、x＝X0は右肺のほぼ中央を通る直線となる。
次に、画像の下半分の行yに対して、次式によってV(y)を計算する。
2. Since the rib cage boundary candidate point sequence obtained by the method described in the above gives the position of the rib cage boundary almost correct in the range of rows yt to yb, the curve obtained by applying Equation 7 to this point sequence is the row yt to yb. Gives an almost correct rib cage boundary in range. On the other hand, the rib cage boundary candidate point sequence obtained by the conventional method often includes a point having a large error, particularly in the upper lung portion. Therefore, even if a curve of the formula 7 is fitted to the rib cage boundary candidate point sequence obtained from the conventional method, the correct rib cage boundary is often not obtained.
5. Detection of lung field bottom line As described above, 2. The uppermost point of the right rib cage boundary candidate point sequence and the left rib cage boundary candidate point sequence obtained from the method described in (1) gives the position of the upper end line of the lung field. However, the lowest point of these point sequences does not necessarily give the position of the lung field bottom line. (Since the ribs are also below the lower end line of the lung field, the position of the lowest point of these point sequences may be lower than the lower end line of the lung field.) It is necessary to use another method.
In the present invention, as in the conventional method, the position in the y direction of the boundary between the right lung and the diaphragm is detected, and this is used as the lung field lower end line.
First, 4. The intersections of the right rib cage boundary and the left rib cage boundary y = N / 2 obtained by the above method are P and Q, and the midpoint of the line segment PQ is C. Further, the x coordinate of the midpoint of the line segment PC is represented by X0. At this time, x = X0 is a straight line passing through almost the center of the right lung.
Next, V (y) is calculated by the following equation for the lower half row y of the image.
医師による胸部単純Ｘ線写真からの肺癌診断を支援するための画像処理手法として、経時的差分（temporal subtraction）と呼ばれる方法がある。この方法は、同一被検者に対して撮影された過去画像と現在画像の２枚の画像の差をとることにより、２枚の画像間に存在する経時変化を強調する。過去画像に結節がなく、現在画像に結節があるとき、経時的差分によって現在画像中の結節が強調される。
過去画像と現在画像の間には、撮影体位やＸ線入射方向の差異に起因する位置ずれが存在するので、差分処理に先立ち、両者の間で位置合わせを行う必要があるが、胸部Ｘ線像における胸郭外の領域は再現性がない領域であることから、過去画像と現在画像間の位置合わせは、胸郭内部の解剖学的構造に関する情報を用いて行う必要がある。それ故、位置合わせに先立ち、過去画像と現在画像のそれぞれから、リブケイジ境界を検出し、胸郭内部領域を抽出する必要がある。本発明のリブケイジ境界検出方法は、画像に依らず安定して正確なリブケイジ境界を検出できることから、この方法を経時的差分における過去画像と現在画像の位置合わせに利用したとき、結節は強調され、かつ、偽陽性の数（差分画像の中で黒く強調された結節候補領域の中で、結節でないものの数）が少ない差分画像を得ることができ、この結果として、医師による肺癌の検出率を向上させることができる。
経時的差分が、過去画像と現在画像の２枚の画像の差分をとることによって結節を強調するのに対して、１枚の胸部Ｘ線像を胸郭中心軸で折り返して左右反転像をつくり、元画像と左右反転像の間で差分をとり結節を強調する方法があり、この方法は対側差分法（contralateral subtraction）と呼ばれている。対側差分法における元画像と左右反転像の位置合わせのためにも、胸郭内部領域を抽出することが重要であり、この目的のためにも、本発明のリブケイジ境界検出方法は利用できる。
There is a method called temporal subtraction as an image processing technique for supporting a lung cancer diagnosis from a chest simple radiograph by a doctor. In this method, a temporal change existing between two images is emphasized by taking a difference between two images of a past image and a current image taken for the same subject. When there is no nodule in the past image and there is a nodule in the current image, the nodule in the current image is emphasized by the temporal difference.
Since there is a positional shift between the past image and the current image due to a difference in the photographing position and the X-ray incident direction, it is necessary to perform alignment between the two before the difference processing. Since the region outside the rib cage in the image is a region having no reproducibility, alignment between the past image and the current image needs to be performed using information regarding the anatomical structure inside the rib cage. Therefore, prior to positioning, it is necessary to detect rib cage boundaries from each of the past image and the current image and extract the internal area of the rib cage. Since the rib cage boundary detection method of the present invention can detect the rib cage boundary stably and accurately regardless of the image, the nodule is emphasized when this method is used for registration of the past image and the current image in the temporal difference, In addition, it is possible to obtain a differential image with a small number of false positives (the number of non-nodule nodule candidate regions highlighted in black in the differential image). As a result, the detection rate of lung cancer by doctors is improved. Can be made.
