技术领域Technical Field
本发明涉及外科前交叉韧带重建手术技术领域，具体涉及一种人群股骨外上髁处髁间线曲率测算系统和方法。The invention relates to the technical field of surgical anterior cruciate ligament reconstruction, and in particular to a system and method for calculating the curvature of an intercondylar line at the lateral epicondyle of a population of femurs.
背景技术Background technique
前交叉韧带（ACL）损伤是运动人群最常见的膝关节损伤之一，在前交叉韧带重建手术中，准确定位股骨隧道和胫骨隧道（尤其是股骨隧道）的起始解剖位点，对于恢复膝关节稳定性、防止撞击和移植物失败至关重要。前交叉韧带重建术中股骨隧道的定位会显著影响手术的效果，然而，在手术中医生发现不同人群的股骨外上髁处髁间线曲率特征不同，会影响现有股骨隧道定位器的稳定固定，影响手术效果。对人群股骨外上髁处髁间线曲率特征进行测量分析，从而指导设计出与解剖结构更匹配，定位更加准确的定位器对精准手术具有重要意义。本发明基于人工智能开发出能够对股骨髁间线近外侧髁点位进行识别并计算出位点的曲率的方法。Anterior cruciate ligament (ACL) injury is one of the most common knee injuries in sports people. In anterior cruciate ligament reconstruction surgery, accurate positioning of the starting anatomical sites of the femoral tunnel and tibial tunnel (especially the femoral tunnel) is crucial to restore knee stability, prevent impact and graft failure. The positioning of the femoral tunnel during anterior cruciate ligament reconstruction will significantly affect the effect of the operation. However, during the operation, doctors found that the curvature characteristics of the intercondylar line at the lateral epicondyle of the femur in different populations are different, which will affect the stable fixation of the existing femoral tunnel locator and affect the effect of the operation. Measuring and analyzing the curvature characteristics of the intercondylar line at the lateral epicondyle of the femur of the population is of great significance for precision surgery, thereby guiding the design of a locator that better matches the anatomical structure and has more accurate positioning. The present invention develops a method based on artificial intelligence that can identify the near-lateral condylar point of the femoral intercondylar line and calculate the curvature of the site.
股骨定位器辅助手术中韧带股骨止点定位的操作，用于定位股骨隧道的位置、控制股骨隧道的方向。目前现有技术中常用股骨的定位器末端夹角并不能良好地与髁间线的生理曲面良好匹配。虽然现有的定位器设计能够使术者初步确定前交叉韧带股骨止点位置并进行骨隧道钻取，但是医生在大量的ACL重建手术中发现，人群中股骨外上髁处髁间窝与腘面夹角存在明显差异。股骨外上髁处髁间窝与腘面夹角，是指外侧髁的上面、髁间窝和腘面交界处的区域，是ACL重建手术中股骨定位器接触的定位部位。对于部分患者，现有的股骨定位器无法精准贴合该股骨位点，不能形成牢固固定，影响手术定位效果。The femoral locator assists in the positioning of the femoral attachment point of the ligament during surgery, and is used to locate the position of the femoral tunnel and control the direction of the femoral tunnel. The angle of the end of the femoral locator commonly used in the current prior art cannot match the physiological curve of the intercondylar line well. Although the existing locator design enables the surgeon to preliminarily determine the position of the femoral attachment point of the anterior cruciate ligament and drill the bone tunnel, doctors have found in a large number of ACL reconstruction surgeries that there are significant differences in the angle between the intercondylar fossa and the popliteal surface at the lateral epicondyle of the femur. The angle between the intercondylar fossa and the popliteal surface at the lateral epicondyle of the femur refers to the area above the lateral condyle, the junction of the intercondylar fossa and the popliteal surface, and is the positioning site that the femoral locator contacts during ACL reconstruction surgery. For some patients, the existing femoral locator cannot accurately fit the femoral site, and cannot form a firm fixation, which affects the surgical positioning effect.
现有的定位方案并没有关注到不同人群的股骨外上髁处髁间线曲率差异。由此设计出来的股骨定位器并不能适用于所有人群，导致现有的股骨定位器与部分患者股骨位点不能精准对接并形成牢固固定。因此，即使患者的股骨定位位点明确，使用传统的股骨定位器也会给定位操作带来一定误差，进而影响手术的效果。The existing positioning scheme does not pay attention to the difference in the curvature of the intercondylar line at the lateral epicondyle of the femur among different populations. The femoral locator designed in this way is not suitable for all populations, resulting in the inability of the existing femoral locator to accurately dock with the femoral site of some patients and form a firm fixation. Therefore, even if the patient's femoral positioning site is clear, the use of a traditional femoral locator will cause a certain error in the positioning operation, thereby affecting the effect of the surgery.
在大量ACL重建手术的操作中，研究人员关注到不同人群的股骨外上髁处髁间线曲率的差异，对此进行了研究。本申请依托于大量数据的统计分析，提出了一种膝关节髁间线测量方法，并将不同曲率结构分型，定制标准致力于解决这个问题。值得注意的是，新的分型结果是在新的研究方法下得到的。本申请开发的智能测算系统能够对股骨外上髁处髁间线近外侧髁点位进行自动智能识别并计算出位点的曲率，利用膝关节CT结果实现了患者外上髁处髁间线曲率的自动测量。这种方法对于人体结构的特征统计也具有临床意义。During a large number of ACL reconstruction surgeries, researchers have noticed the differences in the curvature of the intercondylar line at the lateral epicondyle of the femur among different populations and have conducted research on this. This application, based on the statistical analysis of a large amount of data, proposes a method for measuring the intercondylar line of the knee joint, and classifies different curvature structures, and customizes standards to solve this problem. It is worth noting that the new classification results are obtained under a new research method. The intelligent measurement system developed in this application can automatically and intelligently identify the near-lateral condylar point of the intercondylar line at the lateral epicondyle of the femur and calculate the curvature of the site, and uses the knee joint CT results to realize the automatic measurement of the curvature of the intercondylar line at the lateral epicondyle of the patient. This method is also of clinical significance for the characteristic statistics of human body structure.
