An osteoporosis and osteoporotic vertebral compression fracture identification model, training and prediction method

Abstract

The application provides an osteoporosis and osteoporotic vertebral compression fracture identification model, a training and prediction method, and a technical scheme for predicting and identifying osteoporosis and osteoporotic vertebral compression fracture based on deep learning technology and repeated use of chest CT scan images of patients for lung nodule screening. The first input data of the model includes 2D image information data cut out from a chest CT image slice and 2D segmentation mask information data cut out from a vertebral mask slice; the second input data includes patient age and gender clinical information; a multi-layer feature extraction layer of the model is used for processing the first input data, and an ultimate feature extraction graph is obtained after the multi-layer feature extraction layer; a prediction layer is used for splicing the processed ultimate feature extraction graph and the second input data, and finally, two full connection layers arranged in parallel are used for osteoporosis category prediction and vertebral compression fracture category prediction, respectively.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical image recognition, in particular to a kind of osteoporosis and osteoporotic vertebral compression fracture identification model related technology. BACKGROUND

[0002] With the aggravation of population aging, osteoporosis has become an important public health challenge in China. Osteoporosis patients usually have no obvious symptoms before fracture, therefore, it is of great significance to identify these patients and take appropriate treatment measures in time to prevent further bone loss. At present, bone mineral density (BMD) screening test is the only method to identify osteoporosis patients, but the coverage of such screening is limited.

[0003] In recent years, a large number of chest CT scans are performed annually for lung nodule screening. If these chest CT images are used for opportunistic osteoporosis screening, it provides a huge opportunity for early detection of osteopenia and osteoporosis without additional radiation.

[0004] Vertebral compression fracture (OVCF) as the most common type of osteoporotic fracture of osteoporosis, mainly relies on the way of manual film reading by radiologists to judge each segment of vertebral body, not only time-consuming, but also the accuracy rate is related to the experience of radiologists, therefore, it is of clinical significance to automatically judge each segment of vertebral compression fracture (OVCF) based on vertebral CT images. Using Res2Net module as the basis of feature extraction enhances the multi-scale representation ability, so as to better capture the detailed features at different levels. SUMMARY

[0005] The present application provides an osteoporosis and osteoporotic vertebral compression fracture identification model, training and prediction method, based on deep learning technology, reusing chest CT scan images of patients for lung nodule screening, a technical solution for predicting and identifying osteoporosis and osteoporotic vertebral compression fracture.

[0006] To achieve the above-mentioned purpose, the present application adopts the following technical solution:

[0007] The present application provides an osteoporosis and osteoporotic vertebral compression fracture identification model, the model is used for predicting osteoporosis and vertebral compression fracture according to first input data and second input data;

[0008] The first input data includes 2D image information data cut from chest CT image slices, and 2D segmentation mask information data cut from vertebral mask slices.

[0009] The second input data includes patient age, gender and clinical information;

[0010] The model comprises a plurality of feature extraction layers and a prediction layer;

[0011] The plurality of feature extraction layers are used to process the first input data, and the final feature extraction graph is obtained after the plurality of feature extraction layers;

[0012] The prediction layer comprises an adaptive mean pooling layer and two parallel full connection layers, the adaptive mean pooling layer is used to splice the final feature extraction graph after processing and the second input data, and finally the two parallel full connection layers are used to respectively perform osteoporosis category prediction and vertebral compression fracture category prediction.

[0013] The application also provides a training method of an osteoporosis and osteoporotic vertebral compression fracture identification model, and the method comprises the following steps:

[0014] A training sample set is constructed, and each sample in the training sample set comprises a chest CT image and patient age, gender and clinical information;

[0015] The first input data is preprocessed, and the first input data comprises 2D image information data cut from a chest CT image slice and 2D segmentation mask information data cut from a vertebral mask slice;

[0016] The second input data is preprocessed, and the second input data comprises patient age, gender and clinical information of the sample;

[0017] In the training stage, the manually marked osteoporotic vertebral compression fracture category information of each vertebra and the osteoporosis information extracted from a patient's dual-energy X-ray bone density detection report are simultaneously used as supervision information for model training; the first input data and the second input data are input into the osteoporosis and osteoporotic vertebral compression fracture identification model for training;

