X-ray chest radiography spine offset detection method and system based on prior probability graph

Abstract

The invention relates to an X-ray chest radiography spine offset detection method and system based on a prior probability graph, and the method comprises the steps: obtaining a to-be-detected X-ray chest radiography image, and carrying out the normalization of the transverse pixels of the image; obtaining a corresponding prior probability value through linear interpolation in pre-constructed one-dimensional prior probability distribution based on the normalized position of each column of transverse pixels of the image, constructing a one-dimensional array, copying the one-dimensional array in the vertical direction, and constructing a two-dimensional prior probability graph; inputting an X-ray chest radiograph image to be detected into the GPU accelerated segmentation network to obtain a spine binary mask; and based on the two-dimensional prior probability graph, calculating an average prior probability and an offset probability in the mask area of the spine binary mask, comparing the average prior probability and the offset probability with a preset offset threshold value, if the offset probability is greater than the offset threshold value, determining that the spine is offset, otherwise, determining that the spine is basically centered. Compared with the prior art, the method has the advantages that the offset detection result has better interpretability, stability and quantificaiton.

Description

A method and system for detecting spinal deviation on chest X-ray based on prior probability maps Technical Field

[0001] This invention relates to the field of medical image segmentation technology for the spine, and in particular to a method and system for detecting spinal deviation on chest X-rays based on prior probability maps. Background Technology

[0002] Existing chest X-ray positioning quality control methods are a crucial step in clinical imaging examinations. Their core function is to determine whether the patient's position is compliant during chest X-ray imaging, particularly whether the spine is centered, to ensure image quality meets the needs of subsequent disease diagnosis. The main implementation paths fall into two categories: geometric distance-based methods, which calculate the lateral geometric distance between the spine's center point and the image center as the criterion for judging whether the position is normal; and black-box classification model-based methods, which train a deep learning classification model to directly output a qualitative result of "normal position" or "abnormal position," without requiring explicit judgment criteria.

[0003] Chinese patent CN113017675B proposes an automatic method and system for measuring the pose of an anteroposterior chest radiograph. The method involves acquiring the subject's anteroposterior chest radiograph using a detector, extracting the outer contour of the thoracic cavity and the central axis of the spine from the radiograph using a trained deep learning network, and obtaining the relative deflection angle δ between the subject and the detector through an asymmetric calculation step based on the outer contour of the thoracic cavity and the central axis of the spine. The subject's pose is then determined based on δ, and the chest radiograph pose is corrected. However, this method, through asymmetric calculation, can only reflect local positional deviations and cannot reflect the overall offset of the spinal coverage area, and its adaptability is weak.

[0004] Existing methods suffer from three core shortcomings, failing to meet clinical requirements for the stability, quantification, and adaptability of quality control results: 1. Lack of utilization of statistical priors regarding the lateral position of the spine: Existing methods rely solely on single geometric rules or implicit learning from black-box models, resulting in insufficient stability of quality control results and a lack of interpretability in the judgment logic. For example, black-box models cannot explain why an abnormality is determined. 2. Inability to intuitively quantify the degree of offset: Geometric distance-based methods can only reflect local positional deviations and cannot reflect the offset of the overall spinal coverage area. Black-box classification models only output qualitative labels of normal or abnormal, without specific offset indicators, making it difficult for technicians to determine whether the offset is within an acceptable range and failing to provide accurate references for subsequent re-examinations. 3. Insufficient adaptability to different equipment and field of view sizes: Different hospitals use different imaging equipment and field of view settings, resulting in varying chest X-ray resolutions and image sizes. Existing methods require setting separate judgment thresholds for each size, resulting in poor universality and difficulty in large-scale application across multiple devices and scenarios. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art by providing a method and system for detecting spinal deviation on chest X-rays based on prior probability maps, so that the deviation detection results have better interpretability, stability and quantification.

