TWI719843B - Method for generating model for estimating bone density, method for estimating bone density and electronic system - Google Patents

Abstract

A method for generating a model for estimating bone density is implemented by an electronic system, the electronic system includes an X-ray machine, a dual-energy X-ray absorber and a processing unit, the method includes: the X-ray machine photographs Multiple reference patients’ predetermined bones to generate multiple training image data; the dual-energy X-ray absorber detects the predetermined bones of the reference patients to generate multiple bone density data; and the processing unit is based on the training image data And the bone density data, train a first convolutional neural network model to generate an estimation model, the estimation model is used for the X-ray machine to photograph the predetermined bone of a target patient to be analyzed The imaging data estimates the bone density of the target patient.

Description

Method for generating model for estimating bone density, method for estimating bone density, and electronic system

The present invention relates to a method for generating a model, in particular to a method for generating a model for estimating bone density. The invention also relates to a method for estimating bone density and an electronic system.

Osteoporosis (Femoral neck) fractures caused by osteoporosis has a mortality rate of nearly 20%. There are many causes of osteoporosis, including drugs, diet, and other lifestyle habits, and it is more common in the elderly. The current standard bone density examination is a detailed examination using dual energy x-ray absorptiometry (DEXA).

Bone mineral content (BMC), bone mineral density (BMD), T-score (T-score) and Z-score (Z-score) and other parameters generated by dual-energy X-ray absorbance Numerical value to assist physicians in the diagnosis of osteoporosis. In this test, the patient has to bear the risk of radiation dose. If the patient has previously received a general pelvis x-ray (Pelvis x-ray), there is a risk of a second radiation dose. How to improve the existing technology to reduce the risk of patients receiving radiation dose is the subject of the present invention.

Therefore, the purpose of the present invention is to provide a method for generating a model for estimating bone density.

Another object of the present invention is to provide a method for estimating bone density.

Another object of the present invention is to provide an electronic system.

Therefore, the method of the present invention for generating a model for estimating bone density is implemented by an electronic system that includes an X-ray machine, a dual-energy X-ray absorber, and a processing unit, and the method includes: the X The optical machine photographs a predetermined bone of multiple reference patients to generate multiple training image data; the dual-energy X-ray absorber detects the predetermined bones of the reference patients to generate multiple bone density data; and the processing unit generates multiple bone density data according to the predetermined bones of the reference patients. Training image data and the bone density data, training a first convolutional neural network model to generate an estimation model, the estimation model is used for the X-ray machine to shoot a target patient's predetermined bone generated The image data to be analyzed estimates the bone density of the target patient.

In some embodiments, the electronic system further includes an input unit. Before generating the estimation model and after generating the training image data, the method further includes: the processing unit, for each training image data, selects a circle from the training image data according to a circle selection command received through the input unit The image of the predetermined skeleton; and the processing unit performs a predetermined image processing for each training image data, and the predetermined image processing includes removing unselected images from the training image data.

In the step of generating the estimation model, the processing unit trains the first convolutional neural network model to generate the estimation model based on the training image data and the bone density data subjected to the predetermined image processing .

In some implementations, after the step of performing the predetermined image processing, the method further includes: the processing unit trains a second convolutional neural network model to generate a circle based on the training image data subjected to the predetermined image processing A selection model, and the circle selection model is used to circle out the image of the predetermined bone according to the image data to be analyzed.

In some embodiments, the predetermined bone is a hip joint of a pelvis, a lumbar spine, a pelvis, a ulna, or a femur.

The method for estimating bone density of the present invention is implemented by an electronic system. The electronic system includes an X-ray machine and a processing unit. The method includes: the X-ray machine photographs a predetermined bone of a target patient to generate a to-be-analyzed Image data; and the processing unit uses the estimation model to estimate the bone density of the target patient based on the image data to be analyzed to generate a bone density estimation data.

In some embodiments, the electronic system further includes an output unit. After generating the bone density estimation data, the method further includes: the processing unit determines whether the bone density estimation data meets a predetermined warning condition; and when the processing unit determines that the bone density estimation data meets the predetermined warning condition, the The processing unit outputs a warning message through the output unit.

In some implementation aspects, the predetermined warning condition includes that a T score corresponding to the bone density estimation data is less than a T score threshold.

