Method, system, device, medium and program product for evaluating quality of a spine image

Abstract

The present disclosure provides a quality evaluation method, system, device, medium and program product of a spine image, the quality evaluation method comprising: acquiring a spine image to be detected; performing segmentation processing on the spine image to be detected to obtain a plurality of vertebra images to be detected; determining whether the vertebra image to be detected is a lateral vertebra image using a preset generative large model; acquiring a target number corresponding to the lateral vertebra image in the plurality of vertebra images to be detected corresponding to the spine image to be detected; and evaluating the quality of the spine image to be detected based on the target number. The present disclosure performs segmentation processing on the spine image to be detected to obtain the vertebra image to be detected, determines whether the vertebra image to be detected is a lateral vertebra image through a generative large model, and then evaluates the quality of the spine image to be detected, thereby improving the efficiency of classifying the spine image to be detected, reducing the cost, and improving the accuracy and reliability of the spine image quality evaluation.

Description

Technical Field

[0001] This disclosure relates to the field of spinal image processing technology, and in particular to a method, system, device, medium, and program product for quality evaluation of spinal images. Background Technology

[0002] X-ray images are common data in medical practice, used by doctors to diagnose patients, and also archived and input into downstream systems for further analysis after X-ray imaging. The standardization (normal imaging operation), completeness (images of the entire affected area), and accuracy (images of the areas of diagnostic interest) of the image data are crucial for doctors' diagnoses and for further processing and analysis by downstream information systems.

[0003] In the diagnosis of spinal diseases, lateral spinal X-rays are commonly used. Identifying qualified lateral views is crucial because only qualified views provide accurate and comprehensive spinal imaging information, enabling doctors to make correct diagnostic and treatment decisions. Unqualified views may contain blurry images, missing areas, or incorrect angles, potentially leading to misdiagnosis or missed diagnosis. Furthermore, qualified lateral views are also a prerequisite for downstream information systems to process and analyze data; only by ensuring image quality can the reliability and accuracy of subsequent data analysis be guaranteed.

[0004] Currently, while pre-trained machine learning models can achieve some classification results for X-ray images, their direct application to medical image classification and recognition suffers from high error rates due to a lack of in-depth analysis and understanding of image content and a shortage of relevant medical training data. This fails to meet clinical needs. Furthermore, medical-related machine learning tasks often require annotation by professional medical personnel, which is time-consuming and labor-intensive, especially in the medical field where professional annotation personnel are very expensive. Summary of the Invention

[0005] The technical problem to be solved by this disclosure is to overcome the shortcomings of existing technologies that use machine learning models to classify spinal X-rays, such as low classification efficiency, low accuracy, and high cost, and to provide a method, system, device, medium, and program product for quality evaluation of spinal images.

[0006] This disclosure solves the above-mentioned technical problems through the following technical solution:

[0007] This disclosure provides a method for quality assessment of spinal images, the quality assessment method comprising:

[0008] Acquire the image of the spine to be detected;

[0009] The image of the spine to be detected is segmented to obtain several images of the vertebrae to be detected;

[0010] A pre-defined generative large model is used to determine whether the vertebral image to be detected is a lateral vertebral image;

[0011] Obtain the number of targets corresponding to the lateral vertebral images among several vertebral images to be detected corresponding to the spinal image to be detected;

[0012] The quality of the spine image to be detected is evaluated based on the target number.

[0013] Preferably, the preset generative large model includes a multimodal generative large model and a text generative large model;

[0014] The step of determining whether the vertebral image to be detected is a lateral vertebral image using a preset generative large model includes:

[0015] The image of the vertebra to be detected is input into the multimodal generative large model to obtain the first text description content;

[0016] The first text description is input into the text-generating large model to determine whether the vertebral image to be detected is the lateral vertebral image.

[0017] Preferably, the step of inputting the first text description into the text-generating large model to determine whether the vertebral image to be detected is the lateral vertebral image includes:

[0018] The first text description is input into the text-generating large model to determine whether the vertebral image to be detected is a reference vertebral image.

[0019] In response to the fact that the vertebral body image to be detected is the reference vertebral body image, the feature information of the reference vertebral body image is used as the prompt words of the text-generating large model, so that the text-generating large model can determine whether the vertebral body image to be detected is the lateral vertebral body image.

[0020] Preferably, the step of segmenting the spinal image to be detected to obtain a plurality of vertebral body images to be detected includes:

[0021] The image of the spine to be detected is input into a first preset target detection model to obtain a spinal region image;

[0022] The spinal region image is segmented to obtain several images of the vertebrae to be detected.

[0023] Preferably, the step of segmenting the spinal region image to obtain a plurality of images of the vertebrae to be detected includes:

[0024] The image of the spinal region is segmented to obtain several segmented images to be detected;

[0025] Obtain the first similarity between the first contour information of the segmented image to be detected and the second contour information of the lateral vertebral image;

[0026] Based on the first similarity, it is determined whether the segmented image to be detected is the image of the vertebra to be detected.

[0027] Preferably, the step of segmenting the spinal region image to obtain a plurality of segmented images to be detected includes:

[0028] The spinal region image is segmented using a preset image segmentation model to obtain several segmented images to be detected.

[0029] And / or,

[0030] The step of determining whether the segmented image to be detected is the vertebral body image to be detected based on the first similarity includes:

[0031] In response to the first similarity being less than a first preset similarity, the segmented image to be detected is determined to be the vertebral image to be detected;

[0032] In response to the first similarity being not less than the first preset similarity, it is determined that the segmented image to be detected is not the image of the vertebra to be detected.

[0033] Preferably, the step of evaluating the quality of the spine image to be detected based on the number of targets includes:

[0034] In response to the target quantity being greater than a preset quantity, the quality of the spine image to be detected is determined to be qualified;

[0035] In response to the fact that the target number is not greater than the preset number, the quality of the spine image to be detected is determined to be unqualified.

[0036] Preferably, before the step of segmenting the spinal image to be detected to obtain several vertebral body images to be detected, the method further includes:

[0037] The image of the spine to be detected is input into a second preset target detection model to determine whether there is a foreign object in the image of the spine to be detected.

[0038] In response to the absence of the foreign object in the spinal image to be detected, the step of segmenting the spinal image to be detected to obtain several vertebral images to be detected is performed.

[0039] In response to the presence of the foreign object in the spinal image to be detected, a foreign object region image is obtained based on the second preset target detection model; the foreign object region image is segmented to obtain several images of foreign objects to be detected; a second similarity is obtained between the third contour information of the image of foreign objects to be detected and the fourth contour information of the lateral foreign object image; in response to the second similarity being not less than the second preset similarity, the step of segmenting the spinal image to be detected to obtain several images of vertebrae to be detected is performed.

[0040] Preferably, the quality evaluation method further includes:

[0041] In response to the second similarity being less than the second preset similarity, the image of the foreign object to be detected is input into a multimodal generative large model to obtain the second text description content;

[0042] The second text description is input into a text-generating large model to determine whether the image of the foreign object to be detected is a reference image of the foreign object.

[0043] In response to the fact that the image of the foreign object to be detected is not the reference foreign object image, the step of segmenting the image of the spine to be detected to obtain several images of the vertebrae to be detected is performed.

[0044] In response to the fact that the image of the foreign object to be detected is the reference image of the foreign object, the quality of the image of the spine to be detected is evaluated using the multimodal generative large model.

[0045] Preferably, the step of evaluating the quality of the spine image to be detected using the multimodal generative large model includes:

[0046] The image of the foreign object to be detected is input into the multimodal generative large model to determine whether the image of the foreign object to be detected is a lateral foreign object image;

[0047] In response to the fact that the image of the foreign object to be detected is the lateral foreign object image, the quality of the image of the spine to be detected is determined to be qualified.