While the temporal difference emphasizes the nodule by taking the difference between the two images of the past image and the current image, one chest X-ray image is folded back on the central axis of the rib cage to create a horizontally reversed image, There is a method of enhancing the nodule by taking a difference between the original image and the horizontally reversed image, and this method is called a contralateral subtraction method. The rib cage boundary detection method of the present invention can be used for the purpose of extracting the inner region of the rib cage also for the alignment of the original image and the horizontally reversed image in the contralateral difference method.
A、B、C：連結成分
N：画像の列数
M：画像の行数
A, B, C: Connected components
N: Number of image columns
M: Number of lines in the image
Claims (1)
この後、画像Ｂの左半分において、右エッジ要素の連結成分を求め、面積最大の連結成分の左境界点列（各行において連結成分の最も左側にある画素をつないで得られる道）を右側リブケイジ境界候補点列とし、画像Ｂの右半分において、左エッジ要素の連結成分を求め、面積最大の連結成分の右境界点列（各行において連結成分の最も右側にある画素をつないで得られる道）を左側リブケイジ境界候補点列とし、
次に、右側リブケイジ境界候補点列をLとして、まず、y方向に関してLに包含される右エッジ要素の連結成分を削除した後、左境界点列がLとの間でマージの条件を満たす連結成分があれば、Lとその左境界点列をマージして得られる点列をあらためて右側リブケイジ境界候補点列Lとし、この処理をLに新たな点列がマージされなくなるまで繰り返すことによって最終的な右側リブケイジ境界候補点列を求め、
また、同様な方法で左側リブケイジ境界候補点列を求める。
（２）前記（１）の方法で求まる右側リブケイジ境界候補点列の最上点の行番号と左側リブケイジ境界候補点列の最上点の行番号の平均によって、肺野上端線の位置を与える。
（３）前記（１）の方法で求まる右側リブケイジ境界候補点列と左側リブケイジ境界候補点列のそれぞれに対して、点列上の点の列番号をx、行番号をyとするとき、xがyの4次多項式で与えられると仮定して、点列にカーブフィッティングを適用して得られる曲線を右側リブケイジ境界、左側リブケイジ境界とし、これらと前記（２）の方法で求まる肺野上端線、および、肺野下端線によってリブケイジ境界を与える。
ことを特徴とするリブケイジ境界検出方法。 (1) After applying the first-order differential operator to the chest X-ray image to determine the gradient direction of each pixel, in the left half of the image (right chest), the same gradient direction as the pixel at the lower edge of the rib side surface of the right lung The value of the pixel with the value (referred to as the right edge element) is 1, the value of the other pixels is 0, and in the right half of the image (left chest), the gradient direction is the same as the pixel at the lower edge of the rib side surface of the left lung A binary image B is created by setting the value of a pixel (referred to as a left edge element) having 1 to 1 and the values of other pixels to 0,
Thereafter, in the left half of the image B, the connected component of the right edge element is obtained, and the left boundary point sequence of the connected component with the largest area (the path obtained by connecting the pixels on the leftmost side of the connected component in each row) is the right rib cage. As a boundary candidate point sequence, the connected component of the left edge element is obtained in the right half of the image B, and the right boundary point sequence of the connected component having the largest area (the path obtained by connecting the pixels on the rightmost side of the connected component in each row) Is the left rib cage boundary candidate point sequence,
Next, assuming that the right rib cage boundary candidate point sequence is L, first, after removing the connected component of the right edge element included in L with respect to the y direction, the left boundary point sequence is connected to L satisfying the merge condition If there are components, the point sequence obtained by merging L and its left boundary point sequence is changed to the right rib cage boundary candidate point sequence L, and this process is repeated until no new point sequence is merged into L. A right-side rib cage boundary candidate point sequence,
Further, a left rib cage boundary candidate point sequence is obtained in the same manner.
(2) The position of the upper end line of the lung field is given by the average of the row number of the uppermost point of the right rib cage boundary candidate point sequence obtained by the method of (1) and the row number of the uppermost point of the left rib cage boundary candidate point sequence.
(3) For each of the right rib cage boundary candidate point sequence and the left rib cage boundary candidate point sequence obtained by the method of (1), when the column number of the point on the point sequence is x and the row number is y, x Is the right rib cage boundary and the left rib cage boundary obtained by applying curve fitting to the point sequence, and the upper line of the lung field obtained by the above method (2). And the rib cage boundary is given by the lower lung line.