In some implementation aspects, the predetermined warning condition includes that a Z score corresponding to the bone density estimation data is less than a Z score threshold.

In some implementation aspects, before generating the bone density estimation data and after generating the image data to be analyzed, the method further includes: the processing unit selects the image data to be analyzed by using the circle selection model according to the image data to be analyzed And the processing unit performs a predetermined image processing on the image data to be analyzed, and the predetermined image processing includes removing unselected images from the image data to be analyzed.

In the step of generating the bone density estimation data, the processing unit uses the estimation model to estimate the bone density of the target patient based on the image data to be analyzed that has undergone the predetermined image processing to generate the bone density estimation data.

The electronic system of the present invention includes an X-ray machine, a dual-energy X-ray absorber and a processing unit. The X-ray machine photographs a predetermined bone of a plurality of reference patients to generate a plurality of training image data. The dual-energy X-ray absorber detects the predetermined bones of the reference patients to generate multiple bone density data. The processing unit trains a first convolutional neural network model based on the training image data and the bone density data to generate an estimation model. The estimation model is used according to a target patient photographed by the X-ray machine The to-be-analyzed image data generated by the predetermined bone estimates the bone density of the target patient.

The electronic system of the present invention includes an X-ray machine and a processing unit. The X-ray machine photographs a predetermined bone of a target patient to generate an image data to be analyzed. The processing unit uses the estimation model to estimate the bone density of the target patient according to the image data to be analyzed to generate a bone density estimation data.

The effect of the present invention is that the processing unit trains the first convolutional neural network model according to the training image data and the bone density data to generate the estimation model, so that the target patient only needs to be subjected to the X-ray machine After shooting, the bone density estimation data can be generated based on the image data to be analyzed generated by the X-ray machine, without the need for the dual-energy X-ray absorber to detect, thereby reducing the risk of patients receiving radiation doses, and medical units The degree of reliance on expensive instruments such as the dual-energy X-ray absorber. In addition, the present invention uses the processing unit to perform the predetermined image processing for each training image data and the image data to be analyzed, so as to reduce the recognition by restricting the recognition part. Noise interference can further improve the accuracy of the estimation.

Before the present invention is described in detail, it should be noted that in the following description, similar elements are denoted by the same numbers.

Referring to FIG. 1, a first embodiment of the present invention is implemented by an electronic system 100. The electronic system 100 includes an X-ray machine 1 and a dual energy X-ray absorptiometry (DEXA). , An input unit 3, an output unit 4 and a processing unit 5. The input unit 3, the output unit 4, and the processing unit 5 can be implemented by a keyboard/mouse, a screen, and a host of a personal computer.

1 and 2, the steps of the method for generating a model for estimating bone density of the present invention are described below. First, as shown in step S01, the X-ray machine 1 photographs a predetermined bone of a plurality of reference patients to generate a plurality of training image data. In this embodiment, the predetermined bone is a hip joint (Hip) of a pelvis (Pelvis), but it is not limited to this. In other embodiments, the predetermined bone may also be, for example, the lumbar spine, the bone, and the ulna. Or femur.

Then, as shown in step S02, the dual-energy X-ray absorber 2 detects the predetermined bones of the reference patients to generate multiple bone mineral density (BMD) data. In this embodiment, the data generated by the X-ray machine 1 and the dual-energy X-ray absorber 2 are both in 16-bit Dicom (Digital Imaging and Communications in Medicine) format (the number of bits is not limited to 16 bits) ). In this embodiment, the number of data items of the training image data and the bone density data is, for example, 3,600, but it is not limited thereto. The sequence of step S01 and step S02 is not limited to the sequence exemplified in this embodiment.

Then, as shown in step S03, for each training image data, the processing unit 5 selects the image of the predetermined bone in the training image data according to a circle selection command received through the input unit 3. In this embodiment, a professional medical staff inputs the circle selection instruction to circle the image of the predetermined bone in the training image data.

Then, as shown in step S04, the processing unit 5 performs a predetermined image processing for each training image data, and the predetermined image processing includes removing unselected images from the training image data. In this embodiment, the predetermined image processing further includes cropping, rotation, brightness adjustment, contrast adjustment, and so on.