[0048] This disclosure also provides a quality assessment system for spine images, the quality assessment system comprising:

[0049] The spine image acquisition module is used to acquire images of the spine to be detected.

[0050] The vertebral body image acquisition module is used to segment the image of the spine to be detected to obtain several vertebral body images to be detected.

[0051] The lateral vertebral body determination module is used to determine whether the image of the vertebral body to be detected is a lateral vertebral body image using a preset generative large model;

[0052] The target quantity acquisition module is used to acquire the target quantity corresponding to the lateral vertebral image among several vertebral images to be detected corresponding to the spinal image to be detected;

[0053] The quality evaluation module is used to evaluate the quality of the spine images to be detected based on the target quantity.

[0054] Preferably, the preset generative large model includes a multimodal generative large model and a text generative large model;

[0055] The lateral vertebral body determination module includes:

[0056] The first content acquisition unit is used to input the image of the vertebra to be detected into the multimodal generative large model to obtain the first text description content;

[0057] The lateral vertebral body determination unit is used to input the first text description content into the text generation model to determine whether the vertebral body image to be detected is the lateral vertebral body image.

[0058] Preferably, the lateral vertebral body determination unit includes:

[0059] The reference vertebral body determination subunit is used to input the first text description content into the text generation large model to determine whether the vertebral body image to be detected is a reference vertebral body image;

[0060] The lateral vertebral body determination subunit is used to respond to the fact that the vertebral body image to be detected is the reference vertebral body image, and to use the feature information of the reference vertebral body image as a prompt word of the text generation large model, so that the text generation large model can determine whether the vertebral body image to be detected is the lateral vertebral body image.

[0061] Preferably, the vertebral body image acquisition module includes:

[0062] A region image acquisition unit is used to input the spine image to be detected into a first preset target detection model to obtain a spine region image;

[0063] The vertebral body image acquisition unit is used to segment the image of the vertebral region to obtain several images of the vertebral bodies to be detected.

[0064] Preferably, the vertebral body image acquisition unit includes:

[0065] The image segmentation acquisition subunit is used to segment the spinal region image to obtain several segmented images to be detected.

[0066] The first similarity acquisition subunit is used to acquire the first similarity between the first contour information of the segmented image to be detected and the second contour information of the lateral vertebral image.

[0067] The vertebral image determination subunit is used to determine whether the segmented image to be detected is the vertebral image to be detected based on the first similarity.

[0068] Preferably, the segmented image acquisition subunit is further configured to segment the spinal region image using a preset image segmentation model to obtain a plurality of the segmented images to be detected;

[0069] And / or,

[0070] The vertebral image determination subunit is further configured to determine the segmented image to be detected as the vertebral image to be detected in response to the first similarity being less than the first preset similarity; and to determine the segmented image to be detected as not being the vertebral image to be detected in response to the first similarity being not less than the first preset similarity.

[0071] Preferably, the quality evaluation module includes:

[0072] The first response unit is configured to determine that the quality of the spine image to be detected is qualified in response to the target quantity being greater than a preset quantity.

[0073] The second response unit is used to determine that the quality of the spine image to be detected is unqualified in response to the target number not being greater than the preset number.

[0074] Preferably, the quality evaluation system further includes:

[0075] The foreign object detection module is used to input the spinal image to be detected into a second preset target detection model to determine whether there is a foreign object in the spinal image to be detected.

[0076] The first response module is used to respond to the absence of the foreign object in the spinal image to be detected by calling the vertebral image acquisition module to perform the step of segmenting the spinal image to be detected to obtain several vertebral images to be detected.

[0077] The second response module is used to respond to the presence of the foreign object in the spinal image to be detected, obtain a foreign object region image based on the second preset target detection model; segment the foreign object region image to obtain several images of foreign objects to be detected; obtain a second similarity between the third contour information of the image of foreign objects to be detected and the fourth contour information of the lateral foreign object image; and, in response to the second similarity being not less than the second preset similarity, call the vertebral image acquisition module to perform the step of segmenting the spinal image to be detected to obtain several images of vertebral bodies to be detected.

[0078] Preferably, the quality evaluation system further includes:

[0079] The second content acquisition module is used to input the image of the foreign object to be detected into a multimodal generative large model in response to the second similarity being less than the second preset similarity, so as to obtain the second text description content.

[0080] The reference foreign object determination module is used to input the second text description content into the text generation large model to determine whether the image of the foreign object to be detected is a reference foreign object image;

[0081] The third response module is used to call the vertebral image acquisition module to perform the step of segmenting the vertebral image to obtain several vertebral images in response to the fact that the image of the foreign object to be detected is not the reference foreign object image.

[0082] The fourth response module is used to evaluate the quality of the detected spinal image by employing the multimodal generative large model in response to the fact that the image to be detected is the reference image to be detected.

[0083] Preferably, the fourth response module includes:

[0084] Lateral foreign object determination unit, used to input the image of the foreign object to be detected into the multimodal generative large model, so as to determine whether the image of the foreign object to be detected is a lateral foreign object image;

[0085] An image quality determination unit is used to determine that the quality of the image of the spine to be detected is qualified in response to the fact that the image of the foreign object to be detected is the lateral foreign object image.

[0086] This disclosure also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and for running on the processor, wherein the processor executes the computer program to implement the above-described method for evaluating the quality of spine images.

[0087] This disclosure also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method for evaluating the quality of spine images.

[0088] This disclosure also provides a computer program product, including a computer program that, when executed by a processor, implements the spine image quality assessment method as described above.

[0089] Based on common knowledge in the field, the above-mentioned preferred conditions can be combined arbitrarily to obtain various preferred embodiments of this disclosure.

[0090] The positive and progressive effects of this disclosure are as follows:

[0091] This disclosure involves segmenting a spinal image to obtain a vertebral body image, and using a generative large model to determine whether the vertebral body image is a lateral vertebral body image, thereby achieving quality evaluation of the spinal image to be tested. This improves the efficiency of classifying spinal images to be tested, reduces costs, and enhances the accuracy and reliability of spinal image quality evaluation. Attached Figure Description

[0092] Figure 1 This is a flowchart of the method for evaluating the quality of spinal images according to Embodiment 1 of this disclosure;

[0093] Figure 2 This is a flowchart of step S13 in the method for evaluating the quality of spine images according to Embodiment 2 of this disclosure;

[0094] Figure 3 This is a first example diagram of using a multimodal generative large model to describe the image of the vertebral body to be detected in Embodiment 2 of this disclosure;

[0095] Figure 4 This is a second example diagram illustrating the use of a multimodal generative large model to describe the image of the vertebral body to be detected in Embodiment 2 of this disclosure;

[0096] Figure 5 This is a flowchart of step S132 in the method for evaluating the quality of spine images according to Embodiment 2 of this disclosure;

[0097] Figure 6 This is a flowchart of step S12 in the method for evaluating the quality of spine images according to Embodiment 2 of this disclosure;

[0098] Figure 7 This is an example image of the spinal region obtained after target detection on a lateral spinal X-ray in Embodiment 2 of this disclosure.

[0099] Figure 8 This is an example image of the spinal region obtained after target detection on a non-lateral spinal X-ray in Embodiment 2 of this disclosure.

