MRI lumbar sagittal view image disease diagnosis method based on YOLOv8s algorithm

Abstract

The invention belongs to the technical field of medical diagnosis, and discloses an MRI lumbar sagittal view disease image diagnosis method based on a YOLOv8s algorithm, and the method comprises the following steps: S1, screening collected MRI lumbar sagittal view image data, self-building a VOC2007 data set, and marking the data set; s2, performing data enhancement and other processing on the existing data set; s3, establishing a YOLOv8s deep learning neural network detection model; dividing the data set into a training set, a verification set and a test set according to a proportion of 8: 1: 1, training the YOLOv8s network until convergence, and obtaining a lumbar disease image detection model based on the YOLOv8s network and a weight file; and S4, testing the randomly extracted MRI lumbar vertebra sagittal view image of the patient by using the trained YOLOv8s detection model, outputting a detection result, and evaluating the detection result. According to the invention, the diagnosis of the MRI lumbar sagittal view disease can be realized, the detection precision and efficiency are greatly improved, and the diagnosis accuracy is improved.

Description

Technical Field

[0001] The present invention relates to biomedicine, deep learning and automatic recognition of nuclear magnetic resonance imaging image features, and in particular to a method for diagnosing MRI lumbar sagittal image diseases based on a YOLOv8s algorithm. Background Art

[0002] The lumbar spine is an important connecting structure of the body. It carries the weight of the upper body through the sacrum and pelvis. Due to the long-term bearing effect, it is easy to cause low back pain. According to epidemiological statistics, low back pain ranks in the top 10 of the global disease burden, seriously affecting people's quality of life. Lumbar diseases mainly include lumbar disc herniation, lumbar muscle strain, lumbar degenerative lesions, lumbar spinal stenosis, etc., among which lumbar disc herniation is one of the main diseases of the lumbar spine. The lumbar disc is composed of two basic elements: a central nucleus surrounded by fibrous nuclei located in two vertebrae. When a part of the nucleus pulposus protrudes outward, it squeezes the corresponding nerve roots and causes compression symptoms, which will be accompanied by the occurrence of low back pain symptoms. Early screening of lumbar diseases can greatly improve the diagnosis and treatment rate. It is the best way to accurately predict lumbar diseases, diagnose lumbar diseases early and treat them in time. Magnetic resonance imaging (MRI) can show the images of soft tissues and lumbar bones in the human body more clearly. Medical staff can use MRI to analyze and determine the type and severity of the patient's lesions, so as to decide whether further treatment is needed, track the treatment effect and detect changes in the condition. Medical staff screen and identify lumbar disc diseases based on MRI images, but the process of screening and identifying a large number of lumbar MRI images is time-consuming and cumbersome, and doctors rely too much on previous experience, which may lead to misdiagnosis or missed diagnosis during the diagnosis process. The results of deep learning algorithms for locating and diagnosing MRI lumbar sagittal diseases can assist doctors in making more labor-saving and accurate diagnoses. The present invention belongs to the field of medical imaging technology, and computer-aided diagnosis technology based on deep learning is combined with medical images and applied to medical clinical experiments.

[0003] Lumbar spine diseases have various shapes, so a lot of medical image data of lumbar spine diseases is needed in the model training process. The usual practice is to first screen out the clearest and most characteristic MRI lumbar spine sagittal images and establish a data set, and then use the deep learning method to train a large number of image training sets with the model to generate a trained model; then, construct a new MRI lumbar spine sagittal image verification set to verify the model and obtain the final recognition result. In clinical diagnosis, when doctors observe MRI lumbar spine sagittal medical images, they need to spend a lot of time to make a diagnosis and determine the location and shape of lumbar spine diseases. According to the survey, the current deep learning model has not yet made the above-mentioned related diagnoses.