[0018] In the training stage of the model, the prediction confidence of all CT image slices in each vertebra is counted, and the preliminary prediction results of osteoporosis and osteoporotic vertebral compression fracture of each vertebra are obtained by averaging the confidence; by comparing the prediction accuracy of osteoporosis of all vertebrae in the model training process, the n vertebrae with the highest accuracy are screened out for confidence fusion, as the final osteoporosis prediction result;

[0019] The inference stage of the model fuses the osteoporosis confidence of all slices of the n vertebrae selected in the training stage as the final osteoporosis prediction result; and fuses the osteoporosis vertebral compression fracture confidence of all slices of each vertebra as the osteoporosis vertebral compression fracture prediction result of the corresponding vertebra.

[0020] Meanwhile, the application further provides an osteoporosis and osteoporosis vertebral compression fracture prediction method based on the above model, which comprises the following steps:

[0021] Obtaining the chest CT image of the target, patient age, and gender clinical information;

[0022] Pretreating first input data, which comprises 2D image information data cut out from the chest CT image slices and 2D segmentation mask information data cut out from the vertebra mask slices;

[0023] Pretreating second input data, which comprises the patient age, gender, and clinical information of the sample;

[0024] Inputting the first input data and the second input data into the trained osteoporosis and osteoporosis vertebral compression fracture identification model to obtain the preliminary prediction result of the osteoporosis and osteoporosis vertebral compression fracture of each vertebra;

[0025] Fusing the osteoporosis confidence of all slices of the n vertebrae with the highest osteoporosis accuracy selected in the training stage of the model as the final osteoporosis prediction result;

[0026] Fusing the osteoporosis vertebral compression fracture confidence of all slices of each vertebra as the osteoporosis vertebral compression fracture prediction result of the corresponding vertebra.

[0027] The present application realizes intelligent solution for CT images based on deep learning technology to predict osteoporosis and osteoporosis vertebral compression fracture. The present application has the advantage that it can use the chest CT scan image of the patient for lung nodule screening to predict and identify osteoporosis and osteoporosis vertebral compression fracture. The application of the present application can effectively replace the commonly used methods of bone density screening test, including dual-energy X-ray bone density measurement (DXA) and quantitative CT (QCT), to avoid excessive radiation affecting the body and greatly reduce the cost. Moreover, the present application can make up for the problem in the prior art that the osteoporosis vertebral compression fracture (OVCF) is mainly judged by image doctors through manual film reading, and further improve the efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0028] Figure 1The structural diagram of the osteoporosis and osteoporotic vertebral compression fracture identification model of the application.

[0029] Figure 2 The application example diagram of the osteoporosis and osteoporotic vertebral compression fracture identification model of the application.

[0030] Figure 3 The example diagram of the vertebral sagittal central layer of the chest CT image in the application. DETAILED DESCRIPTION

[0031] The application will be further described below in conjunction with the drawings and specific embodiments.

[0032] The existing osteoporosis diagnosis methods mainly include dual-energy X-ray absorptiometry (DXA) and quantitative computed tomography (QCT).

[0033] Dual-energy X-ray absorptiometry (DXA): DXA is a low-dose X-ray imaging technique that uses two different energy peaks to distinguish between soft tissue and bone. By measuring bone mineral content (mainly calcium), bone density can be calculated.

[0034] This method is commonly used for bone density measurement in the hip, spine and other areas, as it provides accurate results for these regions.

[0035] DXA scanning is fast and safe, with very low radiation exposure, and is considered the gold standard for bone density measurement.

[0036] However, when measuring bone density in the spine, degenerative changes in the vertebral body, arteriosclerosis or vertebral fractures can interfere with the accuracy of the results.

[0037] Quantitative computed tomography (QCT):

[0038] QCT is a special computed tomography (CT) technique that provides three-dimensional images and separately assesses the bone density of cortical bone (the outer hard part of the bone) and cancellous bone (the softer, sponge-like part inside the bone).

[0039] Compared to DXA, QCT can more accurately detect changes in bone density in complex structures such as the spine, and is not affected by spinal degenerative diseases or arteriosclerosis.