Next, as shown in step S05, the processing unit 5 trains a first Convolutional Neural Network (CNN) model based on the training image data and the bone density data that have undergone the predetermined image processing. Generate an estimation model. In this embodiment, the first convolutional neural network model is a model named MobleNetv2, in which each training result will update the parameters through back-propagation, but the first convolutional neural network model Not limited to MobleNetv2. It is supplemented that if the bone density data generated by the dual-energy X-ray absorber 2 is presented in an image file, the processing unit 5 needs to pass, for example, optical character recognition technology (OCR) before performing the training step of step S05. ) Extract the bone density data from the image file generated by the dual-energy X-ray absorber 2, and then execute step S05 according to the textualized bone density data. However, if the bone density data generated by the dual-energy X-ray absorber 2 is in text format, the processing unit 5 can directly execute step S05 without performing text extraction.

Then, as shown in step S06, the processing unit 5 trains a second convolutional neural network model based on the training image data subjected to the predetermined image processing to generate a circle selection model. In this embodiment, the second convolutional neural network model is a segmentation model named U-Net like, in which each training result will update the parameters through Back-propagation, but the first The two-convolutional neural network model is not limited to U-Net like.

The sequence of step S05 and step S06 is not limited to the sequence exemplified in this embodiment.

1 and 3, the following describes the steps of the method for estimating bone density of the present invention. First, as shown in step S11, the X-ray machine 1 photographs a predetermined bone of a target patient to generate an image data to be analyzed.

Then, as shown in step S12, the processing unit 5 selects the image of the predetermined bone in the image data to be analyzed by using the circle selection model described in step S06 according to the image data to be analyzed.

Then, as shown in step S13, the processing unit 5 performs a predetermined image processing on the image data to be analyzed, and the predetermined image processing includes removing unselected images from the image data to be analyzed. In this embodiment, the predetermined image processing further includes cropping, rotation, brightness adjustment, contrast adjustment, and so on.

Then, as shown in step S14, the processing unit 5 uses the estimation model described in step S05 to estimate the bone density of the target patient based on the image data to be analyzed that has undergone the predetermined image processing to generate a bone density Estimated data. In addition, the bone density estimation data is also displayed through the output unit 4.

Next, as shown in step S15, the processing unit 5 determines whether the bone density estimation data meets a predetermined warning condition, if yes, then step S16 is executed, if not, the process ends. In this embodiment, the predetermined warning condition includes a T-score corresponding to the bone density estimation data being less than a T-score threshold, and a Z-score corresponding to the bone density estimation data. ) Is less than a Z-score threshold, and as long as the T-score is less than the T-score threshold or the Z-score is less than the Z-score threshold, step S16 is then executed. The predetermined warning condition is not limited to this embodiment. In other embodiments, the predetermined warning condition further includes, for example, that the bone density estimation data is less than a bone density threshold value.

Finally, as shown in step S16, when the processing unit 5 determines that the bone density estimation data meets the predetermined warning condition, the processing unit 5 outputs a warning message via the output unit 4, the warning message indicating that the T score is less than the T The score threshold or the Z score is less than the Z score threshold to remind medical staff that the bone density of the target patient is abnormally low, thereby providing a warning to patients at high risk of osteoporosis.

It is added that although the present invention is used to estimate the bone density of a patient, it can also be used to estimate the bone mineral content (BMC) of the patient through similar technical means.

In summary, in the present invention, the processing unit 5 trains the first convolutional neural network model based on the training image data and the bone density data to generate the estimation model, so that the target patient only needs to be affected by the X The bone density estimation data can be generated based on the image data to be analyzed by the X-ray machine 1 after shooting by the X-ray machine 1, without the need for the dual-energy X-ray absorber 2 to detect, thereby reducing the patient’s radiation dose. Risks and the degree of dependence of medical units on expensive instruments such as the dual-energy X-ray absorber 2. In addition, the present invention uses the processing unit 5 to perform the predetermined image processing for each training image data and the image data to be analyzed, In order to reduce the noise interference by restricting the recognition part, the accuracy of the estimation can be further improved. Therefore, the objective of the invention can be achieved.

However, the above are only examples of the present invention. When the scope of implementation of the present invention cannot be limited by this, all simple equivalent changes and modifications made in accordance with the scope of the patent application of the present invention and the content of the patent specification still belong to Within the scope of the patent of the present invention.

