Method and device for measuring exophthalmos based on MRI images

Abstract

The application relates to the technical field of image processing and exophthalmos measurement, and provides an exophthalmos measurement method and device based on MRI images. The method comprises the following steps: acquiring an MRI transverse image of an eye socket; determining the positions of the most front edges of two zygomatic processes, the position of the eyeball gravity center and the position of the corneal front edge according to the MRI transverse image of the eye socket; calculating a first distance between the eyeball gravity center and the most front edges of the two zygomatic processes, and a second distance between the eyeball gravity center and the corneal front edge; and taking the sum of the first distance and the second distance as an exophthalmos measurement result. The application realizes exophthalmos measurement based on MRI images, and can improve the safety and accuracy of exophthalmos measurement.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing and exophthalmos measurement, and in particular to an exophthalmos measurement method and device based on MRI images. BACKGROUND

[0002] At present, exophthalmos (exophthalmos) refers to the vertical distance between the corneal vertex of the eyeball and the outer edge of the bilateral orbits, which is an important indicator for measuring the degree of eyeball protrusion. In related technologies, the measurement of exophthalmos is usually performed by using a Hertel exophthalmometer. However, this measurement method has some potential defects and limitations, for example, the repeatability and consistency of this measurement method are poor, and different operators using the Hertel exophthalmometer may obtain different measurement results, which is related to the technology and experience of the operators. In addition, the degree of cooperation of the patient during the examination, such as whether the patient can stably look straight ahead, may also affect the accuracy of the measurement.

[0003] Some literatures also provide a method for measuring exophthalmos based on CT (computed tomography) imaging examination. CT scanning uses X-ray technology, penetrates the human body through X-rays at different angles, and then receives signals by a detector, and generates cross-sectional images after computer processing. Compared with the Hertel exophthalmometer measurement method, this method improves the measurement accuracy and has good consistency.

[0004] However, due to the radiation characteristics of CT, its application in pregnant women and children is relatively limited. In addition, the contrast of CT in soft tissue imaging is relatively low, which may affect the accuracy of the measurement results. Magnetic resonance imaging (MRI) is a medical imaging technology that can provide high-definition views of soft tissues and organs. MRI uses a strong magnetic field and radio frequency pulses to cause hydrogen nuclei (protons) in the body to resonate, and then detects the decay of the resonance signals, and generates images after computer processing. Based on the above principle, MRI does not involve radiation and is relatively safe for the human body. In addition, MRI has higher contrast and resolution in soft tissue imaging. However, there is currently no scheme for measuring exophthalmos based on MRI images. SUMMARY

[0005] The present application provides an exophthalmos measurement method and device based on MRI images, which measures exophthalmos based on magnetic resonance imaging, thereby further improving the safety and accuracy of exophthalmos measurement.

[0006] In a first aspect, the present application provides an exophthalmos measurement method based on MRI images, comprising:

[0007] obtaining an MRI transverse image of the orbit;

[0008] determine the positions of the most anterior edges of the bilateral malar processes according to the MRI transverse image of the eye orbit;

[0009] determine the position of the eyeball barycenter according to the MRI transverse image of the eye orbit;

[0010] determine the position of the corneal anterior edge according to the MRI transverse image of the eye orbit;

[0011] calculate a first distance between the eyeball barycenter and the most anterior edges of the bilateral malar processes, and a second distance between the eyeball barycenter and the corneal anterior edge, and take the sum of the first distance and the second distance as the measurement result of the exophthalmos.

[0012] In a second aspect, the present application provides an exophthalmos measurement device based on MRI images, comprising:

[0013] an acquisition unit configured to acquire an MRI transverse image of the eye orbit;

[0014] a first determination unit configured to determine the positions of the most anterior edges of the bilateral malar processes according to the MRI transverse image of the eye orbit;

[0015] a second determination unit configured to determine the position of the eyeball barycenter according to the MRI transverse image of the eye orbit;

[0016] a third determination unit configured to determine the position of the corneal anterior edge according to the MRI transverse image of the eye orbit;

[0017] a calculation unit configured to calculate a first distance between the eyeball barycenter and the most anterior edges of the bilateral malar processes, and a second distance between the eyeball barycenter and the corneal anterior edge, and take the sum of the first distance and the second distance as the measurement result of the exophthalmos.

[0018] In addition, the present application also provides a terminal comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the exophthalmos measurement method based on MRI images according to the first aspect of the present application when executing the computer program.

[0019] The application provides a protrusion measurement method and device based on MRI images, the method comprises the following steps: acquiring an MRI transverse image of an eye socket, and determining the positions of the most forward edges of bilateral malar processes, the center of gravity of an eyeball and the front edge of a cornea according to the MRI transverse image of the eye socket; calculating a first distance between the center of gravity of the eyeball and the most forward edges of the bilateral malar processes, and a second distance between the center of gravity of the eyeball and the front edge of the cornea, and finally taking the sum of the first distance and the second distance as the measurement result of the protrusion. It can be known that the protrusion is measured based on the MRI image of the eye socket, on the one hand, the safety is improved compared with the CT method because the MRI does not contain radiation, and on the other hand, the measurement result of the protrusion obtained based on the MRI image is more accurate because the MRI has higher contrast and resolution in the imaging of soft tissues. BRIEF DESCRIPTION OF DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without any creative labor.