[0040] The radiation dose of QCT is significantly higher, which is an important consideration for patients who need frequent monitoring.

[0041] Cost issue: QCT is usually much more expensive than DXA, which may limit its widespread application.

[0042] Equipment and technical requirements: not all medical institutions are equipped with equipment that can perform QCT scans, and more professional skills are required to operate this technology.

[0043] And osteoporotic vertebral compression fracture (OVCF) refers to the compression fracture of the vertebral body of the spine caused by osteoporosis. DXA and QCT can measure bone density to help assess the degree of osteoporosis, thereby indirectly indicating the risk of compression fracture. But still need to cooperate with CT and magnetic resonance imaging (MRI) and other screening tools. The problem of relying on the way of artificial film reading by radiologists to judge each segment of the vertebral body is low efficiency and high cost.

[0044] After the epidemic, chest CT scans for lung nodule screening are increasingly valued. Therefore, if chest CT images can be reused for osteoporotic vertebral compression fracture identification and auxiliary diagnosis, it will greatly improve efficiency, reduce diagnosis and treatment costs, and reduce the risk of patient radiation.

[0045] The application provides an osteoporosis and osteoporotic vertebral compression fracture identification model, which is used for osteoporosis category prediction and vertebral compression fracture category prediction according to first input data and second input data;

[0046] The first input data includes 2D image information data cut from a chest CT image slice and 2D segmentation mask information data cut from a vertebral mask slice.

[0047] The second input data includes patient age, gender and clinical information.

[0048] The pre-processing method of the first input data includes the following steps:

[0049] The three-dimensional chest CT image is input into a mature vertebral segmentation model to obtain the CT image corresponding to the vertebral segmentation mask information, and the layer thickness of the CT image is interpolated to the same preset layer thickness value, which is 1 mm in the embodiment.

[0050] Based on the segmentation mask information of each segment of the vertebral body, the sagittal center layer of each segment of the vertebral body and the center coordinates of the mask corresponding to the sagittal center layer are calculated. The sagittal center layer is usually the layer with the most complete and clear vertebral structure.

[0051] In the embodiment of the application, the sagittal center layer of each segment of the vertebral body and the center coordinates of the mask corresponding to the sagittal center layer are calculated, and the specific processing method of this step is:

[0052] For example, Figure 2As shown, the coordinate system of CT image and mask information is set, x is the left-right direction (sagittal plane separates left and right), y is the front-back direction (coronal plane separates front and back), z is the up-down direction (transverse plane separates up and down), and the sagittal center layer is X. The mask is a three-dimensional array of (x, y, z).

[0053] The sum of the elements in the mask on the (y, z) axis is calculated to obtain a one-dimensional array sums.

[0054] By dividing sums by the sum of sums, normalized_sums is obtained. This step converts sums into a probability distribution form normalized_sums, so that all elements normalized_sums add up to 1.

[0055] Calculate the cumulative sum pos of values from 0 to mask.shape[x] but not including mask.shape[x], where each value is multiplied by its corresponding normalized_sums, and pos is rounded to the nearest integer. The sagittal center layer position X is obtained.

[0056] The calculation formula is:

[0057]

[0058] Where:

[0059] N is mask.shape[0], i.e. the number of layers along the left-right direction;

[0060] i represents the index of the i-th layer (counting from 0);

[0061] normalized_sums[i] is the normalized weight of the i-th layer.

[0062] Get all CT image slices within a preset distance from the sagittal center layer, as well as the corresponding 2D mask slices. In this embodiment, the preset distance is 5mm.

[0063] The preset distance is an integer multiple of the preset layer thickness value. The center coordinates of the mask are used to divide the CT image slice blocks of fixed size. Each slice block combines its adjacent upper and lower slice blocks to form a fixed-size three-channel 2D image information data. The 2D image information data is an array with a size of [64, 64, 3].

[0064] Obtain all 2D mask slices within a preset distance from the center layer of the sagittal plane, and cut out a fixed size of 2D segmentation mask information data from the vertebral segmentation mask according to the center coordinates of the obtained mask.