[0100] Figure 9 This is a flowchart of step S122 in the method for evaluating the quality of spine images according to Embodiment 2 of this disclosure;

[0101] Figure 10 This is a first example image of the segmented image to be detected after undergoing black and white transformation processing in Embodiment 2 of this disclosure;

[0102] Figure 11 This is a second example image of the segmented image to be detected after undergoing black-and-white transformation processing in Embodiment 2 of this disclosure;

[0103] Figure 12This is a third example image of the segmented image to be detected after undergoing black-and-white transformation processing in Embodiment 2 of this disclosure;

[0104] Figure 13 This is a fourth example image of the segmented image to be detected after black and white transformation processing in Embodiment 2 of this disclosure;

[0105] Figure 14 This is an example diagram of the second contour information of a lateral vertebral body image in Embodiment 2 of this disclosure;

[0106] Figure 15 Example image of the segmented image to be detected obtained after semantic segmentation of a lateral spinal X-ray in Embodiment 2 of this disclosure;

[0107] Figure 16 This is an example image of the segmented image to be detected obtained after semantic segmentation of a non-lateral spinal X-ray in Embodiment 2 of this disclosure;

[0108] Figure 17 This is a flowchart of the method for evaluating the quality of spinal images according to Embodiment 2 of this disclosure;

[0109] Figure 18 This is an example image of the foreign body region obtained after target detection of a bone nail in a lateral spinal X-ray in Embodiment 2 of this disclosure;

[0110] Figure 19 This is an example image of the foreign body region obtained after target detection of a bone nail in a non-lateral spinal X-ray in Embodiment 2 of this disclosure;

[0111] Figure 20 This is an example image of the foreign body to be detected obtained by semantic segmentation of the bone nail portion in a lateral spinal X-ray in Embodiment 2 of this disclosure.

[0112] Figure 21 This is an example image of the foreign object to be detected obtained after semantic segmentation of a non-lateral spinal X-ray in Embodiment 2 of this disclosure;

[0113] Figure 22 This is a first example image of the foreign object to be detected after black and white transformation processing in Embodiment 2 of this disclosure.

[0114] Figure 23 This is a second example image of the foreign object to be detected after black and white transformation processing in Embodiment 2 of this disclosure;

[0115] Figure 24 This is a third example image of the foreign object to be detected after black and white transformation processing in Embodiment 2 of this disclosure;

[0116] Figure 25This is an example diagram illustrating the textual description of a lateral image of a foreign object to be detected using a multimodal generative large model in Embodiment 2 of this disclosure.

[0117] Figure 26 This is an example diagram illustrating the use of a multimodal generative large model to describe a non-lateral image of a foreign object to be detected in Embodiment 2 of this disclosure.

[0118] Figure 27 This is a flowchart of step S18 in the method for evaluating the quality of spine images according to Embodiment 2 of this disclosure;

[0119] Figure 28 This is a specific example diagram of the quality assessment method for spine images according to Embodiment 2 of this disclosure;

[0120] Figure 29 This is a schematic diagram of the modules of the quality assessment system for spinal images according to Embodiment 3 of this disclosure;

[0121] Figure 30 This is a schematic diagram of the modules of the quality assessment system for spine images according to Embodiment 4 of this disclosure;

[0122] Figure 31 This is a schematic diagram of the structure of the electronic device according to Embodiment 5 of this disclosure. Detailed Implementation

[0123] The present disclosure is further illustrated below by way of embodiments, but the present disclosure is not limited to the scope of the embodiments described herein.

[0124] The prefixes such as "first" and "second" used in this disclosure are merely for distinguishing different descriptive objects and do not limit the position, order, priority, quantity, or content of the described objects. The use of ordinal numbers and other prefixes used to distinguish descriptive objects in this disclosure does not constitute a limitation on the described objects. The description of the described objects is given in the claims or the context of the embodiments, and should not be construed as an unnecessary limitation. Furthermore, in the description of this embodiment, unless otherwise stated, "multiple" means two or more.

[0125] In this embodiment of the disclosure, the collection, storage, use, processing, transmission, provision, and disclosure of user personal information comply with relevant laws and regulations and do not violate public order and good morals.

[0126] Example 1

[0127] This embodiment provides a method for evaluating the quality of spinal images, such as... Figure 1 As shown, the quality assessment method includes:

[0128] S11. Obtain the image of the spine to be detected;

[0129] S12. Segment the image of the spine to be detected to obtain several images of the vertebrae to be detected;

[0130] S13. Use a pre-defined generative large model to determine whether the vertebral image to be detected is a lateral vertebral image;

[0131] S14. Obtain the number of targets corresponding to the lateral vertebral images in several vertebral images corresponding to the spinal image to be detected.

[0132] S15. Evaluate the quality of the spine images to be detected based on the number of targets.

[0133] Specifically, the spine consists of multiple vertebrae, or vertebral bodies, which are divided into thoracic vertebrae and lumbar vertebrae. These vertebrae are often used as an important part of diagnosis and analysis in spinal imaging.

[0134] The image of the spine to be examined can be a spinal X-ray, also known as a spinal X-ray image. An X-ray image is formed by X-rays penetrating human tissue and forming an image on a photosensitive material or digital sensor. X-ray images are commonly used to examine internal human structures, such as bones and lungs.

[0135] Generative large models refer to generative models with a large number of parameters. They are usually based on deep learning and can learn the complex distribution of data to generate new data that is similar to or related to the original data. For example, generative large models can generate new images, describe images, or generate corresponding images based on text, based on learned image features.

[0136] In this embodiment, the spinal image to be detected is segmented to obtain the vertebral body image to be detected. A generative large model is used to determine whether the vertebral body image to be detected is a lateral vertebral body image, thereby realizing the quality evaluation of the spinal image to be detected. This improves the efficiency of classifying the spinal image to be detected, reduces the cost, and improves the accuracy and reliability of the spinal image quality evaluation.

[0137] Example 2

[0138] This embodiment provides a method for evaluating the quality of spinal images, which is a further improvement on Embodiment 1.

[0139] In a feasible scheme, the pre-defined generative large model includes a multimodal generative large model and a text-generative large model;

[0140] like Figure 2 As shown, step S13 includes:

[0141] S131. Input the image of the vertebra to be detected into the multimodal generative large model to obtain the first text description content;

[0142] S132. Input the first text description into the text generation model to determine whether the vertebral image to be detected is a lateral vertebral image.

[0143] Specifically, a multimodal generative large model is a generative large model capable of simultaneously processing and generating multiple modalities of data (such as images, text, and speech). For example, a multimodal generative large model can generate corresponding text descriptions based on input images, or generate corresponding images based on text descriptions, achieving conversion and association between different modalities.

[0144] By leveraging the language generation capabilities of a multimodal generative large model, we can describe the map tiles and assist in the judgment process.

[0145] Because general image segmentation methods have limited accuracy on medical images, they may produce errors or misidentifications. To further verify and identify the content of image patches, multimodal generative large-scale models are introduced. These models are trained on a large amount of general image and text data and a certain amount of medical image and text data, possessing certain medical knowledge, and can provide textual descriptions of images, thereby assisting in determining whether an image patch is a vertebral body.

[0146] Multimodal generative large models, such as GPT-4V (a multimodal model), GLM-4V (a multimodal visual large model), and LlaVA (a large multimodal model), are used to describe the images of the vertebrae to be detected. When using multimodal generative large models, the guiding words (cues) do not need to be preset with specific locations or content; instead, they are allowed to use their integrated knowledge to make judgments.

[0147] like Figure 3 The image shown is a first example of using a multimodal generative large model to generate a text description of an image of a vertebra to be detected. The input consists of the image of the vertebra to be detected and the prompt "describe the details of this vertebral X-ray," and the output is as follows:

[0148] "This X-ray shows a frontal (coronal) view of a vertebral body. Here are some details that can be observed:"

[0149] 1. Vertebral body shape: The vertebral body is rectangular with relatively clear edges, showing a relatively normal anatomical structure. The upper and lower edges are clearly defined, and the two sides and the central part of the vertebral body can be seen.