[0004] Due to the heavy workload of medical staff, which leads to low efficiency and low diagnostic accuracy, there is still room for improvement in the accuracy of identification and detection, the convenience of automatic operation, and automatic qualitative analysis. Therefore, with the help of artificial intelligence deep learning, combined with knowledge of magnetic resonance imaging and lumbar spine diseases, a safe, efficient and accurate diagnostic method to assist physicians is created, thereby providing more effective and accurate assistance for the correct diagnosis of lumbar spine diseases. This is a method for diagnosing MRI lumbar sagittal images based on the YOLOv8s algorithm. Summary of the invention

[0005] The purpose of the present invention is to solve the above-mentioned problems in the field of MRI lumbar sagittal image disease diagnosis, and to provide an MRI lumbar sagittal image disease diagnosis method based on the YOLOv8s algorithm, which can help clinicians of lumbar diseases to make correct diagnosis and treatment and improve the accuracy of lumbar disc disease diagnosis.

[0006] In order to solve the above-mentioned problems in the diagnosis of lumbar vertebrae diseases, the technical solution adopted by the present invention is: a method for diagnosing MRI lumbar vertebrae sagittal images based on the YOLOv8s algorithm.

[0007] A method for diagnosing MRI lumbar sagittal image diseases based on the YOLOv8s algorithm, characterized by comprising:

[0008] S1, construct the VOC2007 data set, randomly select 300 patients from the original unclassified patients, screen out the clearest and most characteristic MRI lumbar sagittal images among the patients, and the sagittal images of all patients are of uniform size, thus forming the MRI lumbar sagittal image VOC2007 data set. Then use LabelImg software to label the data set, mark the locations with symptoms in the lumbar sagittal images, and form a .xml label file.

[0009] S2, traditional data augmentation is performed on the constructed VOC2007 dataset, which is implemented as follows: ① Set scale = (.25, 2) to randomly scale the dataset between 0.25-2; ② Set jitter = .3 to randomly distort the dataset between 0.7 / 1.3-1.3 / 0.7; Set hue = .1, sat = 0.7, val = 0.4 to randomly change the hue, saturation and brightness of the dataset.

[0010] S3, dividing the data set into a training set, a validation set and a test set. Based on the training set and the validation set, the YOLOv8s network is trained until convergence, and a lumbar sagittal image detection model based on the YOLOv8s network and a weight file are obtained; based on the test set, the YOLOv8s detection model is tested to obtain a test result, and the YOLOv8s detection model is evaluated based on the test result. If the evaluation is qualified, the test is terminated.

[0011] S4, verifying the YOLOv8s detection model, and verifying the randomly selected MRI lumbar sagittal images of 300 patients not included in S1 based on the YOLOv8s detection model.

[0012] Preferably, the specific steps of constructing the data set in S1 are as follows: the original unclassified patient MRI lumbar sagittal image contains 30 images of different signals of uniform size, and the image format type is JPG, among which the T2-weighted signal image is more conducive to observing the characteristic morphology of the lesion and accurately judging the lumbar spine disease, and the T2-weighted signal MRI image with the clearest image and obvious image features is screened out as the data set.

[0013] Preferably, the data set is divided in S3, and the specific steps are: dividing the data in the data set into a training set, a validation set and a test set in a ratio of 8:1:1.

[0014] Preferably, the YOLOv8s network includes a backbone part, a neck part and a prediction part. In S3, the specific method for obtaining the YOLOv8s detection model is:

[0015] S3.1, put the .xml label files and original image files annotated by LabelImg software into the VOC2007 folder, and divide the dataset into training set, validation set and test set in a ratio of 8:1:1 through the neural network program.

[0016] S3.2, set training parameters: input image size is 512 pixels × 512 pixels, freeze training for 50 epochs, batch size is 16, unfreeze training for 260 epochs, a total of 260 epochs, batch size is 8; the maximum learning rate is set to 0.01; use the adam optimizer to optimize the network parameters; use the learning rate cosine annealing decay, that is, set 'Freeze_Epoch = 50', 'Freeze_batch_size = 16', 'UnFreeze_Epoch = 260', 'UnFreeze_batch_size = 8', 'optimizer_type = adam'.