[0065] Splice the 2D image information data and the 2D segmentation mask information data of each vertebral body corresponding slice into four channels, and obtain a [4, 64, 64] data block through channel adjustment, which is the first input data.

[0066] As shown in Figure 3 The example of the chest CT image is a sagittal center layer. The center coordinates [y, z] are marked. First, select multiple sagittal layers in the range such as [x-2, x-1, x, x+1, x+2], cut a [64, 64] size block according to the center point [y, z] on the layer, then combine the adjacent layers above and below, and take the x-2 layer as an example, that is, [x-3, x-2, x-1] 3 layers form a [64x64x3] image information block, and the [64x64x1] size mask corresponding to the x-2 layer cut block is used as the segmentation information.

[0067] Each slice combines its upper and lower layers to form a 3-channel picture, which not only retains the information of a single slice, but also increases the spatial correlation of adjacent slices above and below, providing more rich local context information to the deep learning model. The center coordinates [x, y] are cut out from the original picture and the mask to form fixed size [64, 64, 3] image information data and [64, 64, 1] vertebral mask information data, which ensures the consistency and standardization of the input data, so that the model can stably receive the same format data, which is conducive to the convergence and generalization performance in the training process.

[0068] The normalized sums array is used to calculate the cumulative sum pos, and the position x of the sagittal center layer is obtained by rounding. Then the center coordinates [y, z] are obtained from the layer mask, which ensures the accurate positioning of each vertebral segment in space. Based on this center position, the region of interest (ROI) related to the vertebral body can be effectively located and extracted, which improves the accuracy of the model in identifying target structures.

[0069] As shown in Figure 1 The osteoporosis and osteoporotic vertebral compression fracture recognition model of the application includes a plurality of feature extraction layers and a prediction layer.

[0070] The multi-layer feature extraction layer is used to process the first input data, and the final feature extraction graph is obtained after passing through the multi-layer feature extraction layer.

[0071] The prediction layer includes an adaptive mean pooling layer and two full-connection layers arranged in parallel, the adaptive mean pooling layer is used for splicing the final feature extraction graph after processing with the second input data, and finally the two full-connection layers arranged in parallel are used for respectively performing osteoporosis category prediction and vertebral compression fracture category prediction.

[0072] The feature extraction layer of the osteoporosis and osteoporotic vertebral compression fracture identification model includes three combinations of Res2Net module groups and pooling modules, and a fourth Res2Net module group.

[0073] The feature extraction layer mainly adopts the Res2Net module, which adopts a simple and effective multi-scale processing method, and improves the multi-scale representation ability through multiple available receptive fields at a finer granularity level.

[0074] The prediction layer is a multi-task prediction branch, and the two full-connection layers arranged in parallel are used for respectively performing osteoporosis category prediction and vertebral compression fracture category prediction.

[0075] The design of the feature extraction layer of the model considers the importance of the central feature region, and the final feature extraction graph is obtained by splicing and fusing the feature graphs at different stages, which helps to improve the expressiveness of the model.

[0076] The data post-processing mode of the model includes:

[0077] In the training stage of the model, the prediction confidence of all CT image slices in each vertebral body is counted, and the preliminary prediction results of osteoporosis and osteoporotic vertebral compression fracture of each vertebral body are obtained by accumulating and averaging the confidence; by comparing the accuracy of the predicted osteoporosis of all vertebral bodies in the model training process, the n vertebral bodies with the highest accuracy are selected for confidence fusion as the final osteoporosis prediction result.

[0078] In the inference stage of the model, the osteoporosis confidence of all slices of the n vertebral bodies selected in the training stage is fused as the final osteoporosis prediction result; the osteoporotic vertebral compression fracture confidence of all slices of each vertebral body is fused as the osteoporotic vertebral compression fracture prediction result of the corresponding vertebral body.

[0079] In the training stage, the prediction confidence of each vertebral segment slice is counted and fused, and the vertebral body with high accuracy is selected for result fusion, ensuring the reliability of the final prediction result. The same strategy is adopted in the reasoning stage to further ensure the consistency and stability of the output result.