[0021] Figure 1 is an application scenario diagram of the protrusion measurement method based on the MRI image provided by the embodiments of the present application;

[0022] Figure 2 is an implementation flowchart of the protrusion measurement method based on the MRI image provided by the embodiments of the present application;

[0023] Figure 3 is a schematic diagram of the 2D sequence image and the 3D sequence image of the eyeball provided by the embodiments of the present application;

[0024] Figure 4 is a schematic diagram of the position relationship between the most forward edges of the bilateral malar processes, the front edge of the cornea and the center of gravity of the eyeball provided by the embodiments of the present application;

[0025] Figure 5 is a structural schematic diagram of the protrusion measurement device based on the MRI image provided by the embodiments of the present application. DETAILED DESCRIPTION

[0026] In the following description, specific details are set forth such as particular system configurations, techniques, etc., in order to provide a thorough understanding of the embodiments of the present application. However, it should be apparent to those skilled in the art that the present application can be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present application with unnecessary detail.

[0027] To make the objectives, technical solutions, and advantages of the present invention clearer, specific embodiments will be described below in conjunction with the accompanying drawings.

[0028] Figure 1 This is a schematic diagram illustrating an application scenario of the MRI-based exophthalmos measurement method provided in an embodiment of the present invention. For example... Figure 1 As shown, the present invention can be applied to a measurement terminal. The measurement terminal performs key point identification and location positioning on the input orbital MRI transverse image, including locating the position of the anterior edge of the zygomatic process on both sides, locating the position of the eyeball centroid, and locating the position of the anterior edge of the cornea. Then, it calculates the first distance from the position of the eyeball centroid to the line connecting the position of the anterior edge of the zygomatic process on both sides, and the second distance from the position of the eyeball centroid to the position of the anterior edge of the cornea, and uses the sum of the first distance and the second distance as the result of the proptosis measurement.

[0029] It should be noted that on axial MRI images, the vertical distance from the apex of the anterior corneal margin to the line connecting the apex of the bony orbital margin is used as the baseline to measure the eyeball protrusion. In practical applications, the apex of the bony orbital margin is actually the position of the anterior edge of the zygomatic process. There is an anterior edge of the zygomatic process on each side. The vertical distance from the apex of the anterior corneal margin to the line connecting the two anterior edges of the zygomatic processes is the degree of proptosis.

[0030] This invention enables the measurement of proptosis based on orbital MRI images. On the one hand, since MRI does not involve radiation, it is safer than CT. On the other hand, since MRI has higher contrast and resolution in soft tissue imaging, the proptosis measurement results obtained based on MRI images are more accurate.

[0031] It should be noted that CT imaging excels in displaying bone and calcification, but its contrast is lower than that of MRI in soft tissue imaging. Furthermore, the measurement of exophthalmos primarily targets the orbital region, involving more soft tissue imaging. Because MRI images have better soft tissue resolution, it is easier to distinguish between the anterior corneal margin and the closed eyelid, thus allowing for more accurate identification and localization of key points (the anterior margins of the bilateral zygomatic processes, the center of gravity of the eyeball, and the anterior corneal margin), resulting in higher accuracy.

[0032] In addition, it should be noted that during CT or MRI imaging examinations, the target needs to maintain a straight-ahead gaze for a certain period of time. If the target is strabismus, eye movement will introduce certain errors. The MRI-based method for measuring exophthalmos can also effectively correct strabismus in MRI images, obtaining more accurate exophthalmos measurement results. The specific principles and processes are detailed in the following embodiments.

[0033] See Figure 2Fig. 1 shows a flowchart of the implementation of the exophthalmos measurement method based on MRI images according to an embodiment of the present application, which is described in detail as follows:

[0034] Step 201: Obtain an MRI transverse plane image of the eye socket.

[0035] Step 202: Determine the positions of the most forward edges of the two zygomatic processes according to the MRI transverse plane image of the eye socket.

[0036] Step 203: Determine the position of the eyeball center of gravity according to the MRI transverse plane image of the eye socket.

[0037] Step 204: Determine the position of the corneal leading edge according to the MRI transverse plane image of the eye socket.

[0038] Step 205: Calculate the first distance between the eyeball center of gravity and the most forward edges of the two zygomatic processes, and the second distance between the eyeball center of gravity and the corneal leading edge, and take the sum of the first distance and the second distance as the exophthalmos measurement result.

[0039] The specific implementation of the above steps is described as follows:

[0040] First, in step 201, an MRI transverse plane image of the eye socket is obtained.

[0041] The transverse plane (Axial or Transverse Plane), also known as the horizontal plane, is a transverse section parallel to the ground, similar to a slice image obtained by transversely cutting from the top of the head to the bottom of the feet.

[0042] In MRI scanning, different planes are usually scanned to obtain comprehensive diagnostic information. The conventional scanning directions include the transverse plane, the sagittal plane, or the coronal plane. The present application identifies and locates the key points (including the positions of the most forward edges of the two zygomatic processes, the position of the eyeball center of gravity, and the position of the corneal leading edge) based on the transverse plane image, and measures the exophthalmos.

[0043] In the embodiments of the present application, the MRI transverse plane image can be a 2D (two-dimensional) sequence image or a 3D (three-dimensional) sequence image, each of which has its own advantages and disadvantages.

[0044] A 2D sequence image displays internal structures through a two-dimensional section. Common 2D sequences include T1-weighted imaging (T1WI), T2-weighted imaging (T2WI), diffusion-weighted imaging (DWI), etc. The 2D sequence scanning time is relatively short, suitable for rapid evaluation and preliminary diagnosis, and usually has a high signal-to-noise ratio (SNR), good image quality, and is suitable for observing fine structures. However, the 2D sequence cannot provide complete three-dimensional spatial information, and there may be limitations in the comprehensive understanding of complex structures.