[0006] The objective of this invention can be achieved through the following technical solution: a method for detecting spinal deviation on X-ray chest radiographs based on a prior probability map, the method comprising: acquiring an X-ray chest radiograph image to be detected; normalizing the horizontal pixels of the image; based on the normalized position of each column of horizontal pixels in the image, obtaining the corresponding prior probability value in a pre-constructed one-dimensional prior probability distribution through linear interpolation, and constructing a one-dimensional array; copying the one-dimensional array vertically to construct a two-dimensional prior probability map; inputting the X-ray chest radiograph image to be detected into a pre-trained GPU-accelerated segmentation network, and performing thresholding or argm... The ax operation retrieves a binary mask of the spine from the probability map output by the GPU-accelerated segmentation network. Based on the two-dimensional prior probability map, the average prior probability and offset probability within the mask region of the spine binary mask are calculated and compared with a preset offset threshold. If the offset probability is greater than the offset threshold, it is determined to be a spinal offset; otherwise, it is determined to be a basically centered spine. The comparison result is output, and the average prior probability and offset probability are superimposed as percentages and displayed below the chest X-ray image. Mask regions with low prior probabilities are marked in dark and labeled with text prompts, thus completing the detection of spinal offset on the chest X-ray.

[0007] Furthermore, the two-dimensional prior probability graph is stored using a floating-point matrix, with values ​​ranging from [value range missing]. .

[0008] Furthermore, the pre-construction process of the one-dimensional prior probability includes: acquiring a set of chest X-ray images whose posture has been manually confirmed to be normal; labeling the spinal region in each chest X-ray image using a manual or GPU-accelerated segmentation network to obtain the corresponding binary mask; traversing the set of chest X-ray images and its binary mask to obtain the leftmost and rightmost pixel coordinates of the spinal region in the width direction, and normalizing them to an interval. Obtain the leftmost coordinate of the normalized coordinate. with optimal coordinates Based on preset accuracy parameters, set the step size, and use the set step size to... Discretize within the interval to obtain discrete positions Initialize probability Iterate through the pixels of each chest X-ray image, based on all pixels in the image that satisfy... of The probability of each pixel at a given location is calculated by accumulating the probabilities. ; the probability corresponding to each position according to The numerical sorting is used to obtain the ordered sequence as a one-dimensional prior probability distribution.

[0009] Furthermore, in the binary mask, the value is 1 when the pixel is located in the spine region, and 0 otherwise.

[0010] Furthermore, the X-ray chest images in the aforementioned chest image set have different resolutions.

[0011] Furthermore, after obtaining the one-dimensional prior probability distribution, a point with a probability value of 0 is added at the rightmost position.

[0012] Furthermore, when obtaining the corresponding prior probability value through linear interpolation in the pre-constructed one-dimensional prior probability distribution, if the normalized position of a horizontal pixel in the image falls between two probability values ​​in the one-dimensional prior probability, then its prior probability value is obtained through a one-dimensional calculation formula, which is: in, This represents the normalized position of the horizontal pixels in the image. Position in one-dimensional prior probability The corresponding probability value, Position in one-dimensional prior probability The corresponding probability value, For position The corresponding prior probability value.

[0013] Furthermore, before the X-ray chest image to be detected is input into the pre-trained GPU-accelerated segmentation network, preprocessing operations such as window width and window level adjustment, normalization, and scale transformation are performed.

[0014] Furthermore, the pre-trained GPU-accelerated segmentation network is a segmentation network implemented based on PyTorch.

[0015] Further, the system includes: a data preprocessing module: acquiring the chest X-ray image to be detected, normalizing and preprocessing the horizontal pixels of the image; a priori construction module: generating and saving a one-dimensional prior probability using normal body position samples, loading the one-dimensional prior probability when the system starts, and dynamically constructing a two-dimensional prior probability map for chest X-ray images of different widths to be detected; the specific process of constructing the two-dimensional prior probability map includes: based on the normalized position of each column of horizontal pixels in the image, obtaining the corresponding prior probability value through linear interpolation in the pre-constructed one-dimensional prior probability distribution, and constructing a one-dimensional array; copying the one-dimensional array in the vertical direction to construct the two-dimensional prior probability map; and a segmentation module: segmenting the chest X-ray image to be detected... The input is a pre-trained GPU-accelerated segmentation network. Through thresholding or argmax operations, a binary mask of the spine is obtained from the probability map output by the GPU-accelerated segmentation network. The offset probability calculation module calculates the average prior probability and offset probability within the mask region of the spine binary mask based on the two-dimensional prior probability map. The judgment and display module compares the offset probability with a preset offset threshold. If the offset probability is greater than the offset threshold, it is determined to be a spinal offset; otherwise, it is determined to be a basically centered spine. The average prior probability and offset probability are superimposed as percentages and displayed below the chest X-ray image. Mask regions with low prior probabilities are marked in dark colors with accompanying text prompts, thus completing the detection of spinal offset on the chest X-ray.