[0150] 2. Vertebral body bone mineral density: In the imaging, the bone mineral density of the vertebral body appears uniform, with no obvious signs of osteoporosis. The X-ray translucency of the entire vertebral body region is uniform, with no obvious areas of abnormal density.

[0151] 3. Vertebral endplates: The outlines of the superior and inferior endplates are visible, demonstrating the structural integrity of the vertebral body. There is no obvious compression deformation or collapse in the endplate area.

[0152] 4. Intervertebral disc space: Since this is an X-ray of a single vertebra, detailed information about the intervertebral disc is not shown. However, the location of the intervertebral disc space in the upper and lower regions of the image can be inferred.

[0153] 5. Trabecular bone structure: The trabecular bone structure inside the vertebral body shows a certain texture, which is a typical feature of healthy bones, indicating that the bone structure is relatively intact.

[0154] 6. Pathological features: This image shows no obvious signs of fracture, tumor, or other pathological changes.

[0155] like Figure 4 The image shown is a second example of using a multimodal generative large model to generate a text description of the vertebral body image to be detected. The input is the image of the vertebral body to be detected and the prompt "describe the details of this vertebral X-ray," and the output is as follows:

[0156] "This X-ray shows a lateral (sagittal) view of a vertebral body. The following details are observable:"

[0157] 1. Vertebral body shape: The vertebral body is approximately rectangular in shape. When viewed from the side, the upper and lower edges and the anterior and posterior contours of the vertebral body are clearly visible. The height and width of the vertebral body appear uniform, without any obvious signs of compression or collapse.

[0158] 2. Anterior and posterior margins of the vertebral body: The outlines of the anterior and posterior margins are clearly defined, indicating that the vertebral body structure is intact. There are no signs of osteophytes (bone hyperplasia) at the anterior margin, and no abnormal protrusions or collapses are observed at the posterior margin.

[0159] 3. Intervertebral disc space: Because it is a side view, although the edges of the two vertebrae can be observed, this X-ray does not show the complete intervertebral disc space. If the intervertebral disc space were shown, its thickness and structural condition could be seen.

[0160] 4. Bone density: The bone density of the vertebral bodies appears uniform, showing no obvious signs of osteoporosis. The trabecular structure inside the vertebral bodies is clearly visible from the side, exhibiting typical characteristics of healthy bone.

[0161] 5. Pathological features: No obvious signs of vertebral body lesions, such as fractures, tumors, or significant vertebral deformities. The anterior and posterior margins of the vertebral bodies are smooth, with no osteophytes or other abnormal bone structures observed.

[0162] In this scheme, a multimodal generative large model is introduced to describe the vertebral body image to be detected, so as to help determine whether the vertebral body image to be detected is a lateral vertebral body image, thereby improving the accuracy and reliability of the determination of the vertebral body image to be detected.

[0163] In a feasible solution, such as Figure 5 As shown, step S132 includes:

[0164] S1321. Input the first text description into the text-generating large model to determine whether the vertebral image to be detected is a reference vertebral image;

[0165] S1322. In response to the vertebral body image to be detected being a reference vertebral body image, the feature information of the reference vertebral body image is used as a prompt word for the text-generating large model, so that the text-generating large model can determine whether the vertebral body image to be detected is a lateral vertebral body image.

[0166] When using generative large models, in order to obtain the desired output, the model is usually provided with some cue words or guiding words to guide the model's generation process. These cue words can contain information such as task description, expected content or style, helping the model to more accurately understand the task requirements.

[0167] By leveraging the understanding and reasoning capabilities of a text-generating large-scale model, the accuracy of the image patch is further verified to determine if it represents a vertebra. This model possesses powerful language understanding and reasoning abilities, enabling logical analysis and judgment of textual descriptions. By inputting the initial textual description, the general text-generating large-scale model determines whether the description is related to a vertebra. If the model determines it is, the image patch is retained; otherwise, it is excluded. This further filters out images that are not vertebral bodies, improving the overall accuracy of the identification.

[0168] After identifying the vertebral body image to be detected as the reference vertebral body image, the introduction of medical background knowledge allows for a more accurate determination of whether the vertebral body image is a lateral or non-lateral view. First, the different characteristic information of the vertebral bodies in lateral and anteroposterior views is compiled, such as vertebral body width, symmetry, vertebral body height, anterior and posterior margin morphology, pedicles, transverse and spinous processes, spinal canal outline, intervertebral disc space, intervertebral foramen, physiological curvature of the spine, laminae and facet processes, atlas and axis, etc. Then, this characteristic information is used as prompts to guide the text-generating model to judge the first text description. The model infers whether the image is a lateral view based on the description and prompt features. This method combines the model's language understanding ability with medical knowledge, improving the accuracy of the judgment. If the vertebral body image to be detected is not the reference vertebral body image, then the vertebral image to be detected is not a lateral vertebral image, and the quality evaluation of the vertebral image to be detected is unqualified.

[0169] In determining whether a vertebral image to be detected is a lateral vertebral image using a pre-defined generative large model, a step-by-step guided approach was adopted. When handling complex tasks, generative large models may exhibit hallucination phenomena—generating inaccurate or irrelevant content—if presented with too much information at once. By breaking the task down into multiple steps, each focusing on a specific sub-task, the model can more accurately understand and process information at each stage. This makes the model's judgment process more transparent, facilitates the introduction of human medical knowledge for guidance, and improves overall judgment accuracy.

[0170] In this approach, the feature information of the reference vertebral body image is used as prompt words in the text-generating large model. This combines the model's language understanding ability with medical knowledge, thereby improving the accuracy of judging the vertebral body image to be detected.

[0171] In a feasible solution, such as Figure 6 As shown, step S12 includes:

[0172] S121. Input the image of the spine to be detected into the first preset target detection model to obtain the image of the spine region;

[0173] S122. The spinal region image is segmented to obtain several images of the vertebrae to be detected.

[0174] Specifically, in image processing, the first preset target detection model obtains a spinal region image by performing target detection on the spinal image to be detected. Target detection refers to identifying and locating a specific object in an image. For example, in a spinal X-ray, the extent of the spine is detected, which can be a rectangular region.

[0175] The first preset target detection model is a general pre-trained target detection model, such as YOLO (a target detection algorithm), DINO (an unsupervised representation learning method), Grounding-DINO (an open detection model), etc., to perform target detection on spinal X-rays, locate the part of the spine in the X-ray, find the approximate position of the spine in the image, so as to narrow down the scope of interest and exclude other irrelevant image information.

[0176] like Figure 7 The image shown is an example of a spinal region image obtained after target detection on a lateral spinal X-ray.

[0177] like Figure 8 The image shown is an example of a spinal region image obtained after target detection on a non-lateral spinal X-ray.

[0178] In this scheme, a target detection model is used to detect targets in the spinal image to obtain a spinal region image, which narrows the scope of interest in the image. Then, segmentation processing is performed to ensure the accuracy and effectiveness of the segmentation process.

[0179] In a feasible solution, such as Figure 9 As shown, step S122 includes:

[0180] S1221. The spinal region image is segmented to obtain several segmented images to be detected;

[0181] S1222. Obtain the first similarity between the first contour information of the segmented image to be detected and the second contour information of the lateral vertebral body image;

[0182] S1223. Based on the first similarity, determine whether the segmented image to be detected is the image of the vertebra to be detected.