[0017] S3.3, inputting the MRI lumbar sagittal images in the training set into the backbone part, i.e., the CSPdarknet-53 network, by reducing the number of convolutions, mixing the convolution layer and the batch normalization layer and introducing the C2f module, so as to dynamically identify and obtain feature information of different scales;

[0018] S3.4, inputting the feature information of different scales into the neck part, performing upsampling and downsampling in sequence, performing convolution to perform feature fusion on the feature information of different scales in sequence, and obtaining tensor data of different scales;

[0019] S3.5, input the tensor data of different scales into the prediction part, calculate the gradient based on the Task-AlignedAssigner positive and negative sample matching method, introduce the DFL loss function and back propagation, update the gradient in real time and verify it using the validation set to obtain the YOLOv8s detection model. By defining a loss function, the average loss generated by the model prediction can be calculated, that is, in all possible cases, the model's prediction is the average difference between the labels, where the DFL loss function is a loss function for image generation tasks, which can help the generator network learn better image feature representations. The calculation formula of the DFL loss function is as follows: L_DFL = ||F(x)-F(G(z))|| 2 , where L_DFL represents the DFL loss function, F(x) represents the feature representation of the real image x, G(z) represents the image generated by the generator network, and F(G(z)) represents the feature representation of the generated image. ||.|| 2 Represents the square of the Euclidean distance.

[0020] Preferably, the requirement for passing the evaluation in S3 is that the average precision mean of the test results of the YOLOv8s detection model is greater than a set value.

[0021] Preferably, the YOLOv8s detection model is verified, and the randomly selected MRI lumbar sagittal images of 300 patients that do not include S1 are verified based on the YOLOv8s detection model.

[0022] S4.1, randomly select the clearest and most characteristic lumbar sagittal images of patients (excluding the 300 patients described in S1) to verify the YOLO v8s detection model.

[0023] S4.2, input the lumbar sagittal image described in S4.1 into the YOLOv8s detection model, filter the target detection frames based on the confidence, delete the detection frames and features with low confidence, select the prediction frames whose scores meet the confidence, and select the detection frames with the largest scores of the same category in a certain area through non-maximum suppression. That is, set 'confidence=0.5'.

[0024] The MRI lumbar sagittal image disease diagnosis method based on the YOLOv8s algorithm is characterized by collecting and screening the clearest and most prominent MRI lumbar sagittal images, using the YOLOv8s model to detect diseases in the lumbar sagittal MRI images, the location information and confidence of the disease, and judging whether the lumbar intervertebral disc has a disease for the detected candidate frame.

[0025] The beneficial effects of the present invention are:

[0026] The present invention proposes a method for diagnosing lumbar sagittal image diseases of MRI based on the YOLOv8s algorithm. The present invention trains a detection model based on a self-assembled data set and a YOLOv8s network. The model can realize the target detection function, and then diagnose lumbar sagittal diseases of other patients based on the trained YOLOv8s detection model.

[0027] The method of the present invention can accurately identify the location of lumbar sagittal lesions, which can prevent medical personnel from neglecting some tiny lesions due to fatigue or lack of experience during observation, resulting in inaccurate manual diagnosis. It can assist medical personnel in focusing on reviewing the lesion area detected by this method.

[0028] The present invention uses the YOLOv8s algorithm to detect MRI lumbar sagittal image diseases. YOLOv8s uses the Pytorch framework to easily train its own data set. Compared with other detection algorithms, the YOLOv8s detection model has the following advantages in medical diagnosis: first, it performs well in detection accuracy and speed, and can meet the requirements of clinical real-time detection; second, this method has strong robustness to the scale change of the target and can handle lumbar disc lesions of different sizes and shapes.