[0080] As shown in Figure 2 A training method of an osteoporosis and osteoporotic vertebral compression fracture identification model, the method comprising the following steps:

[0081] A training sample set is constructed, each sample in the training sample set containing its chest CT image and patient age, gender and clinical information;

[0082] The first input data, which includes 2D image information data cut out from the chest CT image slice and 2D segmentation mask information data cut out from the vertebral body mask slice, is preprocessed. The preprocessing method of the first input data has been described in detail in the foregoing of the present embodiment, and will not be repeated here.

[0083] The second input data, which includes the patient's age, gender and clinical information of the sample, is preprocessed.

[0084] In the training stage, the artificially labeled osteoporotic vertebral compression fracture category information of each vertebral body and the osteoporosis information extracted from the patient's dual-energy X-ray bone density detection report are used as supervision information for model training; the first input data and the second input data are input into the osteoporosis and osteoporotic vertebral compression fracture identification model for training;

[0085] In the training stage of the model, the prediction confidence of all CT image slices in each vertebral segment is counted, and the preliminary prediction results of osteoporosis and osteoporotic vertebral compression fracture of each vertebral segment are obtained by accumulating and averaging the confidence; by comparing the prediction accuracy of osteoporosis of all vertebral bodies in the model training process, the n vertebral bodies with the highest accuracy are selected for confidence fusion as the final osteoporosis prediction result;

[0086] In the reasoning stage of the model, the osteoporosis confidence of all slices of the n vertebral bodies selected in the training stage is fused as the final osteoporosis prediction result; the osteoporotic vertebral compression fracture confidence of all slices of each vertebral body is fused as the osteoporotic vertebral compression fracture prediction result of the corresponding vertebral body.

[0087] As shown in Figure 2As shown, the present application also provides a method for predicting osteoporosis and osteoporotic vertebral compression fractures. The method is based on the osteoporosis and osteoporotic vertebral compression fracture model described above, and realizes the inference prediction of osteoporosis and osteoporotic vertebral compression fractures using chest CT images.

[0088] Obtain the chest CT image of the target, the patient's age, gender and clinical information;

[0089] Preprocess the first input data, which includes 2D image information data cut from the chest CT image slices, and 2D segmentation mask information data cut from the vertebral mask slices;

[0090] Preprocess the second input data, which includes the patient's age, gender and clinical information of the sample;

[0091] Input the first input data and the second input data into the osteoporosis and osteoporotic vertebral compression fracture recognition model trained by the training method described above, to obtain the preliminary prediction results of osteoporosis and osteoporotic vertebral compression fractures for each vertebral body;

[0092] Fuse the osteoporosis confidence of all slices of the n vertebrae with the highest accuracy of osteoporosis selected in the training stage of the fusion model, as the final osteoporosis prediction result;

[0093] Fuse the osteoporotic vertebral compression fracture confidence of all slices of each vertebral body as the osteoporotic vertebral compression fracture prediction result of the corresponding vertebral body.

[0094] One advantage of the present application is that it can use chest CT scan images for lung nodule screening to predict and identify osteoporosis and osteoporotic vertebral compression fractures. The application can effectively replace the commonly used methods for bone density screening tests, including dual-energy X-ray absorptiometry (DXA) and quantitative CT (QCT), avoiding excessive radiation exposure and significantly reducing costs. It also addresses the problem of relying on image physicians to manually review each vertebral segment for osteoporotic vertebral compression fractures (OVCF), further improving efficiency.

[0095] The clinical data involved in the present application are obtained with full authorization, and the collection, use and processing of relevant information need to comply with relevant national and regional laws, regulations and standards.