[0183] Specifically, shape matching is used to filter out the true cone region and eliminate interference. The individual image patches corresponding to the segmented image to be detected are extracted. Image processing is used to find their contours and perform a black-and-white transformation. For example... Figure 10 The image shown is a first example of the segmented image to be detected after undergoing black-and-white transformation. Figure 11 The image shown is a second example of the segmented image to be detected after undergoing black-and-white transformation. Figure 12 The image shown is a third example of the segmented image to be detected after undergoing black-and-white transformation. Figure 13 The image shown is the fourth example of the segmented image to be detected after undergoing black-and-white transformation. Figure 14 The image shown is an example of the second contour information of a lateral vertebral body image. A pre-prepared standard lateral vertebral body image is obtained. A first similarity is calculated between the segmented image to be detected and the lateral vertebral body image, such as a geometric value like Hausdorff distance (a way to measure the distance between two sets) or Hu moments (a feature descriptor based on image moments). Based on the first similarity, it is determined whether the segmented image to be detected is indeed the vertebral body image to be detected.

[0184] In this scheme, the first similarity is used to determine whether the segmented image to be detected is the vertebral body image to be detected. The segmented image to be detected is then filtered to obtain the vertebral body image to be detected, thereby improving the accuracy and reliability of the vertebral body image to be detected.

[0185] In one feasible embodiment, step S1221 includes:

[0186] A preset image segmentation model is used to segment the spinal region image to obtain several segmented images to be detected.

[0187] Specifically, the spine region detected in the previous step is cropped out, and a general pre-trained image segmentation model, such as SAM (Spatial Attention Map), Medical-SAM (Medical Spatial Attention Map), or PaddleSeg (an end-to-end image segmentation suite), is used to segment patches of different regions in the image. The pre-trained image segmentation model performs semantic segmentation on the spine region image. Semantic segmentation is a computer vision technique that aims to classify each pixel in an image so that pixels of the same category are spatially connected into regions. For example, in a spinal X-ray, semantic segmentation can segment each vertebral body region.

[0188] This step extracts the location and shape of the vertebral bodies from the image, segments the vertebral body region, and obtains more detailed information, but may also introduce noise, segmenting out some non-vertebral or meaningless patches.

[0189] like Figure 15 The image shown is an example of the segmented image to be detected obtained after semantic segmentation of a lateral spinal X-ray.

[0190] like Figure 16 The image shown is an example of the segmented image to be detected obtained after semantic segmentation of a non-lateral spinal X-ray.

[0191] Both the initial object detection model and the initial image segmentation model are general-purpose pre-trained models. General-purpose pre-trained models refer to models pre-trained on a large amount of data. These models typically have broad applicability and can be used in various tasks. Pre-trained models are trained on a specific task or domain and learn general features and patterns. The core idea behind these models is to leverage the training results from large-scale datasets to reduce the training requirements for specific tasks, thereby improving efficiency and performance.

[0192] In this scheme, the image segmentation model is used to segment the spinal region image to obtain several segmented images to be detected, thereby improving the accuracy and reliability of the segmented images to be detected.

[0193] In one feasible embodiment, step S1223 includes:

[0194] In response to a first similarity score being less than a first preset similarity score, the image to be detected and segmented is determined to be a vertebral body image to be detected;

[0195] In response to a first similarity not being less than a first preset similarity, it is determined that the segmented image to be detected is not the image of the vertebra to be detected.

[0196] If the initial similarity is less than a certain threshold, i.e., a first preset similarity, the two images can be considered to have similar contours. The specific threshold can be determined experimentally in practice; for example, the threshold for Hausdorff distance is generally between 20 and 50.

[0197] In this scheme, the relationship between the first similarity and the first preset similarity is used to determine whether the segmented image to be detected is the image of the vertebra to be detected, thereby improving the accuracy and reliability of the image of the vertebra to be detected.

[0198] In one feasible embodiment, step S15 includes:

[0199] If the target number is greater than the preset number, the quality of the spine image to be detected is determined to be qualified.

[0200] If the target number is not greater than the preset number, the quality of the spinal image to be detected is determined to be unqualified.

[0201] Specifically, each possible vertebral body location obtained by segmenting X-ray images (i.e., the vertebral body image to be detected) is determined to be a lateral vertebral image. Then, the number of lateral vertebrae (or lateral vertebral bodies) in each X-ray image is counted. If the number of vertebral body images judged to be lateral in an X-ray image exceeds a certain number (a preset number), it is considered a lateral view and the X-ray image quality is acceptable; otherwise, it is not a lateral view and the X-ray image quality is unacceptable. A typical spinal X-ray includes 5 to 10 vertebral bodies, and in practice, the number of lateral vertebral bodies detected is usually between 2 and 6. The specific threshold can be determined experimentally on the dataset of X-ray images to be classified.

[0202] In this scheme, the quality of the spinal image to be tested is determined by the relationship between the number of target images and the preset number of images, thereby improving the accuracy and reliability of the quality evaluation of the spinal image to be tested.

[0203] In a feasible solution, such as Figure 17 As shown, before step S12, the following steps are also included:

[0204] S123. Input the image of the spine to be detected into the second preset target detection model to determine whether there is a foreign object in the image of the spine to be detected;

[0205] In response to the absence of foreign objects in the spinal image to be detected, step S12 is executed;

[0206] In response to the presence of a foreign object in the spinal image to be detected, step S124 is executed;

[0207] S124. Based on the second preset target detection model, obtain the foreign object region image; perform segmentation processing on the foreign object region image to obtain several foreign object images to be detected; obtain the second similarity between the third contour information of the foreign object image to be detected and the fourth contour information of the lateral foreign object image; in response to the second similarity being not less than the second preset similarity, execute step S12.

[0208] Specifically, foreign bodies, i.e., implants, such as bone screws, can be used to help determine the direction of an X-ray because their geometry appears different on lateral and non-lateral X-rays. Besides bone screws, other types of implants can also be used, as long as their geometry is clearly different on anteroposterior and lateral X-rays, thus aiding in identification.

[0209] The implant's location is detected in the X-ray using a pre-trained general object detection model, i.e., the second preset object detection model. Since implants have relatively defined geometric features, the general object detection model can also locate their positions. This step, using the implant's position and shape, provides important information for determining the orientation of the X-ray.

[0210] like Figure 18 The image shown is an example of a foreign body region image obtained after target detection of a bone screw in a lateral spinal X-ray.

[0211] like Figure 19 The image shown is an example of a foreign body region image obtained after target detection of a bone screw in a non-lateral spinal X-ray.

[0212] Performing semantic segmentation on the detected implant region, i.e., the foreign body region image, is helpful for subsequent shape matching. The method of segmenting the foreign body region image to obtain several images of the foreign body to be detected is similar to the aforementioned method of segmenting the spinal region image to obtain several segmented images of the foreign body to be detected, both of which use semantic segmentation, and will not be described in detail here.

[0213] like Figure 20 The image shown is an example of a foreign body image to be detected obtained by semantic segmentation of the bone nail portion in a lateral spinal X-ray.

[0214] like Figure 21 The image shown is an example of a foreign body image obtained after semantic segmentation of a non-lateral spinal X-ray.

[0215] The segmented implant shape, i.e., the image of the foreign body to be detected, is contour extracted and matched with a standard implant shape template, i.e., a lateral view of the foreign body. The specific type and orientation of the implant are determined by calculating a second similarity. Since the shape of the implant differs significantly between lateral and anteroposterior X-rays, shape matching can help determine the X-ray imaging direction. The calculation and judgment process for the second similarity is similar to that for the first similarity and will not be repeated here.

[0216] like Figure 22 The image shown is a first example of the foreign object to be detected after undergoing black and white transformation processing.