100: electronic system 1: X-ray machine 2: Dual energy X-ray absorber 3: Input unit 4: output unit 5: Processing unit S01~S06: Step S11~S16: steps

Other features and effects of the present invention will be clearly presented in the embodiments with reference to the drawings, in which: FIG. 1 is a schematic diagram of a hardware connection relationship according to an embodiment of the present invention; Figure 2 is a flow chart of this embodiment, illustrating the steps of a method for generating a model for estimating bone density; and Fig. 3 is another flowchart of this embodiment, illustrating the steps of the method for estimating bone density.

S01~S06: Step

Claims (20)

A method for generating a model for estimating bone density is implemented by an electronic system, the electronic system includes an X-ray machine, a dual-energy X-ray absorber and a processing unit, the method includes: the X-ray machine photographs Multiple reference patients’ predetermined bones to generate multiple training image data; the dual-energy X-ray absorber detects the predetermined bones of the reference patients to generate multiple bone density data; and the processing unit is based on the training image data And the bone density data, train a first convolutional neural network model to generate an estimation model, the estimation model is used for the X-ray machine to photograph the predetermined bone of a target patient to be analyzed The imaging data estimates the bone density of the target patient. The method for generating an estimation model for estimating bone density according to claim 1, wherein the electronic system further includes an input unit, and the method is before generating the estimation model and after generating the training image data It also includes: for each training image data, the processing unit selects an image of the predetermined skeleton in the training image data according to a circle selection command received through the input unit; and the processing unit performs a cycle for each training image data Predetermined image processing, the predetermined image processing includes removing unselected images from the training image data; in the step of generating the estimation model, the processing unit is based on The training image data and the bone density data of the predetermined image processing are used to train the first convolutional neural network model to generate the estimation model. The method for generating an estimation model for estimating bone density as described in claim 2, after the step of performing the predetermined image processing, further includes: the processing unit according to the training image data that has undergone the predetermined image processing, A second convolutional neural network model is trained to generate a circle selection model, and the circle selection model is used to circle the image of the predetermined bone according to the image data to be analyzed. The method for generating an estimation model for estimating bone density according to claim 1, wherein the predetermined bone is a hip joint of a pelvis, a lumbar vertebra, a pelvis, a ulna, or a femur. A method for estimating bone density is implemented by an electronic system. The electronic system includes an X-ray machine, a dual-energy X-ray absorber and a processing unit. The method includes: the X-ray machine photographs multiple reference patients. A predetermined bone to generate multiple training image data; the dual-energy X-ray absorber detects the predetermined bone of the reference patient to generate multiple bone density data; the processing unit is based on the training image data and the bone density data , Training a first convolutional neural network model to generate an estimation model; the X-ray machine photographs a predetermined bone of a target patient to generate an image data to be analyzed; and the processing unit uses the image data to be analyzed The estimation model estimates the bone density of the target patient to generate a bone density estimation data. The method for estimating bone density according to claim 5, wherein the electronic system further includes an output unit, and after generating the bone density estimation data, the method further includes: the processing unit judging whether the bone density estimation data is Meets a predetermined warning condition; and when the processing unit determines that the bone density estimation data meets the predetermined warning condition, the processing unit outputs a warning message through the output unit. The method for estimating bone density according to claim 6, wherein the predetermined warning condition includes that a T score corresponding to the bone density estimation data is less than a T score threshold. The method for estimating bone density according to claim 6, wherein the predetermined warning condition includes that a Z score corresponding to the bone density estimation data is less than a Z score threshold. The method for estimating bone density according to claim 5, wherein the electronic system further includes an input unit, and before generating the estimation model and after generating the training image data, the method further includes: the processing unit for For each training image data, an image of the predetermined skeleton is circled from the training image data according to a circle selection command received through the input unit; and the processing unit performs a predetermined image processing for each training image data, the predetermined image The processing includes removing unselected images from the training image data; in the step of generating the estimation model, the processing unit is based on the training image data and the bone densities that have undergone the predetermined image processing Data, training the first convolutional neural network model to generate the estimation model; after the step of performing the predetermined image processing, the method further includes: the processing unit according to the training image data that has undergone the predetermined image processing, Training a second convolutional neural network model to generate a circle selection model; before generating the bone density estimation data and after