[0045] 3D sequence images can provide more stereoscopic and comprehensive perspectives through three-dimensional data acquisition. Common 3D sequences include 3D magnetization preparation gradient echo (3D-MPRAGE) and 3D fast spin echo (3D-FSE). 3D sequences can provide higher spatial resolution and more detailed anatomical structure display, and are particularly suitable for key point detection of fine structures. 3D sequence data can be subjected to multi-planar reconstruction (MPR) to observe lesions from different angles and provide more information. The scanning time of 3D sequences is relatively long, and the cooperation of the detector is required to be higher, and the requirements for the equipment are also higher. In some cases, 3D sequences can obtain multiple cross-sectional images through one scan, thereby reducing the number of scans.

[0046] In the embodiments of the present application, before the identification and positioning of the key points are performed, the MRI transverse images can also be pre-processed, such as automatic segmentation of the eyeball of the MRI transverse images, to achieve more accurate and rapid later positioning and identification.

[0047] Specifically, the automatic segmentation of the eyeball can include automatic segmentation based on 2D images and automatic segmentation based on 3D images. Specifically, a deep learning network model VB-Net is constructed as an eyeball automatic segmentation model, and the collected data set is used to train the eyeball automatic segmentation model, so that the model has the ability to automatically segment the eyeball of the MRI transverse images.

[0048] For example, for 2D sequence images, 2D sequence orbital MRI transverse images from 109 patients are collected prospectively, axial T2WIDIXON in-phase (voxel 0.5x0.5x2.5mm, dimension 320x320x20), and Slicer is used for manual segmentation of the eyeball layer by layer to obtain a training data set. Then the eyeball segmentation model is trained, the parameter setting is a sampling interval of 0.5x0.5x2.5, a cropping size of 64x64x64, image preprocessing is performed by a nearest neighbor interpolation method and adaptive normalization, the pre-constructed deep learning network model VB-Net is trained, and finally the automatic segmentation images are checked one by one, the volumes of the eyeballs under manual segmentation and automatic segmentation are compared, the segmentation efficiency of the model is evaluated, and finally the trained automatic segmentation model capable of automatically segmenting the eyeball of the MRI transverse images is obtained.

[0049] Exemplarily, for the 3D sequence image, 3D T2WI sequence MRI transverse images (voxel 0.5x0.5x0.5, dimension 320x320x80) from 109 patients are prospectively collected, semi-automatic segmentation is performed by using 3DSlicer, the parameter is set as sampling interval 0.5x0.5x2.5, clipping size: 64x64x64, image preprocessing is performed by using the nearest neighbor interpolation method and adaptive normalization, then the pre-constructed deep learning network model VB-Net is trained, finally the automatic segmentation images are checked one by one, the eyeball volumes under manual segmentation and automatic segmentation are compared, the segmentation efficiency of the model is evaluated, and finally the trained automatic segmentation model capable of automatically segmenting the eyeball of the MRI transverse image is obtained.

[0050] In the embodiment of the application, the eyeball automatic segmentation model is used to automatically segment the MRI transverse image of the orbit, to obtain the eyeball image, and then the key point recognition and positioning can be realized based on the segmented eyeball image, so that the complexity of image processing can be greatly reduced and the recognition efficiency can be improved.

[0051] In step 202, the positions of the most forward edges of the bilateral zygomatic processes are determined according to the MRI transverse image of the orbit.

[0052] In the embodiment, the positions of the most forward edges of the bilateral zygomatic processes are recognized and positioned according to the MRI transverse image of the orbit, the eyeball automatic segmentation model is first used to automatically segment the MRI transverse image of the orbit to obtain the eyeball image, and then the trained zygomatic process most forward edge position recognition model is used to recognize the positions of the most forward edges of the bilateral zygomatic processes on the eyeball image.

[0053] Figure 3 The schematic diagram of the 2D sequence image and the 3D sequence image (actually 3D, the plane diagram cannot truly show the 3D effect) of the eyeball is as shown in Figure 3 As shown, the MRI transverse image of the orbit includes a plurality of layer images.

[0054] Specifically, for the 2D sequence image, the eyeball maximum layer image with the largest cross-sectional area can be extracted according to the cross-sectional area of each layer eyeball image in the MRI transverse image of the orbit; the eyeball maximum layer image is input into the trained zygomatic process most forward edge position recognition model to obtain the eyeball maximum layer image with the labeled positions of the most forward edges of the bilateral zygomatic processes.

[0055] Specifically, for the 3D sequence image, each eyeball layer image in the MRI transverse image of the orbit can be input into the trained zygomatic process most forward edge position recognition model to obtain a plurality of edge positions corresponding to each eyeball layer image; the positions of the most forward edges of the bilateral zygomatic processes are selected from the plurality of edge positions corresponding to each eyeball layer image.

[0056] It should be noted that for the 2D sequence image, the most front edge positions of the bilateral zygomatic processes can be considered to be located on the maximum eyeball layer image (i.e., the layer image with the largest eyeball region on each layer image), and the most front edge positions of the bilateral zygomatic processes can be directly obtained by selecting the maximum eyeball layer image and using the trained zygomatic process most front edge position recognition model to recognize. For the 3D sequence image, since one more dimension is considered, the most front edge positions of the bilateral zygomatic processes on the maximum eyeball layer image are not necessarily the most front edge positions, and therefore, the trained zygomatic process most front edge position recognition model needs to be used to recognize each layer image to obtain multiple front edge positions, and then the most front edge positions of the bilateral zygomatic processes are selected from the multiple front edge positions, wherein the distinction between the front edge and the most front edge can be obtained based on a three-dimensional rectangular coordinate system according to three-dimensional coordinate comparison.

[0057] For example, the 2D sequence of the transverse position image of the orbit MRI includes 20 layers, and the maximum eyeball layer image with the largest eyeball cross section needs to be extracted from the 20 layer images, and the most front edge positions of the bilateral zygomatic processes can be obtained by using the zygomatic process most front edge position recognition model to recognize once.