[0016] Compared with existing technologies, the beneficial effects of this invention include: 1. Based on a pre-constructed one-dimensional prior probability distribution, this invention maps the prior probability to each column of pixels according to the width of any image to be detected, and vertically replicates the column probabilities to construct a two-dimensional probability map that matches the image size. Then, by overlapping the spinal mask with the two-dimensional probability map, the offset probability is calculated to determine whether the spine is in a normal position. This achieves the purpose of comprehensively reflecting the overall positional deviation of the spine and providing accurate reference for the degree of offset in clinical practice. In this invention, the method of calculating the average prior probability by overlapping the spinal mask with the two-dimensional probability map averages the prior probability of the entire spinal region, which can obtain continuous normal position probability and offset probability. Compared with the single geometric distance or qualitative label in existing technologies, this invention can more comprehensively reflect the overall positional deviation of the spine and provide accurate reference for the degree of offset in clinical practice.

[0017] 2. This invention has good adaptability to image resolution and field of view. By normalizing the lateral position and performing prior interpolation, this method can be adapted to chest X-ray images of different sizes and field of view without having to set a threshold for each size.

[0018] 3. When outputting the results, this invention displays the average prior probability and offset probability as a percentage superimposed below the chest X-ray image. It also marks masked areas with low prior probabilities in dark colors and provides textual prompts to help technicians quickly determine whether the patient's position is acceptable or requires a retake. It can be directly embedded into the PACS or DR console interface.

[0019] 4. When constructing a one-dimensional prior probability distribution, this invention constructs a prior for the transverse distribution of the spine by statistically analyzing a large number of chest X-rays in normal postures. It encodes the normal geometric position into an explicit probability map, replacing the implicit learning of fuzzy geometric rules or black box models in the prior art. This allows the judgment logic of posture quality control to have a clear statistical basis, and the results are more stable and traceable.

[0020] 5. When constructing a one-dimensional prior probability distribution, this invention adds an extra point with a probability value of 0 at the rightmost position, which ensures that the endpoints are defined during interpolation.

[0021] 6. To address the problem that existing technologies require setting thresholds separately for chest X-rays with different equipment and field of view sizes, resulting in poor versatility and difficulty in large-scale application, this invention achieves efficient compatibility by normalizing the horizontal position, thus breaking free from specific resolution limitations. Based on linear interpolation and vertical replication of a one-dimensional prior distribution, it dynamically constructs a two-dimensional probability map that perfectly matches the size of the image to be tested. This eliminates the need to adjust core parameters for different sizes, enabling efficient compatibility with multiple scenarios and the ability to construct a prior once. Attached Figure Description

[0022] Figure 1 is a flowchart of the method of the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0024] Example 1 This example discloses a method for detecting spinal deviation on chest X-rays based on prior probability maps. The method is shown in Figure 1 and specifically includes steps S1-S6. Each step is described in detail below: Step S1: Obtain the chest X-ray image to be detected and normalize the horizontal pixels of the image.

[0025] Chest X-ray image to be tested Its width is Height is For chest X-rays in DICOM or PNG / JPEG format from PACS or file system.

[0026] The specific expression for normalizing the lateral position is: .

[0027] Step S2: Based on the normalized position of each column of horizontal pixels in the image, obtain the corresponding prior probability value through linear interpolation in the pre-constructed one-dimensional prior probability distribution, construct a one-dimensional array, and copy the one-dimensional array in the vertical direction to construct a two-dimensional prior probability map.

[0028] The two-dimensional prior probability graph is stored using a floating-point matrix, with values ​​ranging from 1 to 2. .

[0029] The preconstruction process of one-dimensional prior probability is as follows, including: collecting a set of chest X-ray images whose posture has been manually confirmed to be normal. Each image may have a different resolution, let's call it the first one. The width of the image is Height is .

[0030] The spinal region in each chest X-ray image is labeled using a segmentation network accelerated by manual methods or GPU, and the corresponding binary mask is obtained. ; where, when pixel When located in the spinal region, ;otherwise, .