The rib cage boundary detection method characterized by the above-mentioned.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JP2007081989A JP4639338B2 (en) | 2007-03-20 | 2007-03-27 | Rib cage boundary detection method from chest X-ray image |
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JP2007073264 | 2007-03-20 | ||
| JP2007081989A JP4639338B2 (en) | 2007-03-20 | 2007-03-27 | Rib cage boundary detection method from chest X-ray image |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| JP2008259522A JP2008259522A (en) | 2008-10-30 |
| JP4639338B2 true JP4639338B2 (en) | 2011-02-23 |
Family
ID=39982501
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| JP2007081989A Active JP4639338B2 (en) | 2007-03-20 | 2007-03-27 | Rib cage boundary detection method from chest X-ray image |
Country Status (1)
| Country | Link |
|---|---|
| JP (1) | JP4639338B2 (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP5267029B2 (en) * | 2007-10-12 | 2013-08-21 | 株式会社ニコン | Illumination optical apparatus, exposure apparatus, and device manufacturing method |
| JP6657893B2 (en) * | 2015-12-09 | 2020-03-04 | 株式会社島津製作所 | Image processing apparatus, program and radiation imaging apparatus |
| US10685258B2 (en) | 2018-06-15 | 2020-06-16 | Shimadzu Corporation | Image processing apparatus, program, and radiographic imaging apparatus |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2002177249A (en) * | 2000-12-13 | 2002-06-25 | Mitsubishi Space Software Kk | Ribcage boundary detecting method of digital chest radiograph |
| JP2002366961A (en) * | 2001-04-03 | 2002-12-20 | Fuji Photo Film Co Ltd | Method and device for detecting image contour |
| JP2003006661A (en) * | 2001-06-22 | 2003-01-10 | Fuji Photo Film Co Ltd | Thoracic contour detector |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH07100060B2 (en) * | 1987-03-30 | 1995-11-01 | コニカ株式会社 | Image measurement method to obtain lung contour lines |
| US5790690A (en) * | 1995-04-25 | 1998-08-04 | Arch Development Corporation | Computer-aided method for automated image feature analysis and diagnosis of medical images |
-
2007
- 2007-03-27 JP JP2007081989A patent/JP4639338B2/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2002177249A (en) * | 2000-12-13 | 2002-06-25 | Mitsubishi Space Software Kk | Ribcage boundary detecting method of digital chest radiograph |
| JP2002366961A (en) * | 2001-04-03 | 2002-12-20 | Fuji Photo Film Co Ltd | Method and device for detecting image contour |
| JP2003006661A (en) * | 2001-06-22 | 2003-01-10 | Fuji Photo Film Co Ltd | Thoracic contour detector |
Also Published As
| Publication number | Publication date |
|---|---|
| JP2008259522A (en) | 2008-10-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107292857B (en) | Image processing apparatus and method, and computer-readable storage medium | |
| TWI520078B (en) | Optical flow tracking method and device | |
| KR101703515B1 (en) | Apparatus and method for target tracking of image | |
| WO2013018575A1 (en) | Image diagnostic device and image correction method | |
| JP5538868B2 (en) | Image processing apparatus, image processing method and program | |
| JP6502511B2 (en) | Calculation device, control method of calculation device, and calculation program | |
| CN104079912A (en) | Image processing apparatus and image processing method | |
| JP6935786B2 (en) | Geometric transformation matrix estimation device, geometric transformation matrix estimation method, and program | |
| JP6598850B2 (en) | Image processing apparatus, image processing method, and image processing program | |
| US11200443B2 (en) | Image processing apparatus, image processing method, and image processing system | |
| WO2019053842A1 (en) | State-of-change detection device | |
| US20190333221A1 (en) | Image processing apparatus, image processing method, and storage medium | |
| JPWO2010095460A1 (en) | Image processing system, image processing method, and image processing program | |
| US8306354B2 (en) | Image processing apparatus, method, and program | |
| JP2009139323A (en) | Travel road surface detecting apparatus for vehicle | |
| JP4639338B2 (en) | Rib cage boundary detection method from chest X-ray image | |
| WO2002047553A1 (en) | Method of detecting rib cage boundary for digital chest image | |
| JP4708740B2 (en) | Image processing apparatus and image processing method | |
| JP6057629B2 (en) | Image processing apparatus, control method thereof, and control program | |
| JP4379706B2 (en) | Region extraction method | |
| JP2011133954A (en) | Edge extraction method and edge extraction device | |
| JP2010009236A (en) | Plane area estimation device and program | |
| WO2016170655A1 (en) | Image processing device, image processing method and image processing program | |
| CN113269732B (en) | Linear object detection method based on characteristic scanning image | |
| JP4613317B2 (en) | Subtraction method for detecting temporal changes from two chest X-ray images that are temporally continuous |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| A977 | Report on retrieval |
Free format text: JAPANESE INTERMEDIATE CODE: A971007 Effective date: 20101015 |
|
| TRDD | Decision of grant or rejection written | ||
| A01 | Written decision to grant a patent or to grant a registration (utility model) |
Free format text: JAPANESE INTERMEDIATE CODE: A01 Effective date: 20101029 |
|
| A01 | Written decision to grant a patent or to grant a registration (utility model) |
Free format text: JAPANESE INTERMEDIATE CODE: A01 |
|
| R150 | Certificate of patent or registration of utility model |
Free format text: JAPANESE INTERMEDIATE CODE: R150 |