[0029] In view of the increasing prevalence of lumbar diseases, insufficient professional medical resources, and time-consuming and inefficient manual diagnosis, this paper proposes a method for diagnosing lumbar sagittal MRI images based on the YOLOv8s algorithm. By using lumbar disease magnetic resonance imaging image data and deep learning and other technologies, the automatic feature extraction and intelligent diagnosis of lumbar disease MRI medical images are realized. The entire diagnostic process is safe, efficient, fast, and has a high diagnostic accuracy rate. This method can accurately identify the specific location of lumbar disc disease, providing doctors with a reliable reference so that they can better formulate treatment plans. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0031] Figure 1 It is a schematic diagram of the diagnostic process of MRI lumbar sagittal image diseases of the present invention;

[0032] Figure 2 It is a schematic diagram of the symptom annotation of the MRI lumbar sagittal image of the present invention;

[0033] Figure 3 This is the F1 graph of the training evaluation parameters of the YOLOv8s target detection model with different data enhancements.

[0034] Figure 4 This is the mAP graph of the training evaluation parameters of different models based on the YOLOv8s target detection model.

[0035] Figure 5 This is the result of the detection and diagnosis based on the YOLOv8s target detection model of the present invention. DETAILED DESCRIPTION

[0036] Various exemplary embodiments of the present invention will now be described in detail. This detailed description should not be considered as limiting the present invention, but should be understood as a more detailed description of certain aspects, features, and embodiments of the present invention.

[0037] It should be understood that the terms described in the present invention are only for describing special embodiments and are not intended to limit the present invention. In addition, for the numerical range in the present invention, it should be understood that each intermediate value between the upper and lower limits of the scope is also specifically disclosed. Each smaller range between the intermediate value in any stated value or stated range and any other stated value or intermediate value in the described range is also included in the present invention. The upper and lower limits of these smaller ranges can be independently included or excluded in the scope.

[0038] Unless otherwise indicated, all technical and scientific terms used herein have the same meanings as commonly understood by those skilled in the art to which the invention belongs. Although the present invention describes only preferred methods and materials, any methods and materials similar or equivalent to those described herein may also be used in the implementation or testing of the present invention. All documents mentioned in this specification are incorporated by reference to disclose and describe the methods and / or materials related to the documents. In the event of a conflict with any incorporated document, the content of this specification shall prevail.

[0039] It will be apparent to those skilled in the art that various modifications and variations may be made to the specific embodiments of the present invention description without departing from the scope or spirit of the present invention. Other embodiments derived from the present invention description will be apparent to those skilled in the art. The present application description and examples are exemplary only.

[0040] The words “include,” “comprising,” “have,” “containing,” etc. used in this document are open-ended terms, meaning including but not limited to.

[0041] The technical solution of the present invention is further described below in conjunction with specific implementation methods.

[0042] Example 1

[0043] The scheme of the present invention mainly embodies the basic idea of ​​intelligent diagnosis, high efficiency, speed and high accuracy. A method for diagnosing MRI lumbar sagittal image diseases based on the YOLOv8s algorithm comprises the following steps:

[0044] (1) Preparation of MRI lumbar sagittal image data: screening and labeling the prepared MRI lumbar sagittal image data and self-assembling the data set;

[0045] (2) Perform traditional data enhancement processing on the labeled data set, and divide the data set after traditional data enhancement processing into training set, validation set and test set;

[0046] (3) Use the training set and validation set in the self-assembled data set to train the YOLOv8s network and obtain the MRI lumbar sagittal image disease detection model and weight file based on the YOLOv8s network;

[0047] (4) Use the YOLOv8s detection model trained in step (3) to detect the test set, output the detection results, and evaluate the detection model;

[0048] (5) Construct a new MRI lumbar sagittal medical image validation set;

[0049] (6) Verify the trained MRI lumbar sagittal medical image disease diagnosis model;

[0050] Preferably, the specific method in step (1) is as follows:

[0051] (1) 300 patients were randomly selected from the original unclassified patients, and the MRI lumbar sagittal images with uniform size, the clearest and most obvious features were screened out to form the MRI lumbar sagittal image VOC2007 dataset.