[0031] Iterate through the set of chest X-ray images and their binary masks to obtain the leftmost and rightmost pixel coordinates of the spinal region in the width direction, and normalize them to intervals. Obtain the leftmost coordinate of the normalized coordinate. with optimal coordinates Specifically, for each chest X-ray image Scan its spinal mask This yields the leftmost and rightmost pixel coordinates of the spine region in the width direction: Normalize it to an interval : Since the selected training set consists of samples with normal body positions, therefore the above This describes the actual coverage area of ​​a normal spine in the image width direction.

[0032] Based on preset accuracy parameters Set step size (For example , ).

[0033] To set the step size Discretize within the interval to obtain a series of positions : .

[0034] Initial probability Iterate through the pixels of each chest X-ray image, based on all pixels in the image that satisfy... of The probability of each pixel at a given location is calculated by accumulating the probabilities. Specifically, for each chest X-ray image For all that satisfy Location ,make: During initialization, After the loop ends, a one-dimensional prior probability distribution is obtained. .

[0035] From an implementation perspective, a key-value pair structure can be used to store... , where the key is a string format with a certain number of decimal places retained.

[0036] To ensure that the endpoints are defined during interpolation, after obtaining the one-dimensional prior probability distribution, a point with a probability of 0 is added at the rightmost position, such as: in, The maximum r for all images in the training set .

[0037] The probability corresponding to each position according to Sort the values ​​by size to obtain an ordered sequence. As a one-dimensional prior probability distribution, it prepares for subsequent interpolation.

[0038] When obtaining the corresponding prior probability value through linear interpolation in a pre-constructed one-dimensional prior probability distribution, if the normalized position of a horizontal pixel in the image falls between two probability values ​​in the one-dimensional prior probability, then its prior probability value is obtained through a one-dimensional calculation formula, which is: in, This represents the normalized position of the horizontal pixels in the image. Position in one-dimensional prior probability The corresponding probability value, Position in one-dimensional prior probability The corresponding probability value, For position The corresponding prior probability value.

[0039] The final one-dimensional array is ,in, .

[0040] The expression for constructing a two-dimensional prior probability graph by copying the one-dimensional array vertically is as follows: .

[0041] Step S3: Input the chest X-ray image to be detected into the pre-trained GPU-accelerated segmentation network, and obtain the spinal binary mask from the probability map output by the GPU-accelerated segmentation network through thresholding or argmax operation. .

[0042] Before the X-ray chest image to be detected is input into the pre-trained GPU-accelerated segmentation network, preprocessing operations such as window width and window level adjustment, normalization, and scale transformation are performed.

[0043] The pre-trained GPU-accelerated segmentation network is a segmentation network implemented based on PyTorch.

[0044] Represents pixels It belongs to the spinal region; Represents pixels This is the background area.

[0045] Step S4: Based on the two-dimensional prior probability map, calculate the average prior probability and offset probability within the mask region of the spine binary mask.

[0046] In this invention, the average prior probability within the spinal region is used as the probability that the spine is in a normal position, defined as follows: If the spine as a whole lies within the normal area, then the area covered by the mask... The value is relatively large. The probability is also relatively high; conversely, if the spine is biased towards a low-probability area, then... Smaller.

[0047] Step S5: Compare the offset probability with the preset offset threshold. If the offset probability is greater than the offset threshold, it is determined that the spine is offset; otherwise, it is determined that the spine is basically centered.

[0048] To facilitate determination and display, this invention introduces an offset probability: when Exceeding the threshold At that time, it is considered that the spinal deviation is significant and the body position is incorrect.

[0049] In practical implementation, threshold A value of 0.5 can be used, or it can be determined through ROC curve analysis based on clinical needs.

[0050] Step S6: Output the comparison results. Display the average prior probability and the offset probability as percentages below the chest X-ray image. Mark the masked areas with low prior probabilities in dark colors and add text prompts to complete the detection of spinal offset on the chest X-ray.

[0051] To help technicians quickly understand the results, this invention displays the two probabilities as a percentage on the visualization screen.