Claims

1. An osteoporosis and osteoporotic vertebral compression fracture identification model, characterized in that, The model is used for osteoporosis category prediction and vertebral compression fracture category prediction according to first input data and second input data; The first input data includes 2D image information data cut out from a chest CT image slice and 2D segmentation mask information data cut out from a vertebral mask slice; The second input data includes patient age and gender clinical information; The model includes a plurality of feature extraction layers and a prediction layer; The plurality of feature extraction layers are used for processing the first input data, and a final feature extraction graph is obtained after the plurality of feature extraction layers; The prediction layer includes an adaptive mean pooling layer and two parallel full connection layers, the adaptive mean pooling layer is used for splicing the final feature extraction graph after processing and the second input data, and finally the two parallel full connection layers are used for osteoporosis category prediction and vertebral compression fracture category prediction, respectively; The pre-processing method of the first input data includes the following steps: input a three-dimensional chest CT image into a mature vertebral segmentation model, obtain the vertebral segmentation mask information corresponding to the CT image, and interpolate the layer thickness of the CT image to the same preset layer thickness value; based on the vertebral segmentation mask information of each segment, the sagittal center layer of each segment is calculated, and the center coordinates of the mask corresponding to the sagittal center layer are calculated; all CT image slices within a preset distance from the sagittal center layer are obtained, and 2D mask slices corresponding to the CT image slices are obtained; the preset distance is an integer multiple of the preset layer thickness value, and the mask center coordinates are used to segment a fixed-size CT image slice block; each slice block combines its adjacent upper and lower slice blocks to form a fixed-size three-channel 2D image information data; all 2D mask slices within a preset distance from the sagittal center layer are obtained, and fixed-size one-channel 2D segmentation mask information data is cut out from the vertebral segmentation mask according to the obtained mask center coordinates; the 2D image information data and the 2D segmentation mask information data corresponding to each vertebral slice are spliced into four-channel first input data.

2. The osteoporosis and osteoporotic vertebral compression fracture identification model according to claim 1, wherein: the preset layer thickness value is 1mm, and the preset distance is 5mm; the 2D image information data is an array with a size of [64, 64, 3]; the 2D segmentation mask information data is an array with a size of [64, 64, 1]; the first input data is a data block with a size of [64, 64, 4] obtained by splicing the 2D image information data and the 2D segmentation mask information data in the third dimension, and a data block with a size of [4, 64, 64] is obtained by adjusting the channel.

3. The model for identifying osteoporosis and osteoporotic vertebral compression fractures according to claim 1, wherein The feature extraction layer includes three combinations of Res2Net module groups and pooling modules, and a fourth Res2Net module group; the feature blocks corresponding to the center feature regions of the results of the processing of the first input data by the three combinations of Res2Net module groups and pooling modules are spliced and fused with the feature maps after the fourth Res2Net module group, to obtain a final feature extraction map.

4. The model for identifying osteoporosis and osteoporotic vertebral compression fractures according to claim 1, wherein The prediction layer is a multi-task prediction branch, and the two full connection layer prediction branches arranged in parallel perform osteoporosis category prediction and vertebral compression fracture category prediction, respectively.

5. The model for identifying osteoporosis and osteoporotic vertebral compression fractures according to claim 1, wherein The data post-processing mode of the model comprises: In the training stage of the model, the prediction confidence of all CT image slices in each vertebral body is counted, and the preliminary prediction results of osteoporosis and the prediction results of osteoporotic vertebral compression fracture of each vertebral body are obtained by accumulating and averaging the confidence; the highest accuracy of n vertebral bodies is selected by comparing the prediction accuracy of osteoporosis of all vertebral bodies in the model training process, and the confidence fusion is used as the final prediction result of osteoporosis; In the inference stage of the model, the osteoporosis confidence of all slices of the n vertebral bodies selected in the training stage is fused as the final prediction result of osteoporosis; the osteoporotic vertebral compression fracture confidence of all slices of each vertebral body is fused as the prediction result of the corresponding vertebral body.