[0052] (2) Label the self-assembled VOC2007 dataset using LabelImg software, mark the locations with symptoms in the lumbar sagittal images, and form an .xml label file. Put the constructed original dataset and .xml label file into the VOC2007 folder.

[0053] Preferably, the traditional data enhancement in step (2) includes any one or more of the following steps:

[0054] (1) Randomly scale the labeled dataset images between 0.25 and 2;

[0055] (2) performing random distortion between 0.7 / 1.3 and 1.3 / 0.7 on the labeled dataset images;

[0056] (3) Randomly change the hue, saturation and brightness of the labeled dataset images to hue=.1, sat=0.7, val=0.4;

[0057] (4) Divide the dataset after traditional data augmentation into training set, validation set and test set;

[0058] Preferably, in step (3), the process of training the YOLOv8s network using the training set and the test set in the self-assembled data set to obtain the MRI lumbar sagittal image disease diagnosis model based on the YOLOv8s network is as follows:

[0059] Preferably, the configuration file of the YOLOv8s network is modified according to the image data category annotated in the self-assembled data set, the number of categories is reset, the images in the training set are standardized and preprocessed, and the processed images are input into the backbone part of the YOLOv8s network to obtain feature information of different scales, the feature information of different scales is input into the neck part for upsampling and downsampling convolution respectively, the feature information of different scales is sequentially fused to obtain tensor data of different scales, and then input into the prediction part for gradient calculation through loss function and back propagation, the gradient is updated in real time and verified using the validation set, and finally an MRI lumbar sagittal image disease detection model and weight file based on the YOLOv8s network are obtained.

[0060] Preferably, the YOLOv8s network training parameters in step (3) are specifically set as follows:

[0061] Set training parameters: input image size is 512 pixels × 512 pixels, freeze training for 50 epochs, batch size is 16, unfreeze training for 260 epochs, a total of 260 epochs, batch size is 8; the maximum learning rate is set to 0.01; use the adam optimizer to optimize the network parameters; use the learning rate cosine annealing decay, that is, set 'Freeze_Epoch = 50', 'Freeze_batch_size = 16', 'UnFreeze_Epoch = 260', 'UnFreeze_batch_size = 8', 'optimizer_type = adam'.

[0062] Preferably, in step (4), the detection model is evaluated, and mAP is used as the accuracy indicator of the detection model, which represents the mean average precision. At the same time, the F1 value is used as an evaluation indicator to determine the relationship between the Recall value and the Precision value, which can be directly generated by a program.

[0063] In order to use only one number to represent the performance of a model, the average precision values ​​of all categories are calculated as the average (average / mean). This new value is the mean average precision mAP (MeanAverage Precision):

[0064]

[0065] Where n is the total number of categories, i is the current category, and AP(i) is the AP of the current category.

[0066] F1-Score is the harmonic mean of precision and recall, taking both into account, so it can more comprehensively evaluate the performance of the model. The higher the F1 score, the better the performance of the model.

[0067]

[0068] Among them, TP is the number of positive samples classified and corrected to positive samples, the number of positive samples misclassified as negative numbers is represented by FN, TN is the number of negative samples classified and corrected to negative numbers, and FP is the number of negative samples misclassified as positive numbers.

[0069] Preferably, the specific steps in step (6) are as follows:

[0070] (1) Randomly select the clearest and most characteristic lumbar sagittal images of patients (excluding the data set described in step (1)) to construct a new validation data set and input it into the YOLOv8s detection model.

[0071] (2) Filter the target detection frames based on confidence, delete the detection frames and features with confidence less than 0.5, filter out the prediction frames whose scores meet the confidence level, and filter out the detection frames with the largest scores of the same category within a certain area through non-maximum suppression.

[0072] (3) Verification results are as follows Figure 5 As shown, if the patient's lumbar intervertebral disc has abnormal symptoms, the program will report "abnormal". If the patient's lumbar intervertebral disc is normal, the program will not report "abnormal".