[0058] For example, the 3D sequence of the transverse position image of the orbit MRI includes 80 layers, and the zygomatic process most front edge position recognition model needs to be used to recognize the 80 layer images respectively to obtain 80 bilateral zygomatic process front edge positions, and then one bilateral zygomatic process most front edge position is selected from the 80 bilateral zygomatic process front edge positions. Of course, in order to improve the recognition efficiency, a certain number of layers including the eyeball region can be selected from the 80 layer images, and then the bilateral zygomatic process front edge positions of the certain number of layers including the eyeball region are recognized.

[0059] In the embodiment of the present application, the implementation of automatically recognizing and determining the most front edge positions of the bilateral zygomatic processes using the trained machine learning model can be as follows: inputting the maximum eyeball layer image into the trained zygomatic process most front edge position recognition model to obtain the maximum eyeball layer image with the labeled most front edge positions of the bilateral zygomatic processes.

[0060] The zygomatic process most front edge position recognition model is constructed based on a hybrid model, and the hybrid model includes a deep residual network for feature extraction and a U-Net convolutional neural network for image segmentation; and the training data source of the hybrid model is a certain number of transverse position images of the orbit MRI containing the labeled most front edge positions of the bilateral zygomatic processes.

[0061] In the embodiment of the present application, whether in the training process or in the actual measurement process, before the maximum eyeball layer image is input into the zygomatic process most front edge position recognition model, the maximum eyeball layer image can be preprocessed, specifically, the maximum eyeball layer image is traversed from front to back to obtain a target coordinate with a first value greater than a preset value; and the maximum eyeball layer image is cropped based on the target coordinate to reduce the maximum eyeball layer image to a preset size.

[0062] In this embodiment, since the frontmost edge of the zygomatic process and the feature part for identifying the frontmost edge of the zygomatic process are located at the front position of the image, the image can be traversed from front to back, and the first large value coordinate is returned, which is exemplarily 1000. Based on this coordinate, the image can be cropped, and after cropping, the size of the data image can be changed from 320x320 to 120x320. Through the above image preprocessing steps, the target range can be reduced, the accuracy of eye segmentation can be improved, the algorithm burden of the model can be reduced, the operation efficiency of the model can be improved, and the model can be facilitated to locate the feature part.

[0063] In this embodiment, the training process of the model needs data labeling, that is, labeling the zygomatic process position in the MRI image. However, since the zygomatic process frontmost edge position required by data labeling is a pixel point coordinate, which is too small for the naked eye, this brings difficulties to the data labeling process, so the way of labeling with different markers is adopted. Specifically, the way of labeling the zygomatic process frontmost edge position in the eye maximum layer image can be region labeling. Correspondingly, the step of obtaining the zygomatic process frontmost edge position from the eye maximum layer image which has been region labeled can include: taking the center of gravity of the two marker regions respectively to obtain two pixel point coordinates indicating the zygomatic process frontmost edge position. Specifically, K-means can be used to cluster the marker data to obtain two marker regions, and then the center of gravity of the two marker regions is taken respectively to finally obtain two required marker points.

[0064] In one specific embodiment, the deep residual network used for feature extraction includes ResNet34; the input channel number of the first layer convolution of the ResNet34 is 1.

[0065] The model architecture proposed in this embodiment is a hybrid of ResNet34 and U-Net, which is designed specifically for medical image segmentation tasks. The model makes full use of the deep learning functions of PyTorch, and is especially suitable for processing magnetic resonance images. The architecture consists of an encoder based on ResNet34 (for feature extraction), a customized decoder (for segmentation tasks), and a special dataset class for processing MRI data.

[0066] In this embodiment, ResNet34 is chosen as the encoder, which is originally famous for its performance in image classification and brings strong feature extraction capability. However, the magnetic resonance imaging images are usually single-channel images, which are different from the three-channel RGB images that ResNet34 is used to. To solve this problem, the first layer of ResNet34 is modified to make the input channel number of the first layer convolution of ResNet34 be 1 to accept single-channel input, which is a key adaptation for handling magnetic resonance imaging data. In addition, the parameters of the last two layers (layer 3 and layer 4) and the fully connected layer of the network can also be trained. This decision ensures that the neural network can be fine-tuned for the actual task, thereby enhancing the ability to capture complex medical images.

[0067] In this embodiment, the decoder can be composed of multiple custom “decoder block” modules, each of which performs a series of upsampling and convolution operations. These modules are able to utilize the skip connections from the corresponding layers of the encoder. This approach helps to recover the spatial information lost during the encoder’s downsampling process, which is a key factor for accurate segmentation. The last convolutional layer of the decoder, which maps the rich upsampled feature maps to the desired localization, finally provides the output of the cheekbone position we need.

[0068] In addition, the NMRDataset class is an important part of the architecture of this embodiment. This class is designed to handle the standard format “.nii.gz” files in medical images, ensuring that the model can effectively process real-world magnetic resonance data. It not only loads and processes MRI images, but also loads and processes the corresponding labels, thereby simplifying the training process. The class also contains a transformation pipeline, which includes important preprocessing steps such as resizing and normalization to ensure the consistency and quality of the input data.

[0069] The model in this embodiment has good integration with the PyTorch ecosystem, with data loaders for efficient batch processing and CUDA for GPU acceleration to ensure efficient training of the model. The training loop is custom designed, using BCEWithLogitsLoss as the loss function and using the adam optimizer with different learning rates for different parts of the model. This meticulous training method allows for more fine-grained control and optimization of the learning process.

