CN-119992109-B - Medical image key point detection method, device, equipment and medium

CN119992109BCN 119992109 BCN119992109 BCN 119992109BCN-119992109-B

Abstract

The invention discloses a medical image key point detection method, device, equipment and medium, which are based on bone anatomy priori information and a deep neural network and are used for detecting hip joint key points in medical CT images. According to the method, the common bone structure missing or damage situation in clinical image data is considered, and the key point detection of a bone defect part is realized by introducing anatomical prior information. And the existing common deep neural network structure for detection is improved, and the balance between the detection precision and the detection efficiency is realized. In addition, the method improves the loss function of the detection network, and uses the skeleton area mask to correct the predicted key point position, so that the position result of the key point prediction accords with the normal cognition.

Inventors

XIAO DEQIANG
FU TIANYU
ZHAO XIRUI
YANG JIAN
FAN JINGFAN
LIN YUCONG
AI DANNI
SHAO LONG
SONG HONG
WANG YUANYUAN

Assignees

北京理工大学

Dates

Publication Date: 20260512
Application Date: 20250120

Claims (7)

1. A medical image keypoint detection method, comprising: Acquiring a CT image to be processed, wherein the CT image to be processed is a CT image of a joint position to be detected of a patient; preprocessing the CT image to be processed to obtain a CT image of a joint to be detected, wherein impurities are removed from the CT image of the joint to be detected, and the impurities comprise artificial prosthesis implants or bone fragments; inputting the joint CT image to be detected into a key point position of the CT image to be processed obtained in a key point detection network model; the key point detection network model comprises a coarse detection network, an image slicing module and a fine detection network; The coarse detection network comprises a personalized key point distribution learning module and a local skeleton segmentation module, wherein the personalized key point distribution learning module is used for fusing general key point distribution information with the CT image characteristics of the joint to be detected to obtain personalized key point distribution information; The image slicing module is used for slicing the bone local area prediction mask to obtain a series of local image slices; the precise detection network is used for processing a series of input local image slices to obtain key point positions of the CT images to be processed, and comprises a local learning module and a global learning module, wherein the global learning module is used for capturing the relevance of the whole inside the slices and the relevance among the slices; The personalized key point distribution learning module learns and obtains the general key point distribution information from a key point distribution data set corresponding to a pre-established joint to be detected; The general key point distribution information is subjected to feature extraction through a convolutional neural network and then is fused with the CT image features of the joint to be detected to obtain the personalized key point distribution information; The general key point distribution information acquisition method comprises the steps of randomly selecting a key point coordinate distribution from the key point distribution data set as a template, aligning other key point distributions to the approximate position of the key point distribution by using a rigid registration method, constructing and obtaining a key point coordinate data set, constructing a piece of volume data for each key point distribution, wherein the size of each volume data is the same as that of a CT joint image after the subsequent pretreatment, assigning each voxel in the volume data to obtain the general key point distribution information, and the formula is as follows: In the formula, Voxel values representing voxel positions to be calculated, The coordinates of the voxels are represented, Is the first The coordinates of the key points of the individual joints, Representing the calculation of the euclidean distance, A logarithmic function based on e is represented.
2. The medical image keypoint detection method of claim 1, wherein the bone local region prediction mask comprises a bone region of a target range around each keypoint in the personalized keypoint distribution information.
3. The method of claim 1, wherein the slicing process includes centering a mask region center of the bone local region prediction mask as a sphere center with a radius Expansion outwards, finding optimisation The formula is as follows: In the formula, Is a minimum value of the radius, Representation search optimization Is a function of (a) and (b), Representing a function that calculates the number of given conditions, Representing voxel values in the three-dimensional image; Taking an circumscribed bounding box for the spherical region, and resampling image pixels in the circumscribed bounding box to a target size to obtain a series of local image slices.
4. The medical image keypoint detection method according to claim 1, wherein the network parameters are optimized by supervising a loss function in the course of the coarse detection network training, and the loss function is formulated as follows: In the formula, Calculating the segmentation loss between the local bone segmentation mask of the output result and the gold standard local bone segmentation mask, wherein the segmentation loss consists of soft price loss; Calculated is the difference in predicted values between the predicted region and the gold standard, consisting of MSE losses.
5. The medical image keypoint detection method according to claim 1, wherein the network parameters are optimized by supervising a loss function in the process of the fine detection network training, and the loss function is formulated as follows: In the formula, Representing a weight map based on the change of the iteration times, wherein the sum weights of the bone region and the non-bone region are different, and the added value of the two is 1; And (3) with The voxel values of the gold standard thermodynamic diagram and the voxel values of the predicted thermodynamic diagram are respectively represented.
6. A medical image keypoint detection device for performing the medical image keypoint detection method according to any one of claims 1 to 5, said device comprising: the image acquisition unit is used for acquiring a CT image to be processed, wherein the CT image to be processed is a CT image of a joint position to be detected of a patient; the image preprocessing unit is used for preprocessing the CT image to be processed to obtain a CT image of the joint to be detected, wherein the CT image of the joint to be detected is subjected to impurity removal, and the impurity comprises artificial prosthesis implants or bone fragments; the key point detection unit is used for inputting the joint CT image to be detected into the key point position of the CT image to be processed obtained in the key point detection network model; the key point detection network model comprises a coarse detection network, an image slicing module and a fine detection network; The coarse detection network comprises a personalized key point distribution learning module and a local skeleton segmentation module, wherein the personalized key point distribution learning module is used for fusing general key point distribution information with the CT image characteristics of the joint to be detected to obtain personalized key point distribution information; The image slicing module is used for slicing the bone local area prediction mask to obtain a series of local image slices; The fine detection network is used for processing a series of input local image slices to obtain key point positions of the CT images to be processed, and comprises a local learning module and a global learning module, wherein the global learning module is used for capturing the correlation of the whole inside the slices and the correlation among the slices.
7. A medical image keypoint detection device, the device comprising a processor and a memory: The memory is used for storing program codes and transmitting the program codes to the processor; the processor is configured to perform the medical image keypoint detection method of any one of claims 1-5 according to instructions in the program code.