[0217] like Figure 23 The image shown is a second example of the foreign object to be detected after undergoing black and white transformation processing.

[0218] like Figure 24 The image shown is a third example of the foreign object to be detected after undergoing black-and-white transformation.

[0219] In this scheme, the presence of foreign objects in the spinal image to be detected is first determined. Then, the second similarity between the image of the foreign object to be detected and the lateral image of the foreign object is calculated. Finally, it is determined whether to segment the spinal image to obtain several images of the vertebrae to be detected. In other words, the presence of foreign objects in the spinal image is determined first, and then the image is segmented to obtain the images of the vertebrae to be detected. The shape of foreign objects is relatively obvious in lateral and non-lateral images; determining the foreign object first improves the efficiency of spinal image detection.

[0220] In a feasible solution, such as Figure 17 As shown, this quality assessment method also includes:

[0221] S16. In response to the second similarity being less than the second preset similarity, the image of the foreign object to be detected is input into the multimodal generative large model to obtain the second text description content.

[0222] S17. Input the second text description into the text generation model to determine whether the foreign object image to be detected is a reference foreign object image;

[0223] In response to the fact that the image of the foreign object to be detected is not the reference image, step S12 is executed;

[0224] In response to the fact that the image of the foreign object to be detected is a reference image, step S18 is executed;

[0225] S18. A multimodal generative large model is used to evaluate the quality of the spinal images to be detected.

[0226] Specifically, a multimodal generative large model is used to describe the extracted images of the foreign objects to be detected. Because these models possess certain general medical knowledge, they can generate textual descriptions of the type and location of the foreign objects based on their shape and characteristics, providing a basis for further judgment.

[0227] When utilizing a multimodal generative large model, target detection bounding boxes obtained through a second preset target detection model can be added to the original X-ray image (i.e., the image of the spine to be detected) to assist the large model in describing the target. Simultaneously, the large model's prompts and guidance do not require preset locations or content, allowing the large model to use its integrated knowledge to make judgments.

[0228] like Figure 25 The image shown is an example of using a multimodal generative large model to generate a text description of a lateral view image of a foreign object to be detected. The input consists of the image of the foreign object to be detected and the prompt "Describe in detail the area framed in this X-ray, other areas are not required," and the output is as follows:

[0229] "In this X-ray, the framed area is concentrated in the lower part of the spine, particularly the surgical implants in the lumbar region. The following is a detailed description of the framed area:"

[0230] 1. The framed steel nails:

[0231] The area is labeled "steel screw 0.65", indicating that it is a steel nail with a confidence level of 0.65.

[0232] Within the frame, two parallel pedicle screws can be seen. These screws are typical fixation instruments used in spinal surgery to stabilize the vertebral body.

[0233] The screw's threads are clearly visible, showing that it was inserted along the anterior-posterior direction of the vertebral body, providing a stable fixation. The screw is relatively long, with threads designed to embed tightly into the bone to ensure secure fixation.

[0234] 2. Connecting components: ……

[0235] like Figure 26 The image shown is an example of using a multimodal generative large model to generate a text description of a non-lateral view image of a foreign object to be detected. The input consists of the image of the foreign object to be detected and the prompt "Describe in detail the area framed in this X-ray, other areas are not needed," and the output is as follows:

[0236] "In this X-ray, two boxes mark the locations of the steel screws implanted during spinal surgery. The following is a detailed description of each boxed area:"

[0237] 1. The red box on the left (labeled "steel screw 0.60"):

[0238] • This box shows an implanted steel nail, with the screw partially embedded in the spinal bone.

[0239] The steel nails have a spiral thread structure, indicating that they are pedicle screws used to fix the vertebral body.

[0240] The screw head is relatively large and clearly visible, indicating the angle and position of implantation. The screw head may have a connecting rod for connection to other fixation devices.

[0241] The steel nail is located on the left side of the spine and at an angle to the midline of the spine. It is usually used to stabilize the vertebrae and prevent abnormal movement of the spine.

[0242] 2. The green box on the right (labeled "steel screw 0.56"):

[0243] • This box also shows an implanted steel nail, similar in shape and position to the steel nail on the left.

[0244] • The outline of the screw is clear, easily showing the structure of the thread and the head of the screw. Its shape and angle are similar to the steel nail on the left...

[0245] In this solution, the second text description is input into a text-generating large model to determine whether the image to be detected is a reference image. For example, it determines whether it is describing a bone nail instead of something else or an abnormal block, thus improving the overall accuracy of recognition.

[0246] In a feasible solution, such as Figure 27 As shown, step S18 includes:

[0247] S181. Input the image of the foreign object to be detected into the multimodal generative large model to determine whether the image of the foreign object to be detected is a lateral foreign object image;

[0248] S182. In response to the fact that the image of the foreign body to be detected is a lateral foreign body image, the quality of the image of the spine to be detected is determined to be qualified.

[0249] S183. In response to the fact that the image of the foreign object to be detected is not a lateral foreign object image, the quality of the image of the spine to be detected is determined to be unqualified.

[0250] Specifically, due to the wide variety of implants, existing generative large-scale models lack sufficient medical device knowledge to describe their details. Therefore, by combining the target detection bounding boxes obtained through a second pre-defined target detection model, the multimodal generative large-scale model is prompted to focus on specific areas, and by combining the image information of the entire X-ray film, a comprehensive judgment is made as to whether it is a lateral view. This step can improve the accuracy of the judgment and compensate for the lack of model knowledge.

[0251] In this scheme, a multimodal generative large model is used to determine whether the image of the foreign object to be detected is a lateral foreign object image, thereby determining whether the quality of the image of the spine to be detected is up to standard, which improves the accuracy and reliability of the quality evaluation of the image of the spine to be detected.

[0252] The working principle of the spine image quality assessment method in this embodiment is explained below with specific examples:

[0253] like Figure 28As shown, the quality evaluation of the spinal image to be detected begins with implant site detection. If no implant is found, the spinal target detection step is performed. If an implant is found, the implant image is segmented and an implant outline is generated to obtain the image of the foreign object to be detected. The image of the foreign object to be detected is matched with the lateral foreign object image to calculate the second similarity. If the second similarity does not meet the threshold condition, the spinal target detection step is performed. If the second similarity meets the threshold condition, a multimodal generative large model is used to describe the image of the foreign object to be detected, obtaining the second text description content. The text generative large model is used to determine whether the image of the foreign object to be detected is a reference image, such as whether it is a bone nail. If the image of the foreign object to be detected is not a reference image, the spinal target detection step is performed. If the image of the foreign object to be detected is a reference image, the multimodal generative large model is used to comprehensively determine the frontal and lateral views of the image of the foreign object to be detected. If the image of the foreign object to be detected is a lateral view, the output is a lateral view, and the quality of the spinal image to be detected is qualified. After performing the spinal target detection step, if no spinal region is found, the output is not a lateral view, and the quality of the spinal image to be detected is unacceptable. If a spinal region is found, a spinal region image is obtained. Semantic segmentation is performed on the spinal region image to obtain the vertebral body image to be detected. The vertebral body image to be detected is matched with the lateral vertebral body image to calculate the first similarity. If the first similarity meets the threshold condition, a multimodal generative model is used to describe the vertebral body image to be detected, and the first text description content is obtained. The text generative model is used to determine whether the vertebral body image to be detected is a reference vertebral body image. If the vertebral body image to be detected is a reference vertebral body image, the text generative model is used to determine whether the vertebral body image to be detected is a lateral vertebral body image. The number of targets corresponding to the lateral vertebral body images in the spinal image to be detected is counted, and the number of targets meets the threshold condition. If the number of targets meets the threshold condition, the output is a lateral view, and the quality of the spinal image to be detected is acceptable. If the number of targets does not meet the threshold condition, the output is not a lateral view, and the quality of the spinal image to be detected is unacceptable.