[0070] In this embodiment, the dataset can be divided into training, validation and test sets in the ratio of 7:2:1, and 200 epochs are run, where an epoch refers to a complete forward and backward propagation of the model on the entire training dataset. In other words, an epoch means that all training samples are used to update the model’s weights once. The loss function converges to near zero on both the training set and the test set.

[0071] The model results of the embodiment can well present the position of the anterior edge of the malar prominence. In the MRI image with artifacts, the recognition ability is superior to that of non-professionals with short-term training. If the distance between the result and the label is less than 1 pixel, the accuracy of the model on the test set reaches 100%. However, the difference of 1 pixel is difficult to distinguish and accurately locate when labeling, so this error is considered to be systematic, and can be improved by accurate labeling and increasing the amount of data in the future.

[0072] In step 203, the eyeball barycenter position is determined according to the transverse orbital MRI image.

[0073] The step of the embodiment is used to determine the eyeball barycenter position, which can be calculated by some mathematical methods. However, in the embodiment, the eyeball is homogeneous, and both the centroid moment and the mass can be neglected in the end. Therefore, the coordinates of the eyeball barycenter position can be obtained by calculating the average of the coordinates of the related points of the eyeball.

[0074] Specifically, for the 2D sequence of the transverse orbital MRI image, the barycenter position of the eyeball must be located on the maximum layer image of the eyeball. Therefore, the barycenter position of the eyeball can be obtained by calculating the average of the positions of all points corresponding to the eyeball image on the maximum layer image of the eyeball. For example, the average of the horizontal axis coordinates of each point on the maximum layer image of the eyeball is obtained, and the horizontal axis coordinate of the barycenter position is obtained. The average of the vertical axis coordinates of each point is obtained, and the vertical axis coordinate of the barycenter position is obtained.

[0075] Specifically, for the 3D sequence of the transverse orbital MRI image, the barycenter position of the eyeball needs to be calculated based on all points on the three-dimensional eyeball surface. First, the coordinates of all points on the eyeball surface in the three-dimensional rectangular coordinate system in the transverse orbital MRI image are obtained. Second, the average of the horizontal axis, the average of the vertical axis and the average of the vertical axis of all points on the eyeball surface are calculated. The position with the average of the horizontal axis, the average of the vertical axis and the average of the vertical axis in the three-dimensional rectangular coordinate system is determined as the barycenter position of the eyeball.

[0076] In step 204, the corneal anterior edge position is determined according to the transverse orbital MRI image.

[0077] The step of the embodiment is used to determine the corneal limbus position. In the embodiment of the application, an important point is how to determine the position of the corneal limbus on the maximum layer image of the eyeball, that is, the position of the top point of the corneal limbus, which is a key factor to ensure the accuracy of the exophthalmos measurement. In some measurement methods in the related art, the determined is the front surface of the eyeball, not the corneal limbus position. This leads to the fact that in the case of non-emmetropia, the position of the front surface of the eyeball is not the position of the corneal limbus, and the exophthalmos should be the vertical distance between the position of the corneal limbus and the front edge of the malar prominence, not the vertical distance between the position of the front surface of the eyeball and the front edge of the malar prominence. If the position of the front surface of the eyeball is determined, measurement error is caused. At present, the measurement in most related art is the position of the front surface of the eyeball, for example, the CT measurement method. This is mainly because the resolution of soft tissue of CT is slightly poor, and it is not easy to identify the position of the corneal limbus.

[0078] In the embodiment, for the 2D sequence of the orbital MRI transverse image, the layer thickness of the 2D transverse image is large, about 2.5 mm. It is considered that the position of the corneal limbus is on the maximum layer image of the eyeball (not affected by the layer thickness of the 2D sequence image), so the front end point of the longest axis of the eyeball region on the maximum layer image of the eyeball can be determined as the position of the corneal limbus. Figure 4 The position relationship between the front edge of the malar prominence, the position of the corneal limbus and the position of the eyeball center provided by the embodiment of the application is shown in the schematic diagram as shown in Figure 4 As shown in the schematic diagram, the eyeball is approximately a sphere, and the cornea is equivalent to an arc protruding on the sphere. The position of the corneal limbus is (x3 ， y3), and the line passing through the position of the corneal limbus and the eyeball center (x3 ， a) is the longest axis of the eyeball region. Therefore, the front end point of the longest axis of the eyeball region on the maximum layer image of the eyeball is the position of the corneal limbus.

[0079] It should be noted that due to the large layer thickness of the 2D transverse image, about 2.5 mm, when the eyeball is volume rendered, the height of the eyeball in the vertical direction is overestimated, so that the stereoscopic recognition of the longest axis of the eyeball is disturbed by the vertical axis. For example, the distance from the farthest point on the surface of the eyeball to the center of gravity may not be the distance from the corneal limbus to the center of gravity, but the distance from the top point of the eyeball to the center of gravity (overestimation of the height of the eyeball in the vertical direction causes wrong recognition). Therefore, in the 2D sequence, only the center of gravity plane, not the entire sphere, needs to be recognized based on the front end point of the longest axis, that is, the position of the corneal limbus.

[0080] For the MRI transverse image of the eye orbit of the 3D sequence, since the layer thickness is small, it usually does not affect the identification of the corneal front edge position, and the position of the point with the farthest distance to the eyeball barycenter position on the eyeball surface can be taken as the corneal front edge position. In addition, since the MRI detection process lasts for a long time, it is difficult for the detector to keep the front view state for a long time, and if the image has a strabismus, that is, the corneal front edge position deviates from the direction of the front view scene, the corneal front edge position may not be located on the maximum layer image of the eyeball, and the above-mentioned corneal front edge position detection method corresponding to the 2D sequence may produce some errors, and the corneal front edge position identification method of traversing all points on the eyeball surface, calculating the distance from each point on the eyeball surface to the eyeball barycenter position, and taking the position of the point with the farthest distance to the eyeball barycenter position as the corneal front edge position is more accurate.