发明内容Summary of the invention
本发明旨在提供一种人群股骨外上髁处髁间线曲率测算系统和方法，所要解决的技术问题至少包括前交叉韧带重建手术中股骨止点定位器不匹配的临床实际问题。The present invention aims to provide a system and method for calculating the curvature of the intercondylar line at the lateral epicondyle of the femur of a population, and the technical problems to be solved at least include the clinical practical problem of mismatch of femoral end point locators in anterior cruciate ligament reconstruction surgery.
为了实现上述目的，本发明提供一种人群股骨外上髁处髁间线曲率测算系统，包括结构分割单元、区域识别单元和曲率计算单元，In order to achieve the above object, the present invention provides a system for calculating the curvature of the intercondylar line at the lateral epicondyle of the femur of a group of people, comprising a structure segmentation unit, a region recognition unit and a curvature calculation unit.
所述的结构分割单元用于从原始CT影像中分割出股骨部分，利用语义分割网络U-net，使用深度学习技术搭建智能分割模型，从CT影像中分割股骨部分，以实现以IOU（Intersection over Union）指标>80%为标准的股骨影像的分割工作；The structure segmentation unit is used to segment the femur from the original CT image, and uses the semantic segmentation network U-net and deep learning technology to build an intelligent segmentation model to segment the femur from the CT image, so as to achieve the segmentation of the femur image with an IOU (Intersection over Union) index > 80% as the standard;
所述的区域识别单元用于从分离的股骨模型中识别出髁间线近外侧髁位点；The region identification unit is used to identify the intercondylar line proximal lateral condylar site from the separated femoral model;
所述的曲率计算单元用于对识别出的位点区域划分曲线簇并计算曲率。The curvature calculation unit is used to divide the identified site area into curve clusters and calculate the curvature.
优选地，所述的结构分割单元用于从原始CT影像中分割出股骨部分，利用语义分割网络U-net，使用深度学习技术搭建智能分割模型，从CT影像中分割股骨部分，以实现以IOU指标>80%为标准的1000套股骨影像的分割工作。Preferably, the structure segmentation unit is used to segment the femur part from the original CT image, using the semantic segmentation network U-net and deep learning technology to build an intelligent segmentation model to segment the femur part from the CT image, so as to achieve the segmentation of 1000 sets of femur images with IOU index>80% as the standard.
优选地，所述的区域识别单元采用基于深度学习的点云处理模型PointNet来从分离的股骨模型中识别出髁间线近外侧髁位点。Preferably, the region identification unit uses a deep learning-based point cloud processing model PointNet to identify the proximal lateral condylar site of the intercondylar line from the separated femoral model.
优选地，所述的区域识别单元从多个分离的股骨模型中识别出髁间线近外侧髁位点的具体方法为：首先对预定数量的股骨CT样本进行人工的结构分割，依次分割并标注出股骨外侧髁、腘面和髁间窝，然后训练深度学习模型对标注出的股骨外侧髁、腘面和髁间窝进行点云识别；在具体的分割标注过程中，首先将股骨远端外侧区域与股骨远端外侧区域以外的部分分割开来，股骨远端外侧区域以外的股骨结构标记为“区域0”；剩余的标注均在股骨远端外侧区域内进行，股骨外侧髁标记为“区域2”，股骨远端外侧与腘面重叠部分标记为“区域3”，髁间窝近外侧髁面标记为“区域4”，其余部分标记为“区域1”；由此将股骨分为五个区域，获取区域2、区域3和区域4的模糊的交叉边界区，即股骨髁间线近外侧髁点位，位于外侧髁、髁间窝和腘面三者边界的交界处的区域；随后使用预定数量的分割标注的数据对机器学习模型进行学习和训练。Preferably, the specific method for the region identification unit to identify the intercondylar line proximal lateral condylar site from a plurality of separated femoral models is: firstly, manually segmenting a predetermined number of femoral CT samples, segmenting and marking the lateral femoral condyle, popliteal surface and intercondylar fossa in sequence, and then training a deep learning model to perform point cloud recognition on the marked lateral femoral condyle, popliteal surface and intercondylar fossa; in the specific segmentation and marking process, firstly, the distal lateral region of the femur is segmented from the part outside the distal lateral region of the femur, and the femoral structure outside the distal lateral region of the femur is marked as "region 0"; The remaining annotations are all performed in the lateral area of the distal femur. The lateral femoral condyle is marked as "Area 2", the overlapping part of the lateral distal femur and the popliteal surface is marked as "Area 3", the intercondylar fossa near the lateral condylar surface is marked as "Area 4", and the rest is marked as "Area 1"; the femur is divided into five regions, and the fuzzy intersection boundary area of Region 2, Region 3 and Region 4 is obtained, that is, the position of the lateral condyle near the intercondylar line of the femur, the area located at the junction of the boundaries of the lateral condyle, intercondylar fossa and popliteal surface; then a predetermined amount of segmented and labeled data is used to learn and train the machine learning model.
优选地，所述的区域识别单元从分离的股骨模型中识别出髁间线近外侧髁位点的具体方法中，首先对大样本量的股骨CT样本进行人工的结构分割。Preferably, in the specific method in which the region identification unit identifies the intercondylar line proximal lateral condylar site from the separated femoral model, firstly, artificial structure segmentation is performed on a large sample size of femoral CT samples.
优选地，所述的区域识别单元从多个分离的股骨模型中识别出髁间线近外侧髁位点的具体方法中，使用大样本量分割标注的数据对机器学习模型进行学习和训练。Preferably, in the specific method in which the region identification unit identifies the proximal lateral condylar site of the intercondylar line from a plurality of separated femoral models, a large sample size of segmented and labeled data is used to learn and train the machine learning model.
优选地，所述的区域识别单元分为两步对区域0、区域1、区域2、区域3和区域4这五个区域进行识别；首先识别并区分“区域0”，随后将“区域1、区域2、区域3和区域4”直接输入深度学习模型进行识别与区分。Preferably, the region identification unit identifies five regions, region 0, region 1, region 2, region 3 and region 4, in two steps; first, "region 0" is identified and distinguished, and then "region 1, region 2, region 3 and region 4" are directly input into the deep learning model for identification and distinction.