[0254] In this embodiment, the spinal image to be detected is segmented to obtain the vertebral body image to be detected. A generative large model is used to determine whether the vertebral body image to be detected is a lateral vertebral body image, thereby realizing the quality evaluation of the spinal image to be detected. This improves the efficiency of classifying the spinal image to be detected, reduces costs, improves the accuracy and reliability of spinal image quality evaluation, reduces the impact of abnormal X-rays generated in spinal-related medical practices on doctors' diagnoses and downstream information processing, has medical interpretability, conforms to the cognition of medical practitioners, and automates the process without the need for cumbersome and costly medical sample annotation.

[0255] Example 3

[0256] This embodiment provides a quality assessment system for spine images, such as... Figure 29 As shown, the quality evaluation system includes:

[0257] The spine image acquisition module 11 is used to acquire images of the spine to be detected.

[0258] The vertebral body image acquisition module 12 is used to segment the vertebral image to be detected in order to obtain several vertebral body images to be detected.

[0259] Lateral vertebral body determination module 13 is used to determine whether the vertebral body image to be detected is a lateral vertebral body image using a preset generative large model;

[0260] The target quantity acquisition module 14 is used to acquire the target quantity corresponding to the lateral vertebral images among several vertebral images to be detected in the spinal image to be detected.

[0261] The quality evaluation module 15 is used to evaluate the quality of the spine images to be detected based on the number of targets.

[0262] In this embodiment, the spinal image to be detected is segmented to obtain the vertebral body image to be detected. A generative large model is used to determine whether the vertebral body image to be detected is a lateral vertebral body image, thereby realizing the quality evaluation of the spinal image to be detected. This improves the efficiency of classifying the spinal image to be detected, reduces the cost, and improves the accuracy and reliability of the spinal image quality evaluation.

[0263] Example 4

[0264] This embodiment provides a quality assessment system for spinal images, which is a further improvement on Embodiment 3.

[0265] In a feasible scheme, the pre-defined generative large model includes a multimodal generative large model and a text-generative large model;

[0266] like Figure 30 As shown, the lateral vertebral body determination module 13 includes:

[0267] The first content acquisition unit 131 is used to input the image of the vertebra to be detected into the multimodal generative large model to obtain the first text description content;

[0268] The lateral vertebral body determination unit 132 is used to input the first text description content into the text-generating large model to determine whether the vertebral body image to be detected is a lateral vertebral body image.

[0269] In one feasible embodiment, the lateral vertebral body determination unit 132 includes:

[0270] Reference vertebral body determination subunit 1321 is used to input the first text description content into the text generation large model to determine whether the vertebral body image to be detected is a reference vertebral body image;

[0271] The lateral vertebral body determination subunit 1322 is used to respond to the vertebral body image to be detected being a reference vertebral body image, and to use the feature information of the reference vertebral body image as prompt words for the text-generating large model, so that the text-generating large model can determine whether the vertebral body image to be detected is a lateral vertebral body image.

[0272] In one feasible embodiment, the vertebral body image acquisition module 12 includes:

[0273] The region image acquisition unit 121 is used to input the image of the spine to be detected into the first preset target detection model to obtain the image of the spine region;

[0274] The vertebral image acquisition unit 122 is used to segment the image of the spinal region to obtain several vertebral images to be detected.

[0275] In one feasible embodiment, the vertebral body image acquisition unit 122 includes:

[0276] The image segmentation acquisition subunit 1221 is used to segment the spinal region image to obtain several segmented images to be detected;

[0277] The first similarity acquisition subunit 1222 is used to acquire the first similarity between the first contour information of the segmented image to be detected and the second contour information of the lateral vertebral body image.

[0278] The vertebral image determination subunit 1223 is used to determine whether the segmented image to be detected is a vertebral image to be detected based on the first similarity.

[0279] In one feasible scheme, the segmentation image acquisition subunit 1221 is further used to segment the spinal region image using a preset image segmentation model to obtain several segmented images to be detected.

[0280] In one feasible solution, the vertebral image determination subunit 1223 is further configured to determine the segmented image to be detected as a vertebral image to be detected in response to a first similarity being less than a first preset similarity; and to determine that the segmented image to be detected is not a vertebral image to be detected in response to a first similarity being not less than a first preset similarity.

[0281] In one feasible solution, the quality assessment module 15 includes:

[0282] The first response unit 151 is used to determine that the quality of the spine image to be detected is qualified in response to the target number being greater than the preset number.

[0283] The second response unit 152 is used to determine that the quality of the spinal image to be detected is unqualified in response to the target number not being greater than a preset number.

[0284] In one feasible solution, the quality assessment system also includes:

[0285] Foreign object determination module 16 is used to input the spinal image to be detected into the second preset target detection model to determine whether there is a foreign object in the spinal image to be detected;

[0286] The first response module 17 is used to respond to the absence of foreign objects in the spinal image to be detected by calling the vertebral image acquisition module to perform segmentation processing on the spinal image to be detected in order to obtain several vertebral images to be detected.

[0287] The second response module 18 is used to respond to the presence of the foreign object in the spinal image to be detected, obtain a foreign object region image based on the second preset target detection model; segment the foreign object region image to obtain several foreign object images to be detected; obtain a second similarity between the third contour information of the foreign object image to be detected and the fourth contour information of the lateral foreign object image; and, in response to the second similarity being not less than the second preset similarity, call the vertebral image acquisition module to perform the step of segmenting the spinal image to be detected to obtain several vertebral images to be detected.

[0288] In one feasible solution, the quality assessment system also includes:

[0289] The second content acquisition module 19 is used to input the image of the foreign object to be detected into the multimodal generative large model in response to the second similarity being less than the second preset similarity, so as to obtain the second text description content.

[0290] Reference foreign object determination module 20 is used to input the second text description content into the text generation large model to determine whether the image of the foreign object to be detected is a reference foreign object image;

[0291] The third response module 21 is used to respond to the fact that the image of the foreign object to be detected is not a reference image of the foreign object, and to call the vertebral image acquisition module to perform segmentation processing on the image of the vertebra to be detected in order to obtain several images of the vertebral body to be detected.

[0292] The fourth response module 22 is used to evaluate the quality of the spine image to be detected by using a multimodal generative large model in response to the image of the foreign object to be detected as a reference image.

[0293] In one feasible embodiment, the fourth response module 22 includes:

[0294] Lateral foreign object determination unit 221 is used to input the image of the foreign object to be detected into a multimodal generative large model to determine whether the image of the foreign object to be detected is a lateral foreign object image;

[0295] Image qualification determination unit 222 is used to determine the quality of the image of the spine to be detected as qualified in response to the fact that the image of the foreign object to be detected is a lateral foreign object image.

[0296] In this embodiment, the spinal image to be detected is segmented to obtain the vertebral body image to be detected. A generative large model is used to determine whether the vertebral body image to be detected is a lateral vertebral body image, thereby realizing the quality evaluation of the spinal image to be detected. This improves the efficiency of classifying the spinal image to be detected, reduces the cost, and improves the accuracy and reliability of the spinal image quality evaluation.

[0297] For the system embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs.

[0298] Example 5

[0299] Figure 31 This is a schematic diagram of the structure of an electronic device according to an example embodiment of the present disclosure. The electronic device includes a memory, a processor, and a computer program stored in the memory and used to run on the processor. When the processor executes the computer program, it implements the spine image quality evaluation method described in any of the above embodiments. Figure 31 The electronic device 90 shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments disclosed herein.