[0081] In step 205, the first distance from the eyeball barycenter position to the line connecting the frontmost edge positions of the two zygomatic arches and the second distance from the eyeball barycenter position to the corneal front edge position are calculated, and the sum of the first distance and the second distance is taken as the exophthalmos measurement result.

[0082] In the embodiment of the application, the distance from the eyeball barycenter position to the line connecting the frontmost edge positions of the two zygomatic arches is calculated, and then the distance from the eyeball barycenter position to the corneal front edge position is added, that is, the distance from the corneal front edge position to the line connecting the frontmost edge positions of the two zygomatic arches is obtained under the condition that the corneal front edge position is in the frontmost position and the frontmost edge positions of the two zygomatic arches and the corneal front edge position are located on the same horizontal plane, and an equivalent exophthalmos value is obtained.

[0083] Specifically, the measurement terminal can obtain the exophthalmos measurement result based on a preset exophthalmos calculation formula, and the exophthalmos calculation formula includes:

[0084]

[0085] wherein d represents the exophthalmos measurement result, (x1, y1) represents the position coordinates of the frontmost edge of one zygomatic arch, (x2, y2) represents the position coordinates of the frontmost edge of the other zygomatic arch, and (x3, y3) represents the position coordinates of the corneal front edge.

[0086] Figure 4 The position relationship between the frontmost edge positions of the two zygomatic arches, the corneal front edge position and the eyeball barycenter position is shown in FIG. 1. Figure 4 As shown in FIG. 1, the coordinates of the frontmost edge positions of the two zygomatic arches are (x1, y1) and (x2, y2) respectively, the corneal front edge position in the front view scene is (x3, y3), and the corneal front edge position in the strabismus is (b, c). The position coordinates of the eyeball barycenter are (x3, a).

[0087] In one embodiment, if the target in the image has a strabismus, the exophthalmos should be (b， c) to (x3 ， a) is (b-x3 ， a) to the vertical distance of the two lines.(b ， c) to (x3 ， a) is (b-x3 2 +(c-a) 2 .

[0088] In fact, assuming that the position of the center of gravity in the sagittal plane changes during eye rotation can be ignored and the shape of the eye does not change, (b-x3 2 +(c-a) 2 can be equivalent to the distance from (x3, y3) to (x3, a), that is, (y3-a) 2 That is, by rotating the eyeball, the corneal front edge is located at the frontmost position, and the frontmost position of the two zygomatic processes and the corneal front edge are located on the same horizontal plane, and the equivalent exophthalmos is calculated. That is, the vertical distance from (x3 ， y3) to the line connecting the frontmost positions of the two zygomatic processes can be obtained.

[0089] The related art indicates that the center of gravity position may have a zero-point several millimeter offset, but according to Figure 3 It can be considered that when the offset is projected onto the axis of the exophthalmos measurement, the offset can be ignored. And based on the calculation results of the embodiments of the present application, the measured values are also similar to the manual measurement values (deviation less than 0.5mm).

[0090] In the embodiments of the present application, for the strabismus scene, the actual corneal front edge coordinates (b ， c) can be converted to the equivalent corneal front edge coordinates (x3 ， y3) in orthovision, so as to realize the correction of exophthalmos in the strabismus scene.

[0091] It should be noted that in some other related technologies, due to the inability to accurately locate the corneal front edge position, the actual positioning is the position of the eyeball front surface (i.e. Figure 4 The intersection position below the midpoint (x3 ， y3) of the sphere), so there is an error, which affects the measurement accuracy.

[0092] The application provides a protrusion degree measurement method based on MRI images, the protrusion degree measurement method comprises the following steps: acquiring an MRI transverse image of an eye socket, determining the positions of the most front edges of bilateral malar processes, the position of a eyeball gravity center and the position of a corneal front edge according to the MRI transverse image of the eye socket, calculating a first distance between the eyeball gravity center and the most front edges of the bilateral malar processes and a second distance between the eyeball gravity center and the corneal front edge, and finally taking the sum of the first distance and the second distance as a protrusion degree measurement result.

[0093] It should be understood that the size of the serial number of each step in the above embodiment does not mean the order of execution, and the execution order of each process should be determined according to its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the application.

[0094] The following is a device embodiment of the application, and for details not described in detail, reference can be made to the corresponding method embodiments described above.

[0095] Figure 5 The structure schematic diagram of the pathogenic microorganism information recognition device provided by the embodiment of the application is shown, only the part related to the embodiment of the application is shown for the convenience of description, and the details are as follows:

[0096] As shown in Figure 5 The protrusion degree measurement device based on MRI images 5 comprises an acquisition unit 51, a first determination unit 52, a second determination unit 53, a third determination unit 54 and a calculation unit 55.

[0097] The acquisition unit 51 is used for acquiring an MRI transverse image of an eye socket;

[0098] The first determination unit 52 is used for determining the positions of the most front edges of bilateral malar processes according to the MRI transverse image of the eye socket;

[0099] The second determination unit 53 is used for determining the position of a eyeball gravity center according to the MRI transverse image of the eye socket;

[0100] The third determination unit 54 is used for determining the position of a corneal front edge according to the MRI transverse image of the eye socket;

[0101] The calculation unit 55 is used for calculating a first distance between the eyeball gravity center and the most front edges of the bilateral malar processes and a second distance between the eyeball gravity center and the corneal front edge, and taking the sum of the first distance and the second distance as a protrusion degree measurement result.