优选地，所述的曲率计算单元计算曲率的具体方法为：首先取股骨髁间线近外侧髁点位的中间1/3部分，做十等分的法向量切面，得到十条生理曲线；随后对所有的十条生理曲线进行拟合，输出沿着目标点位的髁间线的各个曲线的曲率，计算出十条生理曲线的曲率的均方根平均值。Preferably, the curvature calculation unit calculates the curvature in the following specific method: first, take the middle 1/3 of the intercondylar line of the femur near the lateral condyle point, divide the normal vector section into ten equal parts, and obtain ten physiological curves; then, fit all ten physiological curves, output the curvature of each curve along the intercondylar line of the target point, and calculate the root mean square average of the curvatures of the ten physiological curves.
优选地，生理曲线的曲率的具体计算方法为：Preferably, the specific calculation method of the curvature of the physiological curve is:
假设生理曲线上的一个控制点的坐标为（x,y,z），该控制点从生理曲线的一端端点向另一端的端点运动的过程中，坐标（x,y,z）均为时间t的连续可导函数，假设生理曲线的参数方程为：Assume that the coordinates of a control point on the physiological curve are (x, y, z). When the control point moves from one end point of the physiological curve to the other end point, the coordinates (x, y, z) are all continuous differentiable functions of time t. Assume that the parametric equation of the physiological curve is:
x=x(t)；x=x(t);
y=y(t)；y=y(t);
z=z(t)；z=z(t);
对求一阶导数，得到：right Taking the first-order derivative, we get:
对求二阶导数，得到：right Taking the second-order derivative, we get:
生理曲线的曲率半径的计算公式为：The calculation formula of the curvature radius of the physiological curve is:
； ;
其中：in:
； ;
将v和a代入生理曲线的曲率半径的计算公式，整理后得到：Substituting v and a into the calculation formula of the curvature radius of the physiological curve, we get:
。 .
本发明还提供一种人群股骨外上髁处髁间线曲率测算系统的测算方法，包括以下步骤：The present invention also provides a method for calculating the curvature of the intercondylar line at the lateral epicondyle of a group of people, comprising the following steps:
第一步，对患者膝关节进行三维CT扫描，获取DICOM格式的原始影像数据；The first step is to perform a three-dimensional CT scan of the patient's knee joint to obtain the original image data in DICOM format;
第二步，从原始CT影像数据中分割出股骨部分，从分离的股骨模型中识别出髁间线近外侧髁位点，对识别出的位点区域划分曲线簇并计算曲率；The second step is to segment the femur from the original CT image data, identify the intercondylar line near the lateral condyle from the separated femoral model, divide the identified site area into curve clusters and calculate the curvature;
第三步，通过髁间线近外侧髁点位曲率的相关信息，得出相应的曲率分布区间，依此进行曲率特点的分型，从而指导设计出与解剖结构更匹配、定位更加准确牢固的定位器。The third step is to obtain the corresponding curvature distribution range through the relevant information of the curvature of the intercondylar line near the lateral condyle point, and classify the curvature characteristics based on this, so as to guide the design of a locator that better matches the anatomical structure and has more accurate and firm positioning.
优选地，所述的从原始CT影像数据中分割出股骨部分，具体包括：从多套原始CT影像中分割出股骨部分，利用语义分割网络U-net，使用深度学习技术搭建智能分割模型，从CT影像中分割股骨部分，以实现以>80%IOU（Intersection over Union）指标为标准的股骨影像的分割工作。Preferably, the segmenting of the femur part from the original CT image data specifically includes: segmenting the femur part from multiple sets of original CT images, using the semantic segmentation network U-net, using deep learning technology to build an intelligent segmentation model, and segmenting the femur part from the CT image to achieve the segmentation of the femur image with a standard of >80% IOU (Intersection over Union) indicator.
优选地，所述的从原始CT影像数据中分割出股骨部分，具体包括：从大样本量的原始CT影像中分割出股骨部分，利用语义分割网络U-net，使用深度学习技术搭建智能分割模型，从CT影像中分割股骨部分，以实现以>80%IOU指标为标准的1000套股骨影像的分割工作。Preferably, the segmenting of the femur from the original CT image data specifically includes: segmenting the femur from a large sample of original CT images, using the semantic segmentation network U-net, using deep learning technology to build an intelligent segmentation model, and segmenting the femur from the CT image to achieve the segmentation of 1000 sets of femur images with a standard of >80% IOU indicator.
优选地，所述的从分离的股骨模型中识别出髁间线近外侧髁位点，具体包括：采用基于深度学习的点云处理模型PointNet来从多个分离的股骨模型中识别出髁间线近外侧髁位点。Preferably, the step of identifying the proximal lateral condyle site of the intercondylar line from the separated femoral models specifically includes: using a deep learning-based point cloud processing model PointNet to identify the proximal lateral condyle site of the intercondylar line from multiple separated femoral models.