[0052] Specifically, for example, it can be displayed below the image as: in 92.3%: indicating that there is a 92.3% probability that the spine is in the normal position area; out 7.7%: indicating that there is a 7.7% probability that the spine is deviated.

[0053] In addition, it outputs judgment prompts. If the judgment is that the spine is misaligned, it will display a message such as "Spine misalignment, please consider retaking the test"; if the body position is normal, it will display "Position is qualified".

[0054] Example 2 This example is based on the X-ray chest radiograph spinal deviation detection method based on prior probability maps disclosed in Example 1 above. It discloses an X-ray chest radiograph spinal deviation detection system based on prior probability maps. The system includes: a data preprocessing module: acquiring the X-ray chest radiograph image to be detected, normalizing and preprocessing the horizontal pixels of the image; a prior construction module: generating and saving a one-dimensional prior probability using normal posture samples, loading the one-dimensional prior probability when the system starts, and dynamically constructing a two-dimensional prior probability map for X-ray chest radiograph images of different widths; the specific process of constructing the two-dimensional prior probability map includes: based on the normalized position of each column of horizontal pixels in the image, obtaining the corresponding prior probability value through linear interpolation in the pre-constructed one-dimensional prior probability distribution, and constructing a one-dimensional array; and then... The system performs a vertical copy to construct a two-dimensional prior probability map. The segmentation module inputs the chest X-ray image to be detected into a pre-trained GPU-accelerated segmentation network. Through thresholding or argmax operations, it obtains a binary mask of the spine from the probability map output by the GPU-accelerated segmentation network. The offset probability calculation module calculates the average prior probability and offset probability within the masked region of the spine binary mask based on the two-dimensional prior probability map. The judgment and display module compares the offset probability with a preset offset threshold. If the offset probability is greater than the threshold, it is judged as spinal offset; otherwise, it is judged as the spine being basically centered. The average prior probability and offset probability are displayed as percentages below the chest X-ray image, and masked regions with low prior probabilities are marked in dark colors with accompanying text prompts, thus completing the detection of spinal offset on the chest X-ray.

[0055] For details regarding the above modules, please refer to the relevant descriptions and effects in Example 1 for further understanding.

[0056] Based on Example 1, Example 3 provides an electronic device, including one or more processors and a memory, wherein the memory stores one or more programs, and the one or more programs include instructions for executing the X-ray chest radiograph spinal deviation detection method as described above based on prior probability maps.

[0057] At the hardware level, the electronic device includes a processor, internal bus, network interface, memory, and non-volatile memory, and may also include other hardware required for the business. The processor reads the corresponding computer program from the non-volatile memory into memory and then runs it to implement the above-mentioned method for detecting spinal deviation on X-ray chest radiographs based on prior probability maps. Of course, in addition to software implementation, this invention does not exclude other implementation methods, such as logic devices or a combination of hardware and software, etc. That is to say, the execution subject of the following processing flow is not limited to individual logic units, but can also be hardware or logic devices.

[0058] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0059] Computer-readable media include both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0060] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for detecting spinal deviation on chest X-rays based on prior probability maps, characterized in that, The method includes: acquiring a chest X-ray image to be detected and normalizing the horizontal pixels of the image; based on the normalized position of each column of horizontal pixels in the image, obtaining the corresponding prior probability value through linear interpolation in a pre-constructed one-dimensional prior probability distribution, and constructing a one-dimensional array; copying the one-dimensional array vertically to construct a two-dimensional prior probability map; inputting the chest X-ray image to be detected into a pre-trained GPU-accelerated segmentation network, and obtaining a binary mask of the spine from the probability map output by the GPU-accelerated segmentation network through thresholding or argmax operation; based on the two-dimensional prior probability map, calculating the average prior probability and offset probability within the mask region of the binary mask of the spine, and comparing them with a preset offset threshold; if the offset probability is greater than the offset threshold, it is determined to be a spinal offset, otherwise it is determined to be a basically centered spine; outputting the comparison result, displaying the average prior probability and offset probability as a percentage superimposed below the chest X-ray image, and marking the mask region with low prior probability in dark color with text prompts, thus completing the detection of spinal offset on the chest X-ray.