[0102] In an embodiment, the transverse orbital MRI image is a 2D sequence image, and the first determination unit 52 is specifically configured to extract an eyeball maximum layer image with the largest cross-sectional area according to the cross-sectional area of each eyeball image in the transverse orbital MRI image;

[0103] The eyeball maximum layer image is input into the trained most anterior edge position recognition model of the malar process to obtain an eyeball maximum layer image with the most anterior edge positions of the two malar processes labeled.

[0104] The second determination unit 53 is specifically configured to calculate the average of the positions of all points corresponding to the eyeball image on the eyeball maximum layer image to obtain the center of gravity position of the eyeball.

[0105] The third determination unit 54 is specifically configured to determine the front end position of the longest axis of the eyeball region on the eyeball maximum layer image as the corneal front edge position.

[0106] In an embodiment, the transverse orbital MRI image is a 3D sequence image, and the first determination unit 52 is specifically configured to input each eyeball layer image in the transverse orbital MRI image into the trained most anterior edge position recognition model of the malar process to obtain a plurality of front edge positions corresponding to each eyeball layer image; and select the most anterior edge positions of the two malar processes from the plurality of front edge positions corresponding to each eyeball layer image.

[0107] The second determination unit 53 is specifically configured to obtain the coordinates of all points on the eyeball surface in the transverse orbital MRI image in a three-dimensional rectangular coordinate system; calculate the horizontal axis mean, the vertical axis mean and the vertical axis mean of all points on the eyeball surface; and determine the position with the horizontal axis mean, the vertical axis mean and the vertical axis mean in the three-dimensional rectangular coordinate system as the center of gravity position of the eyeball.

[0108] The third determination unit 54 is specifically configured to traverse all points on the eyeball surface, calculate the distance from each point on the eyeball surface to the center of gravity position of the eyeball, and determine the position of the point with the farthest distance to the center of gravity position of the eyeball as the corneal front edge position.

[0109] In an embodiment, the most anterior edge position recognition model of the malar process is constructed by a hybrid model, the hybrid model includes a deep residual network for feature extraction and a U-Net convolutional neural network for image segmentation; the training data source of the most anterior edge position recognition model of the malar process is a certain number of layer images of transverse orbital MRI images with the most anterior edge positions of the two malar processes labeled; the deep residual network for feature extraction includes ResNet34; the input channel number of the first layer convolution of the ResNet34 is 1.

[0110] In an embodiment, the calculation unit 55 is specifically configured to obtain a protopsis measurement result based on a preset protopsis calculation formula, and the protopsis calculation formula includes:

[0111]

[0112] wherein d represents the exophthalmos measurement result, (x1 ， y1) represents the position coordinates of the most forward edge of the one side malar process, (x2 ， y2) represents the position coordinates of the most forward edge of the other side malar process, (x3 ， y3) represents the position coordinates of the corneal leading edge.

[0113] It can be known from the above that the application provides an exophthalmos measurement device based on MRI images, the most forward edge positions of the two side malar processes, the eyeball barycenter position and the corneal leading edge position are determined according to the MRI transverse position images of the eye socket, the first distance between the eyeball barycenter position and the most forward edge positions of the two side malar processes and the second distance between the eyeball barycenter position and the corneal leading edge position are calculated, and finally the sum of the first distance and the second distance is taken as the exophthalmos measurement result. It can be known that the exophthalmos is measured based on the MRI images of the eye socket, on the one hand, the safety is improved compared with the CT mode because the MRI does not contain radiation, and on the other hand, the exophthalmos measurement result obtained based on the MRI images is more accurate because the MRI has higher contrast and resolution in the imaging of soft tissues.

[0114] The application also provides a measurement terminal, which can include a processor, a memory and a computer program stored in the memory and executable on the processor. The processor implements the steps in each of the above exophthalmos measurement methods based on MRI images, such as steps 201 to 205 shown in the figure. Figure 2 Alternatively, the processor implements the functions of each unit in each of the above device embodiments, such as the functions of units 41 to 45 shown in the figure. Figure 4

[0115] For example, the computer program can be divided into one or more units, which are stored in the memory and executed by the processor to complete the application. The one or more units can be a series of computer program instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program in the measurement terminal.

[0116] The measurement terminal can be a desktop computer, a notebook, a palm computer and a cloud server, etc. The measurement terminal can include, but is not limited to, a processor and a memory. Those skilled in the art can understand that the measurement terminal can include more or fewer components than those shown, or combine certain components, or different components, for example, the measurement terminal can also include input and output devices, network access devices, buses, etc.​

[0117] The processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0118] The memory can be an internal storage unit of the measurement terminal, such as a hard disk or a memory of the measurement terminal. The memory can also be an external storage device of the measurement terminal, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. Further, the memory can also include both the internal storage unit and the external storage device of the measurement terminal. The memory is used to store the computer program and other programs and data required by the terminal. The memory can also be used to temporarily store data that has been output or will be output.

[0119] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above functional units and modules is exemplified, and in actual application, the above functions can be completed by different functional units and modules according to needs, that is, the internal structure of the apparatus is divided into different functional units or modules to complete all or part of the above described functions. Each functional unit and module in the embodiment can be integrated in one processing unit, or each unit can be physically present separately, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit. In addition, the specific names of each functional unit and module are only for easy distinction, and do not limit the protection scope of the present application. The specific working process of the units and modules in the above system can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0120] In the above embodiments, the description of each embodiment has its own emphasis, and the parts not described or recorded in detail in a certain embodiment can be referred to the relevant description of other embodiments.