[0300] like Figure 31 As shown, the electronic device 90 can be manifested as a general-purpose computing device, such as a server device. The components of the electronic device 90 may include, but are not limited to: at least one processor 91, at least one memory 92, and a bus 93 connecting different system components (including memory 92 and processor 91).

[0301] Bus 93 includes a data bus, an address bus, and a control bus.

[0302] The memory 92 may include volatile memory, such as random access memory (RAM) 921 and / or cache memory 922, and may further include read-only memory (ROM) 923.

[0303] The memory 92 may also include a program tool 925 (or utility) having a set (at least one) program module 924, including but not limited to: an operating system, one or more application programs, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment.

[0304] The processor 91 executes various functional applications and data processing by running computer programs stored in the memory 92, such as the spinal image quality evaluation method provided in any of the above embodiments.

[0305] Electronic device 90 can also communicate with one or more external devices 94 (e.g., keyboard, pointing device, etc.). This communication can be performed through input / output (I / O) interface 95. Furthermore, electronic device 90 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public network, such as the Internet) via network adapter 96. As shown, network adapter 96 communicates with other modules of electronic device 90 via bus 93. It should be understood that, although not shown in the figure, other hardware and / or software modules can be used in conjunction with electronic device 90, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID (disk array) systems, tape drives, and data backup storage systems.

[0306] It should be noted that although several units / modules or sub-units / modules of the electronic device have been mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to embodiments of this disclosure, the features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided and embodied by multiple units / modules.

[0307] Example 6

[0308] This disclosure also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the spine image quality evaluation method provided in any of the above embodiments.

[0309] The readable storage medium may be more specifically adopted, including but not limited to: portable disk, hard disk, random access memory, read-only memory, erasable programmable read-only memory, optical storage device, magnetic storage device, or any suitable combination thereof.

[0310] Example 7

[0311] This disclosure also provides a computer program product, including a computer program that, when executed by a processor, implements the spine image quality evaluation method described in any of the preceding embodiments.

[0312] The program code for executing the computer program product of this disclosure can be written in any combination of one or more programming languages, and the program code can be executed entirely on a user device, partially on a user device, as a stand-alone software package, partially on a user device and partially on a remote device, or entirely on a remote device.

[0313] While specific embodiments of this disclosure have been described above, those skilled in the art should understand that these are merely illustrative examples, and the scope of protection of this disclosure is defined by the appended claims. Those skilled in the art can make various changes or modifications to these embodiments without departing from the principles and essence of this disclosure, but all such changes and modifications fall within the scope of protection of this disclosure.

Claims

1. A method for quality assessment of spinal images, characterized in that, The quality evaluation method includes: Acquire the image of the spine to be detected; The image of the spine to be detected is segmented to obtain several images of the vertebrae to be detected; A pre-defined generative large model is used to determine whether the vertebral image to be detected is a lateral vertebral image; Obtain the number of targets corresponding to the lateral vertebral images among several vertebral images to be detected corresponding to the spinal image to be detected; The quality of the spine image to be detected is evaluated based on the target quantity. Before the step of segmenting the image of the spine to be detected to obtain several images of the vertebrae to be detected, the method further includes: The image of the spine to be detected is input into a second preset target detection model to determine whether there is a foreign object in the image of the spine to be detected. In response to the absence of the foreign object in the spinal image to be detected, the step of segmenting the spinal image to be detected to obtain several vertebral images to be detected is performed. In response to the presence of the foreign object in the spinal image to be detected, a foreign object region image is obtained based on the second preset target detection model; the foreign object region image is segmented to obtain several images of foreign objects to be detected; a second similarity is obtained between the third contour information of the image of foreign objects to be detected and the fourth contour information of the lateral foreign object image; in response to the second similarity being not less than the second preset similarity, the step of segmenting the spinal image to be detected to obtain several images of vertebrae to be detected is performed.

2. The method for quality assessment of spinal images as described in claim 1, characterized in that, The preset generative large model includes a multimodal generative large model and a text generative large model; The step of determining whether the vertebral image to be detected is a lateral vertebral image using a preset generative large model includes: The image of the vertebra to be detected is input into the multimodal generative large model to obtain the first text description content; The first text description is input into the text-generating large model to determine whether the vertebral image to be detected is the lateral vertebral image.

3. The method for quality assessment of spinal images as described in claim 2, characterized in that, The step of inputting the first text description into the text generation model to determine whether the vertebral image to be detected is the lateral vertebral image includes: The first text description is input into the text-generating large model to determine whether the vertebral image to be detected is a reference vertebral image. In response to the fact that the vertebral body image to be detected is the reference vertebral body image, the feature information of the reference vertebral body image is used as the prompt word of the text-generating large model, so that the text-generating large model can determine whether the vertebral body image to be detected is the lateral vertebral body image.

4. The method for evaluating the quality of spinal images as described in any one of claims 1-3, characterized in that, The step of evaluating the quality of the spine image to be detected based on the target quantity includes: In response to the target quantity being greater than a preset quantity, the quality of the spine image to be detected is determined to be qualified; In response to the fact that the target number is not greater than the preset number, the quality of the spine image to be detected is determined to be unqualified.

5. The method for evaluating the quality of spinal images as described in any one of claims 1-3, characterized in that, The quality assessment method also includes: In response to the second similarity being less than the second preset similarity, the image of the foreign object to be detected is input into a multimodal generative large model to obtain the second text description content; The second text description is input into a text-generating large model to determine whether the image of the foreign object to be detected is a reference image of the foreign object. In response to the fact that the image of the foreign object to be detected is not the reference foreign object image, the step of segmenting the image of the spine to be detected to obtain several images of the vertebrae to be detected is performed. In response to the fact that the image of the foreign object to be detected is the reference image of the foreign object, the quality of the image of the spine to be detected is evaluated using the multimodal generative large model.

6. A quality assessment system for spine images, characterized in that, The quality evaluation system includes: The spine image acquisition module is used to acquire images of the spine to be detected. The vertebral body image acquisition module is used to segment the image of the spine to be detected to obtain several vertebral body images to be detected. The lateral vertebral body determination module is used to determine whether the image of the vertebral body to be detected is a lateral vertebral body image using a preset generative large model; The target quantity acquisition module is used to acquire the target quantity corresponding to the lateral vertebral image among several vertebral images to be detected corresponding to the spinal image to be detected; The quality evaluation module is used to evaluate the quality of the spine image to be detected based on the target quantity. The quality evaluation system also includes: The foreign object detection module is used to input the spinal image to be detected into a second preset target detection model to determine whether there is a foreign object in the spinal image to be detected. The first response module is used to respond to the absence of the foreign object in the spinal image to be detected by calling the vertebral image acquisition module to perform the step of segmenting the spinal image to be detected to obtain several vertebral images to be detected. The second response module is used to respond to the presence of the foreign object in the spinal image to be detected, obtain a foreign object region image based on the second preset target detection model; segment the foreign object region image to obtain several images of foreign objects to be detected; obtain a second similarity between the third contour information of the image of foreign objects to be detected and the fourth contour information of the lateral foreign object image; and, in response to the second similarity being not less than the second preset similarity, call the vertebral image acquisition module to perform the step of segmenting the spinal image to be detected to obtain several images of vertebral bodies to be detected.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and for running on the processor, characterized in that, When the processor executes the computer program, it implements the method for evaluating the quality of spinal images according to any one of claims 1-5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method for evaluating the quality of spinal images as described in any one of claims 1-5.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the method for evaluating the quality of spine images as described in any one of claims 1-5.

Images