优选地，所述的从分离的股骨模型中识别出髁间线近外侧髁位点的具体方法为：首先对预定数量的股骨CT样本进行人工的结构分割，依次分割并标注出股骨外侧髁、腘面和髁间窝，然后训练深度学习模型对标注出的股骨外侧髁、腘面和髁间窝进行点云识别；在具体的分割标注过程中，首先将股骨远端外侧区域与股骨远端外侧区域以外的部分分割开来，股骨远端外侧区域以外的股骨结构标记为“区域0”；剩余的标注均在股骨远端外侧区域内进行，股骨外侧髁标记为“区域2”，股骨远端外侧与腘面重叠部分标记为“区域3”，髁间窝近外侧髁面标记为“区域4”，其余部分标记为“区域1”；由此将股骨分为五个区域，获取区域2、区域3和区域4的模糊的交叉边界区，即股骨髁间线近外侧髁点位，位于外侧髁、髁间窝和腘面三者边界的交界处的区域；随后使用预定数量的分割标注的数据对机器学习模型进行学习和训练。Preferably, the specific method for identifying the lateral condyle site near the intercondylar line from the separated femoral model is: first, manually segmenting a predetermined number of femoral CT samples, segmenting and marking the lateral femoral condyle, popliteal surface and intercondylar fossa in turn, and then training a deep learning model to perform point cloud recognition on the marked lateral femoral condyle, popliteal surface and intercondylar fossa; in the specific segmentation and marking process, firstly, the lateral region of the distal femur is segmented from the part outside the lateral region of the distal femur, and the femoral structure outside the lateral region of the distal femur is marked as "region 0"; the remaining marked The annotations were all performed in the lateral region of the distal femur, with the lateral femoral condyle marked as "region 2", the overlapping part of the distal femoral lateral side and the popliteal surface marked as "region 3", the intercondylar fossa near the lateral condylar surface marked as "region 4", and the rest marked as "region 1". The femur was divided into five regions, and the fuzzy intersection boundary area of regions 2, 3, and 4 was obtained, that is, the intercondylar line of the femur near the lateral condyle point, the region located at the junction of the boundaries of the lateral condyle, intercondylar fossa, and popliteal surface. Subsequently, a predetermined number of segmented and labeled data were used to learn and train the machine learning model.
优选地，所述的从分离的股骨模型中识别出髁间线近外侧髁位点的具体方法中，首先对预定数量的股骨CT样本进行人工的结构分割，Preferably, in the specific method for identifying the intercondylar line proximal lateral condylar site from the separated femoral model, firstly, a predetermined number of femoral CT samples are manually segmented.
优选地，所述的从分离的股骨模型中识别出髁间线近外侧髁位点的具体方法中，使用大样本量分割标注的数据对机器学习模型进行学习和训练。Preferably, in the specific method of identifying the proximal lateral condylar site of the intercondylar line from the separated femoral model, a large sample size of segmented and labeled data is used to learn and train the machine learning model.
优选地，所述的从分离的股骨模型中识别出髁间线近外侧髁位点的具体方法中，分为两步对区域0、区域1、区域2、区域3和区域4这五个区域进行识别；首先识别并区分“区域0”，随后将“区域1、区域2、区域3和区域4”直接输入深度学习模型进行识别与区分。Preferably, in the specific method of identifying the proximal lateral condylar site of the intercondylar line from the separated femoral model, the five regions of region 0, region 1, region 2, region 3 and region 4 are identified in two steps; first, "region 0" is identified and distinguished, and then "region 1, region 2, region 3 and region 4" are directly input into the deep learning model for identification and distinction.
优选地，所述的对识别出的位点区域划分曲线簇并计算曲率的具体方法为：首先取股骨髁间线近外侧髁点位的中间1/3部分，做十等分的法向量切面，得到十条生理曲线；随后对所有的十条生理曲线进行拟合，输出沿着目标点位的髁间线的各个曲线的曲率，计算出十条生理曲线的曲率的均方根平均值。Preferably, the specific method of dividing the identified site area into curve clusters and calculating the curvature is: first take the middle 1/3 of the femoral intercondylar line near the lateral condyle point, make a normal vector section divided into ten equal parts, and obtain ten physiological curves; then fit all ten physiological curves, output the curvature of each curve along the intercondylar line of the target point, and calculate the root mean square average of the curvatures of the ten physiological curves.
与现有技术相比，本发明的有益效果是：Compared with the prior art, the present invention has the following beneficial effects:
本发明所述的人群股骨外上髁处髁间线曲率测算系统能够对股骨外上髁处髁间线近外侧髁点位进行识别并计算出位点的曲率。本发明通过深度学习自动化分析大样本量股骨数据，计算并统计相应的髁间线所在法向量面位点曲率数据，依此进行曲率特点的分型，从而设计出与解剖结构更匹配、定位更加准确牢固的定位器，以解决股骨定位器无法与股骨定位点精准贴合并形成牢固固定的问题。本申请开发的智能测算系统能够对股骨外上髁处髁间线近外侧髁点位进行自动识别并计算出位点的曲率，利用膝关节三维CT重建结果实现了患者外上髁处髁间线曲率的自动测量。通过股骨自动分割、重点区域识别、曲面参数测算共三个步骤，实现了一键自动化计算曲率半径，输入DICOM格式CT数据，输出曲率半径的数值及3D可视化结果。这种方法对于人体结构的特征统计也具有临床意义。The curvature calculation system of the intercondylar line at the lateral epicondyle of the femur described in the present invention can identify the point of the intercondylar line near the lateral condyle of the femur and calculate the curvature of the site. The present invention automatically analyzes a large sample of femoral data through deep learning, calculates and statistics the curvature data of the normal vector surface site where the corresponding intercondylar line is located, and classifies the curvature characteristics accordingly, so as to design a locator that is more matched with the anatomical structure and more accurately and firmly positioned, so as to solve the problem that the femoral locator cannot accurately fit with the femoral positioning point and form a firm fixation. The intelligent measurement system developed in this application can automatically identify the point of the intercondylar line near the lateral condyle of the femur and calculate the curvature of the site, and realizes the automatic measurement of the curvature of the intercondylar line at the lateral epicondyle of the patient using the three-dimensional CT reconstruction results of the knee joint. Through the three steps of automatic segmentation of the femur, identification of key areas, and calculation of curved surface parameters, the one-key automatic calculation of the radius of curvature is realized, the DICOM format CT data is input, and the numerical value of the radius of curvature and the 3D visualization result are output. This method also has clinical significance for the characteristic statistics of human body structure.
附图说明BRIEF DESCRIPTION OF THE DRAWINGS
附图用来提供对本发明技术方案的进一步理解，并且构成说明书的一部分，与本申请的具体实施方式一起用于解释本发明的技术方案，并不构成对本发明技术方案的限制。The accompanying drawings are used to provide a further understanding of the technical solution of the present invention and constitute a part of the specification. Together with the specific implementation methods of the present application, they are used to explain the technical solution of the present invention and do not constitute a limitation on the technical solution of the present invention.