[0072] (7) The present invention is based on the MRI lumbar sagittal image disease diagnosis method of the YOLOv8s algorithm. In order to verify that the selected algorithm is superior to other algorithms in medical diagnosis, the present invention also selects YOLOv4-tiny-CBEM, YOLOv5s, YOLOv5m, and YOLOv7 to perform diagnostic tests on the lumbar sagittal image datasets.

[0073] (8) The comparison results are shown in Table 1.

[0074] (9) Table 1 Results comparison chart

[0075]

[0076] As can be seen from Table 1, the YOLOv8s model after traditional data enhancement of the lumbar sagittal data set has higher mean average precision (mAP), F1 value and accuracy (Precision) in disease diagnosis than other models. Compared with the YOLOv4-tiny-CBEM, YOLOv5s, YOLOv5m, and YOLOv7 models, the mAP value increased by 0.39%, 0.44%, 0.25%, and 1%, respectively, the F1 value increased by 2%, 2%, 1%, and 4%, respectively, and the diagnostic Precision value increased by 5.13%, 5.72%, 7.98%, and 9.39%, respectively. It can be seen that the YOLO v8s model has good performance in the diagnosis of lumbar disc disease and has achieved good results.

[0077] We will further improve the method of the present invention into a complete set of intelligent diagnosis process for MRI lumbar spine medical images, continuously expand our MRI lumbar spine medical image data set, continuously update the data source for the present invention, optimize the iterative process of model training and verification, and continuously tune and update the parameters of the deep learning network model, thereby greatly improving the diagnostic accuracy of MRI lumbar spine medical images of the present invention. By studying a large number of cases and mining the important information and inherent laws of the disease, a complete set of lumbar disease feature induction and disease diagnosis models will be established, and finally artificial intelligence-assisted diagnosis of lumbar diseases will be realized.

[0078] The above is only a preferred embodiment of the present invention, and does not limit the present invention in any form. Although the present invention has disclosed the preferred embodiment as above, it is not used to limit the present invention. Any technician familiar with the profession can make some changes or modify the above technical content into an equivalent implementation scheme without departing from the scope of the technical solution of the present invention. However, any simple modification, equivalent change and modification made to the above implementation content according to the technical essence of the present invention without departing from the content of the technical solution of the present invention still falls within the scope of the technical solution of the present invention.