[0121] Those skilled in the art can understand that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized in electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0122] In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0123] The above-described embodiments are only used to illustrate the technical solutions of the present application, not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that the technical solutions recorded in the foregoing embodiments can be modified, or some technical features can be replaced by equivalents; and these modifications or replacements do not make the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A method for measuring exophthalmos based on MRI images, characterized by, The method comprises the following steps: obtaining an orbital MRI transverse image; determining the positions of the most anterior edges of the bilateral malar processes according to the orbital MRI transverse image; determining the position of the eyeball barycenter according to the orbital MRI transverse image; determining the position of the corneal anterior edge according to the orbital MRI transverse image; calculating a first distance between the position of the eyeball barycenter and the positions of the most anterior edges of the bilateral malar processes, and a second distance between the position of the eyeball barycenter and the position of the corneal anterior edge, and taking the sum of the first distance and the second distance as a measurement result of exophthalmos.

2. The MRI image-based exophthalmos measurement method of claim 1, wherein, The orbital MRI transverse image is a 2D sequence image. The step of determining the positions of the most anterior edges of the bilateral malar processes according to the orbital MRI transverse image comprises the following steps: extracting an eyeball maximum layer image with the largest cross-sectional area from each layer eyeball image in the orbital MRI transverse image according to the cross-sectional areas of the layer eyeball images; inputting the eyeball maximum layer image into a trained malar process most anterior edge position recognition model to obtain an eyeball maximum layer image with the positions of the most anterior edges of the bilateral malar processes labeled.

3. The MRI image-based exophthalmos measurement method of claim 2, wherein, The step of determining the position of the eyeball barycenter according to the orbital MRI transverse image comprises the following steps: calculating the average of the positions of all points corresponding to the eyeball image on the eyeball maximum layer image to obtain the position of the eyeball barycenter.

4. The MRI image-based exophthalmos measurement method of claim 2, wherein, The step of determining the position of the corneal anterior edge according to the orbital MRI transverse image comprises the following steps: determining the position of the front end of the longest axis of the eyeball region on the eyeball maximum layer image as the position of the corneal anterior edge.

5. The MRI image-based exophthalmos measurement method of claim 1, wherein, The orbital MRI transverse image is a 3D sequence image. The step of determining the positions of the most anterior edges of the bilateral malar processes according to the orbital MRI transverse image comprises the following steps: inputting each eyeball layer image in the orbital MRI transverse image into a trained malar process most anterior edge position recognition model to obtain a plurality of anterior edge positions corresponding to each eyeball layer image; selecting the positions of the most anterior edges of the bilateral malar processes from the plurality of anterior edge positions corresponding to each eyeball layer image.

6. The MRI image-based exophthalmos measurement method according to claim 5, wherein, The step of determining the position of the eyeball barycenter according to the orbital MRI transverse image comprises the following steps: obtaining the coordinates of all points on the eyeball surface in the orbital MRI transverse image in a three-dimensional rectangular coordinate system; calculating the horizontal axis mean, the vertical axis mean and the vertical axis mean of all points on the eyeball surface, and determining the position with the coordinates of the horizontal axis mean, the vertical axis mean and the vertical axis mean in the three-dimensional rectangular coordinate system as the position of the eyeball barycenter.

7. The MRI image-based exophthalmos measurement method according to claim 6, wherein, The step of determining the position of the corneal anterior edge according to the orbital MRI transverse image comprises the following steps: traversing all points on the eyeball surface to calculate the distances of the points on the eyeball surface to the position of the eyeball barycenter; determining the position of the point with the farthest distance to the position of the eyeball barycenter as the position of the corneal anterior edge.

8. The MRI image-based exophthalmos measurement method according to claim 2 or 5, wherein, The malar process most anterior edge position recognition model is constructed by a hybrid model, the hybrid model comprising a deep residual network for feature extraction and a U-Net convolutional neural network for image segmentation; the training data source of the malar process most anterior edge position recognition model is a certain number of layer images of orbital MRI transverse images with the positions of the most anterior edges of the bilateral malar processes labeled; the deep residual network for feature extraction comprises a ResNet34; the input channel number of the first layer convolution of the ResNet34 is 1.

9. The MRI image-based exophthalmos measurement method according to any one of claims 1 to 7, wherein The first distance from the eyeball barycenter position to the line connecting the most front edge positions of the two zygomatic processes and the second distance from the eyeball barycenter position to the cornea front edge position are calculated, and the sum of the first distance and the second distance is included as the exophthalmos measurement result: The exophthalmos measurement result is obtained based on a preset exophthalmos calculation formula, and the exophthalmos calculation formula includes: wherein d represents the exophthalmos measurement result, (x1 ， y1) represents the position coordinates of the most anterior edge of the one side malar process, (x2 ， y2) represents the position coordinates of the most anterior edge of the other side malar process, (x3 ， y3) represents the position coordinates of the corneal limbus.

10. A protrusion measurement device based on MRI images, characterized by, The exophthalmos measurement result is obtained based on a preset exophthalmos calculation formula, and the exophthalmos calculation formula includes: An acquisition unit is configured to acquire an MRI transverse image of an eye socket; A first determination unit is configured to determine the most front edge positions of the two zygomatic processes according to the MRI transverse image of the eye socket; A second determination unit is configured to determine the eyeball barycenter position according to the MRI transverse image of the eye socket; A third determination unit is configured to determine the cornea front edge position according to the MRI transverse image of the eye socket; A calculation unit is configured to calculate the first distance from the eyeball barycenter position to the line connecting the most front edge positions of the two zygomatic processes and the second distance from the eyeball barycenter position to the cornea front edge position, and the sum of the first distance and the second distance is included as the exophthalmos measurement result.

Images