图1是本发明所述的人群股骨外上髁处髁间线曲率测算系统的结构示意图。FIG1 is a schematic structural diagram of a system for calculating the curvature of the intercondylar line at the lateral epicondyle of the femur of a human population according to the present invention.
具体实施方式Detailed ways
在下文中更详细地描述了本发明以有助于对本发明的理解。The present invention is described in more detail hereinafter to facilitate understanding of the present invention.
如图1所示，本发明所述的人群股骨外上髁处髁间线曲率测算系统包括结构分割单元、区域识别单元和曲率计算单元；As shown in FIG1 , the intercondylar line curvature measurement system at the lateral epicondyle of the femur of a human population of the present invention comprises a structure segmentation unit, a region identification unit and a curvature calculation unit;
所述的结构分割单元用于从大样本量的原始CT影像中分割出股骨部分，利用语义分割网络U-net，使用深度学习技术搭建智能分割模型，从CT影像中分割股骨部分，以实现以IOU（Intersection over Union）指标>80%为标准的股骨影像的分割工作；The structure segmentation unit is used to segment the femur from the original CT images with a large sample size, and uses the semantic segmentation network U-net and deep learning technology to build an intelligent segmentation model to segment the femur from the CT images, so as to achieve the segmentation of the femur image with an IOU (Intersection over Union) index > 80% as the standard;
所述的区域识别单元用于从分离的股骨模型中识别出髁间线近外侧髁位点；该位点处于多个区域的模糊边界，边界较难界定。同时，该位点处曲面在各个人身上存在差异性，传统计算方法难以识别该位点，也难以泛化相应的识别规则。针对该位点的识别，目前没有可靠的先例可供参考。通过系统性分析，本申请采用基于深度学习的点云处理模型PointNet来解决这一难题。首先对预定数量（例如200套）的股骨CT样本进行了人工的结构分割，依次分割并标注出股骨外侧髁、腘面、髁间窝等部位，然后训练深度学习模型对这些部位进行点云识别。在具体的分割标注过程中，首先将股骨远端外侧区域与其余部分分割开来，股骨远端外侧的其余股骨结构标记为“区域0”。剩余的标注均在股骨远端外侧区域内进行，股骨外侧髁标记为“区域2”，股骨远端外侧与腘面重叠部分（腘面外侧半）标记为“区域3”，髁间窝近外侧髁面标记为“区域4”，其余部分标记为“区域1”。由此将股骨分为五个区域，获取区域2/3/4的模糊的交叉边界区，即股骨髁间线近外侧髁点位，位于外侧髁、髁间窝和腘面三者边界的交界处的区域。随后使用预定数量（例如200套）的分割标注的数据对机器学习模型进行学习和训练。在最初的训练中，本申请将五个区域直接输入深度学习模型，一次性学习识别五个区域，得到的识别效果不甚理想，出现了区域杂乱的情况，无法达到区域分割及识别位点的需求。为了优化区域识别效果，本申请采用了新的方法，分为两步进行五个区域的识别。首先识别并区分“区域0”与“区域1234”，随后将“区域1234”直接输入深度学习模型进行识别与区分。相较于一次性识别五个区域，改进后的训练方法可以较为准确地区分不同区域，能够满足本申请位点识别的需求。The region recognition unit is used to identify the intercondylar line near the lateral condyle site from the separated femoral model; this site is at the fuzzy boundary of multiple regions, and the boundary is difficult to define. At the same time, the surface at this site is different in each person, and it is difficult for traditional computing methods to identify this site and generalize the corresponding recognition rules. There is currently no reliable precedent for the identification of this site. Through systematic analysis, this application adopts a point cloud processing model PointNet based on deep learning to solve this problem. First, a predetermined number (for example, 200 sets) of femoral CT samples are manually segmented, and the lateral femoral condyle, popliteal surface, intercondylar fossa and other parts are segmented and marked in turn, and then the deep learning model is trained to perform point cloud recognition on these parts. In the specific segmentation and annotation process, the lateral region of the distal femur is first separated from the rest, and the remaining femoral structures on the lateral side of the distal femur are marked as "region 0". The remaining annotations are all performed in the lateral area of the distal femur, with the lateral condyle of the femur marked as "area 2", the overlapping part of the lateral side of the distal femur and the popliteal surface (lateral half of the popliteal surface) marked as "area 3", the intercondylar fossa near the lateral condylar surface marked as "area 4", and the rest marked as "area 1". The femur is thus divided into five regions, and the fuzzy cross-border area of region 2/3/4 is obtained, that is, the intercondylar line of the femur near the lateral condyle point, and the area at the junction of the boundaries of the lateral condyle, intercondylar fossa and popliteal surface. Subsequently, a predetermined number (for example, 200 sets) of segmented and annotated data are used to learn and train the machine learning model. In the initial training, this application directly inputs the five regions into the deep learning model, learns and recognizes the five regions at one time, and the recognition effect obtained is not ideal, and the region is messy, which cannot meet the requirements of regional segmentation and identification sites. In order to optimize the regional recognition effect, this application adopts a new method to identify the five regions in two steps. First, "Region 0" and "Region 1234" are identified and distinguished, and then "Region 1234" is directly input into the deep learning model for identification and distinction. Compared with identifying five regions at one time, the improved training method can distinguish different regions more accurately and meet the needs of site identification in this application.
所述的曲率计算单元用于对识别出的位点区域划分曲线簇并计算曲率，通过股骨分割与股骨髁间线近外侧髁点位分割，切割出的区域将较为完好的呈现。首先取股骨髁间线近外侧髁点位的中间1/3部分，对其做十等分的法向量切面，得到十条生理曲线。随后对所有生理曲线进行拟合，输出沿着目标点位的髁间线的各个曲线的曲率，计算出十条生理曲线的曲率的均方根平均值。The curvature calculation unit is used to divide the identified site area into curve clusters and calculate the curvature. Through the femoral segmentation and the segmentation of the femoral intercondylar line near the lateral condyle point, the cut-out area will be presented more intact. First, take the middle 1/3 part of the femoral intercondylar line near the lateral condyle point, divide it into ten equal parts, and obtain ten physiological curves. Then fit all physiological curves, output the curvature of each curve along the intercondylar line of the target point, and calculate the root mean square average of the curvature of the ten physiological curves.