Claims

1. A method for diagnosing MRI lumbar sagittal images based on the YOLOv8s algorithm, characterized in that: The following steps are involved: S1, construct the VOC2007 dataset. Randomly select 300 patients from the original unclassified patients, and screen out the clearest and most characteristic MRI lumbar sagittal images of the patients. The sagittal images of all patients are of uniform size, thus forming the MRI lumbar sagittal image VOC2007 dataset. Then use LabelImg software to label the dataset, mark the locations with symptoms in the lumbar sagittal images, and form a .xml label file; S2, traditional data augmentation is performed on the constructed VOC2007 dataset, which is implemented as follows: ① Set scale = (.25, 2) to randomly scale the dataset between 0.25-2; ② Set jitter = .3 to randomly distort the dataset between 0.7 / 1.3-1.3 / 0.7; Set hue = .1, sat = 0.7, val = 0.4 to randomly change the hue, saturation and brightness of the dataset. S3, dividing the data set into a training set, a validation set and a test set. Based on the training set and the validation set, the YOLOv8s network is trained until convergence, and a lumbar sagittal image detection model based on the YOLOv8s network and a weight file are obtained; Based on the test set, the YOLOv8s detection model is tested to obtain a test result, and the YOLOv8s detection model is evaluated based on the test result, and the test is terminated if the evaluation is qualified; The YOLOv8s network includes a backbone part, a neck part and a prediction part. In S3, the specific method for obtaining the YOLOv8s detection model is: S3.1, put the .xml label files and original image files annotated by LabelImg software into the VOC2007 folder, and divide the dataset into training set, validation set and test set in a ratio of 8:1:1 through the neural network program. S3.2 Set training parameters: input image size is 512 pixels × 512 pixels, freeze training for 50 epochs, batch size is 16, unfreeze training for 260 epochs, a total of 260 epochs, batch size is 8; the maximum learning rate is set to 0.01; use the adam optimizer to optimize the network parameters; use the learning rate cosine annealing decay, that is, set 'Freeze_Epoch = 50', 'Freeze_batch_size = 16', 'UnFreeze_Epoch = 260', 'UnFreeze_batch_size = 8', 'optimizer_type = adam'. S3.3, inputting the MRI lumbar sagittal images in the training set into the backbone part, i.e., the CSPdarknet-53 network, by reducing the number of convolutions, mixing the convolution layer and the batch normalization layer and introducing the C2f module, so as to dynamically identify and obtain feature information of different scales; S3.4, inputting the feature information of different scales into the neck part, performing upsampling and downsampling in sequence, performing convolution to perform feature fusion on the feature information of different scales in sequence, and obtaining tensor data of different scales; S3.5, input the tensor data of different scales into the prediction part, calculate the gradient based on the Task-AlignedAssigner positive and negative sample matching method, introduce the DFL loss function and back propagation, update the gradient in real time and verify it using the validation set to obtain the YOLOv8s detection model. By defining a loss function, the average loss generated by the model prediction can be calculated, that is, in all possible cases, the model's prediction is the average difference between the labels, where the DFL loss function is a loss function for image generation tasks, which can help the generator network learn better image feature representations. The calculation formula of the DFL loss function is as follows: L_DFL = ||F(x)-F(G(z))|| 2 , where L_DFL represents the DFL loss function, F(x) represents the feature representation of the real image x, G(z) represents the image generated by the generator network, and F(G(z)) represents the feature representation of the generated image. ||.|| 2 Represents the square of the Euclidean distance. S4, verifying the YOLOv8s detection model, and verifying the randomly selected MRI lumbar sagittal images of 300 patients not included in S1 based on the YOLOv8s detection model. S4.1, randomly select the clearest and most characteristic lumbar sagittal images of patients (excluding the 300 patients described in S1) to verify the YOLOv8s detection model. S4.2, input the lumbar sagittal image described in S4.1 into the YOLOv8s detection model, filter the target detection frames based on the confidence, delete the detection frames and features with low confidence, select the prediction frames whose scores meet the confidence, and select the detection frames with the largest scores of the same category in a certain area through non-maximum suppression. That is, set 'confidence=0.5'.

2. The MRI lumbar sagittal image disease diagnosis method based on the YOLOv8s algorithm according to claim 1, characterized in that: The method for preparing the MRI lumbar sagittal image VOC2007 dataset is as follows: each patient's MRI scans will present a lumbar sagittal image that is blurred or has unclear features due to the scanned marginal tissue or bones, and such images are not suitable for diagnosis and cannot be used to judge the disease. Therefore, for each patient, the clearest and most distinctive lumbar sagittal image is selected to prepare the dataset.

3. The MRI lumbar sagittal image disease diagnosis method based on the YOLOv8s algorithm according to claim 1, characterized in that: The requirement for passing the evaluation in S3 is that the average precision mean of the test results of the YOLOv8s detection model is greater than or equal to a set value.

4. The MRI lumbar sagittal image disease diagnosis method based on the YOLOv8s algorithm according to claim 1, characterized in that: The processing method in S4.2 is: initialize various parameters and remove detection frames with detection confidence less than 0.5; set 'nms_iou=0.3', and remove the overlapping detection frames in the detection by non-maximum suppression method.

5. A method for diagnosing MRI lumbar sagittal image diseases based on the YOLOv8s algorithm, characterized in that: The clearest and most prominent MRI lumbar sagittal images are collected and screened, and the YOLOv8s model is used to detect the disease in the lumbar sagittal MRI images, the location information and confidence of the disease, and the detected candidate boxes are judged whether the lumbar intervertebral disc has a disease.

Images