Description

Medical image key point detection method, device, equipment and medium Technical Field The invention relates to the technical field of key point detection, in particular to a medical image key point detection method, device, equipment and medium based on anatomical prior information. Background In modern computer-assisted orthopaedic surgery, hip key points are critical to the planning of surgery in this modern form of surgery. In conventional clinical practice, hip key points are manually marked on a CT image by an expert, and the marking process consumes a great deal of time and labor cost. In addition, the manual marking of key points is highly dependent on experience and skills of experts, and the problem of large individual difference of marking results exists. Therefore, a high-efficiency and accurate automatic hip joint key point detection method is urgently needed clinically. However, because the resolution of the CT image of the hip joint is higher, a larger error exists in detecting key points on the image with high resolution, and because the clinical hip joint data most contain defects of bone structures, the missing bone structure information also greatly reduces the detection effect of an automatic method on the key points of the hip joint. Thus, accurate real-time detection of hip keypoints under bone defect conditions remains a challenge. Current methods for automatically detecting hip joint key points include a map set-based method and a learning-based method, but these methods have certain limitations. First, the detection effect of the atlas-based method is greatly affected by the constructed atlas, and the atlas has limited expression ability, so that it is difficult to express the distribution of features not in the atlas. Secondly, the method based on reinforcement learning is difficult to detect a plurality of target key points simultaneously due to the self intelligent training mode, and the application of the method in clinical scenes is greatly limited due to the low detection efficiency. The learning-based method has relatively better detection effect, and the currently commonly used method comprises a Mask RCNN-based method and a U-shaped network-based method, but the training parameters of the Mask RCNN method adopting an anchor frame are increased due to higher resolution of the CT image of the hip joint. In addition, most researchers use the network to directly learn the characteristics of the CT image to detect the key points, but neglect the correlation between the hip joint key points and bones, so that the detected key point distribution does not accord with objective rules, and the key point detection precision is lower. Disclosure of Invention In view of the foregoing, the present invention provides a medical image keypoint detection method, apparatus, device and medium for overcoming or at least partially solving the foregoing problems. The invention provides the following scheme: a medical image keypoint detection method comprising: Acquiring a CT image to be processed, wherein the CT image to be processed is a CT image of a joint position to be detected of a patient; preprocessing the CT image to be processed to obtain a CT image of a joint to be detected, wherein impurities are removed from the CT image of the joint to be detected, and the impurities comprise artificial prosthesis implants or bone fragments; inputting the joint CT image to be detected into a key point position of the CT image to be processed obtained in a key point detection network model; the key point detection network model comprises a coarse detection network, an image slicing module and a fine detection network; The coarse detection network comprises a personalized key point distribution learning module and a local skeleton segmentation module, wherein the personalized key point distribution learning module is used for fusing general key point distribution information with the CT image characteristics of the joint to be detected to obtain personalized key point distribution information; The image slicing module is used for slicing the bone local area prediction mask to obtain a series of local image slices; The fine detection network is used for processing a series of input local image slices to obtain key point positions of the CT images to be processed, and comprises a local learning module and a global learning module, wherein the global learning module is used for capturing the correlation of the whole inside the slices and the correlation among the slices. Preferably, the personalized key point distribution learning module learns and obtains the general key point distribution information from a key point distribution data set corresponding to a pre-established joint to be detected; And extracting features of the general joint key point distribution information to be detected through a convolutional neural network, and then fusing the feature extraction with the CT image features of the join