优选地，生理曲线的曲率的具体计算方法为：Preferably, the specific calculation method of the curvature of the physiological curve is:
假设生理曲线上的一个控制点的坐标为（x,y,z），该控制点从生理曲线的一端端点向另一端的端点运动的过程中，坐标（x,y,z）均为时间t的连续可导函数，假设生理曲线的参数方程为：Assume that the coordinates of a control point on the physiological curve are (x, y, z). When the control point moves from one end point of the physiological curve to the other end point, the coordinates (x, y, z) are all continuous differentiable functions of time t. Assume that the parametric equation of the physiological curve is:
x=x(t)；x=x(t);
y=y(t)；y=y(t);
z=z(t)；z=z(t);
对求一阶导数，得到：right Taking the first-order derivative, we get:
对求二阶导数，得到：right Taking the second-order derivative, we get:
生理曲线的曲率半径的计算公式为：The calculation formula of the curvature radius of the physiological curve is:
； ;
其中：in:
将v和a代入生理曲线的曲率半径的计算公式，整理后得到：Substituting v and a into the calculation formula of the curvature radius of the physiological curve, we get:
。 .
优选地，所述的结构分割单元用于从大样本量的原始CT影像中分割出股骨部分，利用语义分割网络U-net，使用深度学习技术搭建智能分割模型，从CT影像中分割股骨部分，以实现以IOU（Intersection over Union）指标>80%为标准的1000套股骨影像的分割工作。Preferably, the structure segmentation unit is used to segment the femur part from the original CT images with a large sample size, and use the semantic segmentation network U-net and deep learning technology to build an intelligent segmentation model to segment the femur part from the CT images to achieve the segmentation of 1,000 sets of femur images with an IOU (Intersection over Union) index > 80% as the standard.
IOU（Intersection over Union）指标是指预测值与实际值的交集面积除以它们的并集面积。IOU指标常用于目标检测和图像分割等任务中，用于衡量算法对目标位置或区域的准确度。具体来说，IOU指标可以用来衡量预测的目标与真实目标之间的重叠程度，值越大表示两者的重叠程度越高，准确度越高。The IOU (Intersection over Union) metric refers to the intersection area of the predicted value and the actual value divided by their union area. The IOU metric is often used in tasks such as target detection and image segmentation to measure the accuracy of the algorithm for the target location or area. Specifically, the IOU metric can be used to measure the degree of overlap between the predicted target and the real target. The larger the value, the higher the overlap between the two and the higher the accuracy.
优选地，所述的区域识别单元用于从分离的股骨模型中识别出髁间线近外侧髁位点。Preferably, the region identification unit is used to identify the proximal lateral condyle site of the intercondylar line from the separated femoral model.
本发明还提供一种人群股骨外上髁处髁间线曲率测算系统的测算方法，包括以下步骤：The present invention also provides a method for calculating the curvature of the intercondylar line at the lateral epicondyle of a group of people, comprising the following steps:
第一步，对患者膝关节进行三维CT扫描，获取DICOM格式原始影像数据；The first step is to perform a three-dimensional CT scan of the patient's knee joint to obtain the original image data in DICOM format;
第二步，从原始CT影像数据中分割出股骨部分，从分离的股骨模型中识别出髁间线近外侧髁位点，对识别出的位点区域划分曲线簇并计算曲率；The second step is to segment the femur from the original CT image data, identify the intercondylar line near the lateral condyle from the separated femoral model, divide the identified site area into curve clusters and calculate the curvature;
第三步，通过髁间线近外侧髁点位曲率的相关信息，得出相应的曲率分布区间，依此进行曲率特点的分型，从而指导设计出与解剖结构更匹配、定位更加准确牢固的定位器，以解决股骨定位器无法与股骨定位点精准贴合并形成牢固固定的问题，由此指导设计的定位器有益于前交叉韧带重建手术中股骨定位的稳定操作。The third step is to obtain the corresponding curvature distribution range through the relevant information of the curvature of the proximal lateral condyle point of the intercondylar line, and classify the curvature characteristics accordingly, so as to guide the design of a locator that better matches the anatomical structure and has more accurate and firm positioning, so as to solve the problem that the femoral locator cannot accurately fit with the femoral positioning point and form a firm fixation. The locator designed with this guidance is beneficial to the stable operation of femoral positioning in anterior cruciate ligament reconstruction surgery.
在一个具体实施例中，所述的测算方法包括以下步骤：In a specific embodiment, the calculation method comprises the following steps:
S1、从原始CT影像中分割出股骨部分；S1, segment the femur from the original CT image;
S2、从分离的股骨模型中识别出髁间线近外侧髁位点；S2, identify the intercondylar line proximal to the lateral condyle site from the separated femoral model;
S3、对识别出的位点区域划分曲线簇并计算曲率。S3. Divide the identified site area into curve clusters and calculate the curvature.
本申请利用此方法自动化分析股骨数据，计算并统计相应的髁间线两侧曲率数据，依此进行曲率特点的分型，从而设计出与解剖结构更匹配、定位更加准确牢固的定位器，以解决股骨定位器无法与股骨定位点精准对接并形成牢固固定的问题。本申请提出的智能测算方法能够对股骨外上髁处髁间线近外侧髁点位进行自动识别并计算出位点的曲率，利用CT结果实现了患者外上髁处髁间线曲率的自动测量。通过股骨自动分割、重点区域识别、曲面参数测算共三个步骤，实现了一键自动化计算曲率半径，输入DICOM格式CT数据，输出曲率半径的数值及3D可视化结果。这种方法对于人体结构的特征统计也具有临床意义。This application uses this method to automatically analyze femoral data, calculate and count the corresponding curvature data on both sides of the intercondylar line, and classify the curvature characteristics accordingly, so as to design a locator that is more compatible with the anatomical structure and more accurately and firmly positioned, so as to solve the problem that the femoral locator cannot accurately dock with the femoral positioning point and form a firm fixation. The intelligent measurement method proposed in this application can automatically identify the near-lateral condylar point of the intercondylar line at the lateral epicondyle of the femur and calculate the curvature of the site, and realize the automatic measurement of the curvature of the intercondylar line at the lateral epicondyle of the patient using CT results. Through three steps of automatic femoral segmentation, key area identification, and surface parameter calculation, one-click automatic calculation of the radius of curvature is realized, DICOM format CT data is input, and the numerical value of the radius of curvature and 3D visualization results are output. This method is also of clinical significance for the characteristic statistics of human structure.