2. The method for detecting spinal deviation on chest X-ray based on prior probability maps according to claim 1, characterized in that, The two-dimensional prior probability graph is stored using a floating-point matrix, with values ​​ranging from [value range missing]. 。 3. The method for detecting spinal deviation on chest X-ray based on prior probability maps according to claim 1, characterized in that, The pre-construction process of the one-dimensional prior probability includes: acquiring a set of chest X-ray images whose posture has been manually confirmed to be normal; annotating the spinal region in each chest X-ray image using a segmentation network accelerated by a manual process or a GPU to obtain the corresponding binary mask; traversing the set of chest X-ray images and its binary mask to obtain the leftmost and rightmost pixel coordinates of the spinal region in the width direction, and normalizing them to an interval. Obtain the leftmost coordinate of the normalized coordinate. with optimal coordinates Based on preset accuracy parameters, set the step size, and use the set step size to... Discretize within the interval to obtain discrete positions Initialize probability Iterate through the pixels of each chest X-ray image, based on all pixels in the image that satisfy... of The probability of each pixel at a given location is calculated by accumulating the probabilities. ; the probability corresponding to each position according to The numerical sorting is used to obtain the ordered sequence as a one-dimensional prior probability distribution.

4. The method for detecting spinal deviation on chest X-ray based on prior probability maps according to claim 3, characterized in that, In the binary mask, the value is 1 when the pixel is located in the spine region, and 0 otherwise.

5. The method for detecting spinal deviation on chest X-ray based on prior probability maps according to claim 3, characterized in that, The chest X-ray images in the collection have different resolutions.

6. The method for detecting spinal deviation on chest X-ray based on prior probability maps according to claim 3, characterized in that, After obtaining the one-dimensional prior probability distribution, a point with a probability value of 0 is added at the rightmost position.

7. The method for detecting spinal deviation on chest X-ray based on prior probability maps according to claim 1, characterized in that, When obtaining the corresponding prior probability value through linear interpolation in a pre-constructed one-dimensional prior probability distribution, if the normalized position of a horizontal pixel in the image falls between two probability values ​​in the one-dimensional prior probability, then its prior probability value is obtained through a one-dimensional calculation formula, which is: in, This represents the normalized position of the horizontal pixels in the image. Position in one-dimensional prior probability The corresponding probability value, Position in one-dimensional prior probability The corresponding probability value, For position The corresponding prior probability value.

8. The method for detecting spinal deviation on chest X-ray based on prior probability maps according to claim 1, characterized in that, Before the X-ray chest image to be detected is input into the pre-trained GPU-accelerated segmentation network, preprocessing operations such as window width and window level adjustment, normalization, and scale transformation are performed.

9. The method for detecting spinal deviation on chest X-ray based on prior probability maps according to claim 1, characterized in that, The pre-trained GPU-accelerated segmentation network is a segmentation network implemented based on PyTorch.

10. A spinal deviation detection system based on X-ray chest radiographs using prior probability maps, characterized in that, The system includes: a data preprocessing module: acquiring the chest X-ray image to be detected, normalizing and preprocessing the horizontal pixels of the image; a priori construction module: generating and saving a one-dimensional prior probability using normal body position samples, loading the one-dimensional prior probability when the system starts, and dynamically constructing a two-dimensional prior probability map for chest X-ray images of different widths to be detected; the specific process of constructing the two-dimensional prior probability map includes: based on the normalized position of each column of horizontal pixels in the image, obtaining the corresponding prior probability value through linear interpolation in the pre-constructed one-dimensional prior probability distribution, and constructing a one-dimensional array; copying the one-dimensional array in the vertical direction to construct the two-dimensional prior probability map; and a segmentation module: inputting the chest X-ray image to be detected into... A pre-trained GPU-accelerated segmentation network obtains a binary mask of the spine from the probability map output by the GPU-accelerated segmentation network through thresholding or argmax operations. An offset probability calculation module calculates the average prior probability and offset probability within the masked region of the spine binary mask based on the two-dimensional prior probability map. A judgment and display module compares the offset probability with a preset offset threshold; if the offset probability is greater than the threshold, it is determined to be a spinal offset; otherwise, it is determined to be a basically centered spine. The average prior probability and offset probability are superimposed as percentages and displayed below the chest X-ray image, with masked regions of low prior probability marked in dark and accompanied by textual prompts, thus completing the detection of spinal offset on the chest X-ray.

Images