6. A method of training an osteoporosis and osteoporotic vertebral compression fracture identification model, characterized by, The method comprises the following steps: A training sample set is constructed, and each sample in the training sample set comprises a chest CT image and patient age, gender and clinical information; The first input data is preprocessed, and the first input data comprises 2D image information data cut from a chest CT image slice and 2D segmentation mask information data cut from a vertebral body mask slice; The second input data is preprocessed, and the second input data comprises patient age, gender and clinical information of the sample; In the training stage, the osteoporotic vertebral compression fracture category information of each vertebral body marked by artificial marking and the osteoporosis information extracted from a patient dual-energy X-ray bone density detection report are used as the supervision information for model training; the first input data and the second input data are input into the osteoporosis and osteoporotic vertebral compression fracture identification model of any one of claims 1-5 for training; In the training stage of the model, the prediction confidence of all CT image slices in each vertebral body is counted, and the preliminary prediction results of osteoporosis and the prediction results of osteoporotic vertebral compression fracture of each vertebral body are obtained by accumulating and averaging the confidence; the highest accuracy of n vertebral bodies is selected by comparing the prediction accuracy of osteoporosis of all vertebral bodies in the model training process, and the confidence fusion is used as the final prediction result of osteoporosis; Inference stage of the model, by fusing the osteoporosis confidence of all slices of the n vertebrae selected in the training stage as the final osteoporosis prediction result; by fusing the osteoporosis vertebral compression fracture confidence of all slices of each vertebra as the corresponding vertebra osteoporosis vertebral compression fracture prediction result; The preprocessing method of the first input data comprises the following steps: Input the three-dimensional chest CT image into a mature vertebra segmentation model, obtain the vertebra segmentation mask information corresponding to the CT image, and interpolate the layer thickness of the CT image to the same preset layer thickness value; Based on the vertebra segmentation mask information of each segment, the sagittal center layer of each segment and the center coordinates of the mask corresponding to the sagittal center layer are calculated; All CT image slices within a preset distance from the sagittal center layer are obtained, and 2D mask slices corresponding to the CT image slices are obtained; The preset distance is an integer multiple of the preset layer thickness value, and the center coordinates of the mask are used to segment a fixed-size CT image slice block; each slice block combines its adjacent upper and lower slice blocks to form a fixed-size three-channel 2D image information data; All 2D mask slices within a preset distance from the sagittal center layer are obtained, and a fixed-size one-channel 2D segmentation mask information data is cut from the vertebra segmentation mask according to the center coordinates of the mask; The 2D image information data and the 2D segmentation mask information data corresponding to each vertebra slice are spliced into four-channel first input data.

7. A method for predicting osteoporosis and osteoporotic vertebral compression fractures, characterized by, The method comprises the following steps: Obtain the chest CT image of the target, the patient's age, and the gender clinical information; Preprocess the first input data, which contains 2D image information data cut from the chest CT image slices and 2D segmentation mask information data cut from the vertebra mask slices; Preprocess the second input data, which contains the patient's age, gender, and clinical information of the sample; Input the first input data and the second input data into the osteoporosis and osteoporosis vertebral compression fracture recognition model trained by the training method of claim 6 to obtain the preliminary prediction results of the osteoporosis and osteoporosis vertebral compression fracture of each vertebra; By fusing the osteoporosis confidence of all slices of the n vertebrae with the highest accuracy of osteoporosis selected in the training stage as the final osteoporosis prediction result; By fusing the osteoporosis vertebral compression fracture confidence of all slices of each vertebra as the corresponding vertebra osteoporosis vertebral compression fracture prediction result; The preprocessing method of the first input data comprises the following steps: Input the three-dimensional chest CT image into a mature vertebra segmentation model, obtain the vertebra segmentation mask information corresponding to the CT image, and interpolate the layer thickness of the CT image to the same preset layer thickness value; Based on the vertebra segmentation mask information of each segment, the sagittal center layer of each segment and the center coordinates of the mask corresponding to the sagittal center layer are calculated; All CT image slices within a preset distance from the sagittal center layer are obtained, and 2D mask slices corresponding to the CT image slices are obtained; The preset distance is an integer multiple of the preset layer thickness value, and the center coordinates of the mask are used to segment a fixed-size CT image slice block; each slice block combines its adjacent upper and lower slice blocks to form a fixed-size three-channel 2D image information data; All CT image slices within a preset distance from a center layer of a sagittal plane and 2D mask slices corresponding to the CT image slices are acquired; The preset distance is an integer multiple of a preset layer thickness value, and a CT image slice block of a fixed size is divided based on a center coordinate of the mask; each slice block combines its adjacent upper and lower slice blocks to form 2D image information data of a fixed size and three channels; All 2D mask slices within a preset distance from a center layer of a sagittal plane are acquired, and a 2D mask information data of a fixed size and one channel is cut from the mask based on the center coordinate of the mask; The 2D image information data and the 2D mask information data of each slice corresponding to a vertebra are spliced into first input data of four channels.

Images