generating the image data to be analyzed, the processing unit further includes: the processing unit uses the image data to be analyzed according to the image data to be analyzed. The circle selection model circle selects the image of the predetermined bone in the image data to be analyzed; and the processing unit performs the predetermined image processing on the image data to be analyzed, and the predetermined image processing includes the unselected image data in the image data to be analyzed Image removal; in the step of generating the bone density estimation data, the processing unit uses the estimation model to estimate the bone density of the target patient based on the image data to be analyzed that has undergone the predetermined image processing to generate the bone density Density estimation data. The method for estimating bone density according to claim 5, wherein the predetermined bone is a hip joint of a pelvis, a lumbar vertebra, a spine, a ulna, or a femur. An electronic system comprising: an X-ray machine; a dual-energy X-ray absorber; and a processing unit; the X-ray machine photographs a predetermined bone of multiple reference patients to generate multiple pens Training image data; the dual-energy X-ray absorber detects the predetermined bones of the reference patients to generate multiple bone density data; the processing unit trains a first convolution based on the training image data and the bone density data The neural network model generates an estimation model for estimating the bone density of the target patient based on a to-be-analyzed image data generated by the X-ray machine photographing the predetermined bone of a target patient. The electronic system according to claim 11, further comprising an input unit; for each training image data, the processing unit selects the image of the predetermined bone in the training image data according to a circle selection command received through the input unit The processing unit performs a predetermined image processing for each training image data, the predetermined image processing includes removing unselected images in the training image data; wherein, the processing unit is based on the predetermined image processing performed Waiting for the training image data and the bone density data to train the first convolutional neural network model to generate the estimation model. The electronic system according to claim 12, wherein the processing unit trains a second convolutional neural network model based on the training image data subjected to the predetermined image processing to generate a circle selection model, the circle selection model It is used to circle the image of the predetermined bone according to the image data to be analyzed. The electronic system according to claim 11, wherein the predetermined bone is a hip joint of a pelvis, a lumbar spine, a spine, an ulna, or a femur. An electronic system comprising: an X-ray machine; a dual-energy X-ray absorber; and a processing unit; the X-ray machine photographs a predetermined bone of a plurality of reference patients to generate multiple training image data; the dual-energy X-ray The absorptiometer detects the predetermined bones of the reference patients to generate multiple bone density data; the processing unit trains a first convolutional neural network model based on the training image data and the bone density data to generate an estimate Model; the X-ray machine photographs a predetermined bone of a target patient to generate an image data to be analyzed; the processing unit uses the estimation model to estimate the bone density of the target patient based on the image data to be analyzed to generate a bone density Estimated data. The electronic system according to claim 15, further comprising an output unit; the processing unit determines whether the bone density estimation data meets a predetermined warning condition; when the processing unit determines that the bone density estimation data meets the predetermined warning condition, The processing unit outputs a warning message through the output unit. The electronic system according to claim 16, wherein the predetermined warning condition includes that a T score corresponding to the bone density estimation data is less than a T score threshold. The electronic system according to claim 16, wherein the predetermined warning condition includes that a Z score corresponding to the bone density estimation data is less than a Z score threshold value. The electronic system according to claim 15, further comprising an input unit; for each training image data, the processing unit selects the image of the predetermined bone in the training image data according to a circle selection command received through the input unit The processing unit performs a predetermined image processing for each training image data, the predetermined image processing includes removing unselected images in the training image data; wherein, the processing unit is based on the predetermined image processing performed Training the first convolutional neural network model to generate the estimation model based on the training image data and the bone density data; wherein, the processing unit trains a first convolutional neural network model based on the training image data subjected to the predetermined image processing Two convolutional neural network models generate a circle selection model; the processing unit uses the circle selection model to circle out the image of the predetermined bone in the image data to be analyzed according to the image data to be analyzed; the processing unit performs the analysis on the image data to be analyzed The image data is subjected to the predetermined image processing, and the predetermined image processing includes removing the unselected images from the image data to be analyzed; wherein, the processing unit uses the image data to be analyzed based on the image data to be analyzed that has undergone the predetermined image processing. The estimation model estimates the bone density of the target patient to generate the bone density estimation data. The electronic system according to claim 15, wherein the predetermined bone is a hip joint of a pelvis, a lumbar spine, a spine, an ulna, or a femur.

Images