在一个具体实施例中，本申请通过股骨自动分割、重点区域识别、曲面参数测算共三个步骤，实现了一键自动化计算曲率半径，输入DICOM格式CT数据，输出曲率半径的数值及3D可视化结果。In a specific embodiment, the present application realizes one-click automatic calculation of the curvature radius through three steps: automatic femur segmentation, key area identification, and surface parameter measurement, inputs DICOM format CT data, and outputs the numerical value of the curvature radius and 3D visualization results.
第一步、股骨自动分割：Step 1: Automatic segmentation of femur:
（1）实现方法(1) Implementation method
基于深度学习U-Net网络，从Dicom格式的CT文件中，自动分割出股骨3D模型。Based on the deep learning U-Net network, the femur 3D model is automatically segmented from the Dicom format CT file.
（2）验证策略(2) Verification strategy
定量评价：Quantitative evaluation:
计算交并比（Intersection over Union，IoU）Calculate Intersection over Union (IoU)
IoU = 自动分割和金标准股骨体素点的交集/自动分割和金标准股骨体素点的并集。IoU = Intersection over Union of automatic segmentation and gold standard femur voxel points/Intersection over Union of automatic segmentation and gold standard femur voxel points.
经验证，本发明提出的方法的交并比达95%以上，证明本发明提出的方法在股骨区域的自动分割方面精确度较高。It has been verified that the intersection-over-union ratio of the method proposed in the present invention is above 95%, which proves that the method proposed in the present invention has high accuracy in automatic segmentation of the femur region.
定性评价：对自动分割结果进行形态学分析，通过3D slicer等医学图像查看软件，将分割结果叠加覆盖至CT图像，查看分割的准确性，尤其是骨骼边缘是否对齐。Qualitative evaluation: Perform morphological analysis on the automatic segmentation results, and overlay the segmentation results on the CT image through medical image viewing software such as 3D slicer to check the accuracy of the segmentation, especially whether the bone edges are aligned.
第二步、外侧髁、腘面、髁间窝区域自动识别：Step 2: Automatically identify the lateral condyle, popliteal surface, and intercondylar fossa:
第一步的自动分割为第二步的外侧髁、腘面、髁间窝区域自动识别提供了可靠的基础。The automatic segmentation in the first step provides a reliable basis for the automatic identification of the lateral condyle, popliteal surface, and intercondylar fossa in the second step.
（1）实现方法(1) Implementation method
基于深度学习PointNet网络，从第一步分割出的股骨3D模型中，自动识别外侧髁、腘面、髁间线等重点区域。Based on the deep learning PointNet network, key areas such as the lateral condyle, popliteal surface, and intercondylar line are automatically identified from the femoral 3D model segmented in the first step.
（2）验证策略(2) Verification strategy
定量评价：计算交并比（IoU）Quantitative evaluation: calculating the intersection over union (IoU)
IoU = 区域自动识别和金标准区域的交集/区域自动识别和金标准区域的并集。IoU = Intersection over Union of automatically identified regions and gold standard regions / Intersection over Union of automatically identified regions and gold standard regions.
本发明提出的方法的交并比达89%以上，该方法在重点区域的识别方面精确度较高。The intersection-over-union ratio of the method proposed in the present invention is over 89%, and the method has high accuracy in identifying key areas.
定性评价：通过CloudCompare等3D点云查看软件，查看区域分割的准确性。Qualitative evaluation: Use 3D point cloud viewing software such as CloudCompare to check the accuracy of region segmentation.
第三步、曲面参数测算：Step 3: Calculation of surface parameters:
（1）实现方法(1) Implementation method
通过如下7个计算步骤，识别髁间线、拟合髁间线、创建截取平面、拟合截线、计算曲率半径、识别髁间线位置、识别髁间线与腘面相交位置，实现了曲面参数测算。其中，曲率半径根据拟合的截线计算得到。The following seven calculation steps are used to calculate the surface parameters: identifying the intercondylar line, fitting the intercondylar line, creating a cutting plane, fitting the cutting line, calculating the radius of curvature, identifying the position of the intercondylar line, and identifying the intersection of the intercondylar line and the popliteal surface. The radius of curvature is calculated based on the fitted cutting line.
（2）验证策略(2) Verification strategy
计算完毕后，自动生成“计算结果3D可视化.html”文件，使用浏览器打开此文件，查看截面位置的准确性和每条截线的曲率半径数值。After the calculation is completed, the "Calculation Results 3D Visualization.html" file is automatically generated. Use a browser to open this file to check the accuracy of the section position and the curvature radius value of each section line.
如果截面位置正确，则曲率半径计算结果正确，因为曲率半径基于截线拟合计算，且计算方法遵循严格的数学公式。If the section position is correct, the curvature radius calculation result is correct, because the curvature radius is calculated based on the cross-section fitting and the calculation method follows a strict mathematical formula.
这个验证策略不仅提供了定性评价，同时也为医生提供了直观的数据支持，使得医生在手术前期能够更好地了解患者的骨骼结构，为手术方案的制定提供了重要参考。This verification strategy not only provides qualitative evaluation, but also provides doctors with intuitive data support, enabling them to better understand the patient's bone structure before surgery and providing an important reference for the formulation of surgical plans.
以上描述了本发明优选实施方式，然其并非用以限定本发明。本领域技术人员对在此公开的实施方案可进行并不偏离本发明范畴和精神的改进和变化。The above describes the preferred embodiments of the present invention, but it is not intended to limit the present invention. Those skilled in the art may make improvements and changes to the embodiments disclosed herein without departing from the scope and spirit of the present invention.
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410749487.7ACN118335297B (en) | 2024-06-12 | 2024-06-12 | System and method for calculating curvature of intercondylar line at lateral epicondyle of femur of a population |
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410749487.7ACN118335297B (en) | 2024-06-12 | 2024-06-12 | System and method for calculating curvature of intercondylar line at lateral epicondyle of femur of a population |
| Publication Number | Publication Date |
|---|---|
| CN118335297Atrue CN118335297A (en) | 2024-07-12 |
| CN118335297B CN118335297B (en) | 2024-09-06 |
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410749487.7AActiveCN118335297B (en) | 2024-06-12 | 2024-06-12 | System and method for calculating curvature of intercondylar line at lateral epicondyle of femur of a population |
| Country | Link |
|---|---|
| CN (1) | CN118335297B (en) |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120323335A1 (en)* | 2011-06-16 | 2012-12-20 | Zimmer, Inc. | Femoral component for a knee prosthesis with improved articular characteristics |
| US20150105698A1 (en)* | 2013-10-16 | 2015-04-16 | Somersault Orthopedics Inc. | Method for knee resection alignment approximation in knee replacement procedures |
| AU2020101836A4 (en)* | 2020-08-14 | 2020-09-24 | Xi'an university of posts and telecommunications | A method for generating femoral x-ray films based on deep learning and digital reconstruction of radiological image |
| CN113409301A (en)* | 2021-07-12 | 2021-09-17 | 上海精劢医疗科技有限公司 | Point cloud segmentation-based femoral neck registration method, system and medium |
| CN117689891A (en)* | 2024-01-17 | 2024-03-12 | 中国医学科学院北京协和医院 | Method for segmenting and classifying a target lumbar vertebral structure using a two-stage model |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120323335A1 (en)* | 2011-06-16 | 2012-12-20 | Zimmer, Inc. | Femoral component for a knee prosthesis with improved articular characteristics |
| US20150105698A1 (en)* | 2013-10-16 | 2015-04-16 | Somersault Orthopedics Inc. | Method for knee resection alignment approximation in knee replacement procedures |
| AU2020101836A4 (en)* | 2020-08-14 | 2020-09-24 | Xi'an university of posts and telecommunications | A method for generating femoral x-ray films based on deep learning and digital reconstruction of radiological image |
| CN113409301A (en)* | 2021-07-12 | 2021-09-17 | 上海精劢医疗科技有限公司 | Point cloud segmentation-based femoral neck registration method, system and medium |
| CN117689891A (en)* | 2024-01-17 | 2024-03-12 | 中国医学科学院北京协和医院 | Method for segmenting and classifying a target lumbar vertebral structure using a two-stage model |
| Title |
|---|
| 孙月芳 等: "股骨远端及后髁曲率半径的性别差异及其意义", 《吉林大学学报（医学版）》, no. 1, 28 January 2010 (2010-01-28), pages 160* |
| 陈笠, 邓倩妮, 王成焘: "不同插值算法应用于骨科三维造型的适应性研究", 医用生物力学, no. 02, 30 June 2002 (2002-06-30)* |
| Publication number | Publication date |
|---|---|
| CN118335297B (en) | 2024-09-06 |
| Publication | Publication Date | Title |
|---|---|---|
| CN111529063B (en) | Operation navigation system and method based on three-dimensional reconstruction multi-mode fusion | |
| CN114419618B (en) | Preoperative planning system for total hip replacement based on deep learning | |
| US9842394B2 (en) | Detection of anatomical landmarks | |
| WO2022142741A1 (en) | Total knee arthroplasty preoperative planning method and device | |
| US11410769B1 (en) | Tactile solutions integration for patient specific treatment | |
| CN107563383A (en) | A kind of medical image auxiliary diagnosis and semi-supervised sample generation system | |
| US11883219B2 (en) | Artificial intelligence intra-operative surgical guidance system and method of use | |
| CN107847275A (en) | For preoperative planning positioning the system and method for soft tissue | |
| Polizzi et al. | Automatic cephalometric landmark identification with artificial intelligence: An umbrella review of systematic reviews | |
| CN109727251A (en) | The system that lung conditions are divided a kind of quantitatively, method and apparatus | |
| CN108309334A (en) | A kind of data processing method of spine X-ray image | |
| CN114711796A (en) | Double-lower-limb load potential line measuring and diagnosing system and intelligent measuring method | |
| CN117442334A (en) | A method based on deep learning to assist segment selection for adolescent idiopathic scoliosis surgery | |
| CN114596251B (en) | A knee joint X-ray image quality control method and device | |
| CN118335297B (en) | System and method for calculating curvature of intercondylar line at lateral epicondyle of femur of a population | |
| Liu et al. | Prior-based 3D U-Net: A model for knee-cartilage segmentation in MRI images | |
| CN118453111B (en) | Intraoperative knee joint gap adjusting device and knee joint operation equipment | |
| CN114693602A (en) | Knee joint dynamic tension balance assessment method and device | |
| CN110895818B (en) | A method and device for feature extraction of knee joint contours based on deep learning | |
| CN113781453B (en) | A kind of scoliosis progression prediction and method and device based on X-ray films | |
| Wang | [Retracted] Medical Imaging Diagnosis of Anterior Cruciate Ligament Injury Based on Intelligent Finite‐Element Algorithm | |
| CN114612400A (en) | Knee joint femoral replacement postoperative evaluation system based on deep learning | |
| Liu et al. | A personalized insertion centers preoperative positioning method for minimally invasive surgery of cruciate ligament reconstruction | |
| Nakatsu et al. | Automatic carpal site detection method for evaluation of rheumatoid arthritis using deep learning | |
| He et al. | Automatic quantification of morphology on magnetic resonance images of the proximal tibia |
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |