Application of YOLO-v8 model based on lumbar X-ray in grading diagnosis of lumbar facet joint of osteoarthritis

Application of YOLO-v8 model based on lumbar X-ray in grading diagnosis of lumbar facet joint of osteoarthritis | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Application of YOLO-v8 model based on lumbar X-ray in grading diagnosis of lumbar facet joint of osteoarthritis baisen chen, yuyu sun, jiaming cui, tianqi wu, guanhua xu, zhiming cui This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7540599/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Purpose Lumbar facet joint of osteoarthritis (LFJOA) can cause intractable low back pain in patients. Early evaluation of the status of LFJOA is very important for subsequent treatment. This paper discusses the automatic segmentation and detection of LFJOA by studying the characteristics of artificial intelligence technology and its potential application in medical image analysis. Methods The ability to detect inflammation has been significantly enhanced in recent years due to deep learning technology, especially models based on object detection. This study collected 987 lateral lumbar X-ray from 987 patients, each of which was manually divided into five lumbar facet joint segments. According to the computed tomography (CT) image of each patient, the classification annotation was carried out based on weishaupt standard. Then, the you only look once (YOLO)-v8 model was used for hierarchical diagnosis. Precision, recall, f1 score, mean average precision (map)50, and map50-95 were used to evaluate the model's performance. Additionally, the research examined how this technology could be applied in clinical settings. Results In detecting facet arthritis, the YOLO-v8 model reached a map50 of 0.694, a map50-95 of 0.286, an F1 score of 0.64, a precision rate of 0.71, and a recall of 0.689. Conclusion YOLO-v8 has diagnostic value in detecting the severity of LFJOA. Future research should the model’s classification potential to enhance its clinical application settings, and help spinal surgeons more effectively diagnose the severity of lumbar facet arthritis, so as to formulate accurate treatment plans. Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 1 Introduction Lumbar facet joint of osteoarthritis (LFJOA) also known as lumbar facet osteoarthritis, involves intervertebral joints located at the back of the spine. Its pathological characteristics are cartilage degeneration, subchondral bone sclerosis and reactive proliferation of joint edges [ 1 , 2 ]. The condition is prevalent in middle-aged and senior populations, often leading to low back pain and movement difficulties [ 3 , 4 ]. In recent years, the incidence of LFJOA has been on the rise, seriously reducing the daily life and work status of patients [ 5 , 6 , 7 ], so the diagnosis and treatment of disease has been one of the research hotspots. Magnetic resonance imaging (MRI) can image articular cartilage, synovium, joint cavity effusion and other soft tissues, but its sensitivity to show the edge of bone cortex is low, and often underestimates the severity of LFJOA [ 8 , 9 ]. For diagnosing LFJOA, computed tomography (CT) is the most reliable imaging method, which can clearly show the changes of bone tissue and surrounding soft tissue [ 10 ]. Weishaupt proposed a system for diagnosing and grading LFJOA in CT images [ 11 ]. Indicators of the system encompass narrowing of the joint space, formation of osteophytes, hypertrophy of the articular process, erosion of subarticular bone, subchondral cysts, and the vacuum joint phenomenon [ 12 ]. However, the classification of LFJOA needs to be judged by clinicians and radiologists with rich clinical experience. X-ray is a commonly used imaging examination method, which is used to evaluate the lumbar bone structure [ 10 ]. Lumbar X-ray has the advantages of fast and convenient, small radiation, wide application range, etc. Therefore, this paper proposes a conjecture, whether to find an effective means to automatically identify the facet joint position and assess the severity in the lateral position of lumbar X-ray, in order to develop a treatment plan promptly and avert complications. At present, the application of artificial intelligence algorithm combined with imaging radiomics in osteoarthritis mainly focuses on shoulder joint, hip and knee joint and other large joints [ 13 , 14 ]. Martin Magnéli et al. can recognize and grade humeral head glenohumeral arthritis (GHOA) on X-ray by collecting 7139 shoulder X-ray to train the neural network model [ 15 ]. Another study proposed a neural network model to predict the progress of hip OA and determine its precision based on hip X-ray plain film and patient data [ 16 ]. These results indicate that the deep learning model has great potential in the use of X-ray to identify and assess the severity of inflammation. You only look once (YOLO)-v8 is the latest target detection model developed by Ultralytics, which provides model variants dedicated to detection, segmentation, key point estimation and other tasks [ 17 ]. Currently, research has gathered breast X-rays from female participants, employed the YOLO-v8 target detection algorithm to identify breast microcalcifications, and examined its performance benefits and clinical applications [ 18 ]. This shows that YOLO-v8 has great potential in identifying lesions on X-ray images. Therefore, we focus on facet arthritis disease. Utilizing the lateral X-ray of lumbar spine, we use the YOLO-v8 model to automatically identify and assess the severity of lumbar facet joints. Through the hierarchical diagnosis of disease by this model, doctors can customize personalized treatment plans for each patient. This means that the treatment method can be adjusted according to the classification of patients, so as to optimize the treatment effect and reduce unnecessary treatment and its related risks and costs. 2 Research methods Approved by the ethics review committee of Nantong First People's Hospital (No. 2024KT411), this study did not require informed consent due to its retrospective design. As shown in Fig. 1 , the research process for this study involves collecting images, preprocessing them, training the model, and evaluating its performance. The following paragraphs will elaborate on these processes (Fig. 1 flow chart). 2.1 Study cohort This research was conducted retrospectively. The subjects were selected from patients treated in our hospital from May 2023 to June 2025. The inclusion criteria were as follows: (1) patients who had received lateral lumbar X-ray examination and lumbar CT in our hospital at the same time within one week; (2) The age of patients was 18–90 years old; Exclusion criteria included: (1) the patient had previously received lumbar surgery; (2) The image is unclear or poor quality; (3) The region of interest (ROI) is difficult to cut. 2.2 Data acquisition The X-ray examination was performed by UDR 780i pro of Shanghai Lianying company. The CT examination utilized the Philips Ingenuity Core CT. Experienced spinal surgeons conducted the image acquisition to maintain process consistency. All obtained images are in the form of gray-scale digital imaging and JPG files with original resolution and color depth. In addition to the images, the corresponding clinical data, age, gender and body mineral index (BMI) of the patients were also obtained. 2.3 Image processing In the preprocessing stage, the image is enhanced by panning, zooming, and flipping the image. Two spine surgeons with clinical experience recognize and segment the facet joint area on lateral lumbar spine X-ray. Combined with the patient's lumbar CT image, the patient is graded according to the weishaupt standard [ 11 ]. The weishaupt classification of 0–1 is defined as the nosevere group, and the classification of 2–3 is defined as the severe group. Finally, after reviewing and discussing the circled results and grading results, the ROI area was determined as the area designated as lumbar facet joint in the training set. 2.4 YOLO-v8 network and training parameters This paper selects YOLO-v8 algorithm for image segmentation and processing to build a model (Fig. 2 shows the YOLO-v8 framework) The balance between performance and precision is considered in the design of YOLO-v8. During training, pre-trained weight models of various dimensions can be chosen as the initial model based on requirements. In this study, the initial model selected is YOLO-v8n, known for being the most 'lightweight' model, and its precision will be inferior to other weight models. Considering the amount of data we have, we can still choose YOLO-v8n, which can accept 640×640 pixels of image input. During training, the protocol involves setting the initial learning rate to 0.001 and using the SGD optimizer. Using image enhancement technology to enhance the adaptability of the model; It includes random scaling (ranging from 0.8x to 1.2x), rotation (between − 90°and + 90°), cropping, vertical and horizontal flipping, and mosaic. To align with the YOLO-v8n model's input requirements, the image is resized to 640×640 pixels before the input model is trained and assessed. The training also involves a batch size of 32 and 50 epochs. After 40 epochs, the training duration generally stabilized as precision stopped improving. These parameters are chosen based on YOLO-v8's default settings and are determined after evaluating the GPU environment's capacity for images and computing power, in order to achieve the best possible results and maintain a smooth training process. 2.5 Performance evaluation Performance was evaluated by calculating precision, recall and mean precision (map). The map is employed to evaluate how accurately the model detects facet joint areas and to assess its classification capabilities. Here, two thresholds are used to determine the 50% mean precision of evaluation (map 50) and the average precision (50%-95%) (map50-95). Precision measures the number of samples accurately identified as positive, while recall measures the number of samples correctly identified as positive. The map is applied to gauge the model's average detection precision across different intersection over union (IOU) thresholds, divided into map 50 (IOU threshold fixed at 0.5) and map 50–95 (the average value calculated in increments of 0.05 from IOU threshold 0.5 to 0.95). Compared with map 50, map 50–95 is a more stringent evaluation index. It can more comprehensively evaluate the positioning ability of the model, and is suitable for tasks requiring high-precision positioning. 2.6 Computer parameters The computing resources allocated by the virtual machine used in this research are completed by the 13th Gen Intel (R) core (TM) i9-13980hx (2.20 GHz) processor. The operating system used is windows 11 home edition. Utilizing the NVIDIA GEFORCE RTX 4060 laptop GPU, the accelerated computing environment for graphics processing implements deep learning processes and performance evaluations through PyTorch 2.0. 3 Results The imaging data and clinical data of 987 participants who received lumbar lateral and lumbar CT in Nantong first people's Hospital from May 2023 to June 2025 were collected. We finally determined the clinical data of a total of 987 patients. Figure 3 illustrates the process of including and excluding all participants. 3.1 Patient baseline data sheet On average, the patients were 61.96 ± 12.26 years old with a BMI of 24.96 ± 3.35 (kg/m 2 ) (Relevant parameters are shown in Table 1 ). There were 4133 segments with nosevere LFJOA and 802 segments with severe LFJOA. Table 1 Patient's baseline data sheet Variables Gender Male 488(49.4%) Female 499(50.6%) Age(years) 61.96 ± 12.26 BMI 24.96 ± 3.35 (BMI: body mass index) 3.2 Facet joint grading visualization A number of images were chosen at random from the dataset, and the training weights were applied for inference testing. Figure 4 displays example of test results. The blue box represents the ultimate convergence result of the generated prediction box. White text on a blue background indicates the category name. 3.3. Performance of the model The dataset consists of 987 X-rays, including 4935 facet joint regions. Figure 5 shows the Precision/Confidence (PC) curve and Precision/Recall (PR) curve of the model, which are used to help observe the ratio of correct recognition of lumbar facet joint area and the confidence level of classification prediction of the model. When the PR curve approaches the upper right corner, it means the model's prediction precision is high. In the PC curve analysis, the right end reflects the model's high-confidence predictions. Precision is nearly 1 when confidence reaches 0.67. In contrast, on the left side of the figure, precision decreases significantly when confidence falls below around 0.2. The evaluation of the model's performance on the test data set resulted in a precision of 0.71, a recall of 0.689, and an F1 score of 0.64. Figure 6 presents additional data, including frame loss and class loss. The train/box_loss curve demonstrates a decreasing pattern, suggesting that the model's boundary box prediction error reduces progressively during training. The metrics/map50 curve quickly ascends and then stabilizes, suggesting that the model is effective at object detection, particularly with an IOU threshold of 0.5. The metrics/map50-95 curve serves as a stricter evaluation metric, with its stability indicating the model's overall accuracy. Both the box_loss and DFL loss curves demonstrate a similar downward trend during training and eventually stabilize, which indicates that the training and verification performance of the model are consistent, and there is no obvious sign of over fitting. The model's map50 and map50-95 scores were 0.694 and 0.286, respectively. 4 Discussion With the advantages of single-stage design structure and multi-scale detection of YOLO-v8 model [ 19 ], this study successfully realized the accurate identification and grading evaluation of lumbar facet joints. The experimental data confirmed that yolo-v8 is expected to become an effective tool for large-scale clinical screening work, and solve the important deficiencies in the current technical scheme. At present, the research of artificial intelligence combined with osteoarthritis mainly focuses on the automatic detection and classification of knee osteoarthritis. The research of s Sheik Abdullah et al. Accurately identify and assess the severity of knee OA in digital X-ray images [ 20 ]. At present, there is no research on judging the severity of LFJOA combined with automatic detection and identification. Early evaluation of the extent of lumbar facet joint injuries aids in developing treatment plans for patients [ 21 ]. Therefore, YOLO-v8 is chosen as the detection model in this study. The model used in this study takes the efficient feature extraction network as the backbone, it realizes the efficient capture of image spatial features and context information, and significantly improves the detection precision. The model adopts the multi-level feature fusion strategy, which not only effectively improves the recognition rate of different size targets, but also greatly enhances the model’s stability and generalization. The multi task joint loss function can perform cooperative optimization on three core tasks: target position, classification and recognition, and confidence prediction. This improvement makes the adjustment of model parameters more accurate, so as to comprehensively improve the detection performance. It is particularly worth noting that the excellent small target detection performance provided by the single-level detection architecture adopted by YOLO-v8 provides reliable technical support for accurately identifying the lumbar facet joint region. A common inquiry is if the detection outcomes from the deep learning approach align perfectly with human judgment. Based on the evaluation, the map50 is noted to be 0.694, which can basically meet the requirements of image classification. The map50-95 value is only 0.286, suggesting that the model does not perform well with high IOU thresholds. Consequently, doctors still need to make the final judgment when using this auxiliary detection technology. There are several limitations to our study. First, the data set is limited to 987 patients in one institution, which may limit the applicability of our results; Future validation should include a larger multi-center queue. Second, although the map50 of the model is about 0.7, the precision is still not enough to completely replace the standard clinical diagnosis. It is emphasized that the model needs to be improved and integrated with other imaging modes. Multimodal data (such as clinical data and MRI images) should be considered to enhance the performance of the model. The study tackled these gaps, laying the groundwork for the integration of more efficient and precise automated detection systems into clinical workflows, and opened new avenues for diagnosing and treating LFJOA. In short, this study solved the key problem of automatic detection and classification of LFJOA by integrating large-scale medical image data sets and deep learning target detection technology. Based on the YOLO-v8 algorithm framework, the precise positioning and grading evaluation of lumbar facet joints are realized, which provides a theoretical basis for clinical auxiliary diagnosis. After the system is integrated into the existing diagnosis and treatment process, it can assist physicians to quickly identify the lesion areas that need to be focused on in lumbar lateral X-ray film, so as to improve the precision of early diagnosis of LFJOA. The subsequent research will aim to broaden the multi-center image data collection and investigate the clinical utility of the multimodal model. By improving these key links, this study will promote the development of LFJOA screening technology in a more efficient and accurate direction, and ultimately improve the diagnosis and treatment effect of patients. Declarations Author Contribution BSC, JMC and ZMC contributed to the study design, participated in the review process and prepared the manuscript. YYS and JMC contributed to collecting the relevant literature, generating figures and critical interpretation. TQW and GHX processed and analyzed data. BSC, ZMC and GHX conceived the paper and modified the manuscript. All authors read and approved the final manuscript. References Wang J, Yang Z, He X, Wang Y, Luo D, Xu W, Zhang H, & Zhou X (2024) DNM3OS/miR-127-5p/CDH11, activates Wnt3a/beta-catenin/LEF-1 pathway to form a positive feedback and aggravate spine facet joint osteoarthritis. Noncoding RNA Res 9(2), 294-306 Wang J, Lu Q, Mackay MJ, Liu X, Feng Y, Burton DC, & Asher MA (2022) Spontaneous Facet Joint Osteoarthritis in NFAT1-Mutant Mice: Age-Dependent Histopathologic Characteristics and Molecular Mechanisms. J Bone Joint Surg Am 104(10), 928-940 Pang H, Chen S, Klyne DM, Harrich D, Ding W, Yang S, & Han FY (2023) Low back pain and osteoarthritis pain: a perspective of estrogen. Bone Res 11(1), 42 Perolat R, Kastler A, Nicot B, Pellat JM, Tahon F, Attye A, Heck O, Boubagra K, Grand S, & Krainik A (2018) Facet joint syndrome: from diagnosis to interventional management. Insights Imaging 9(5), 773-789 Kalichman L, & Hunter DJ (2007) Lumbar facet joint osteoarthritis: a review. Semin Arthritis Rheum 37(2), 69-80 Tan L, Du X, Tang R, Rong L, & Zhang L (2024) Preoperative Adjacent Facet Joint Osteoarthritis Is Associated with the Incidence of Adjacent Segment Degeneration and Low Back Pain after Lumbar Interbody Fusion. Asian Spine J 18(1), 21-31 Jiang J, Zhang J, Wu C, Guo X, Chen C, Bao G, Sun Y, Chen J, Xue P, Xu G, & Cui Z (2018) Up-regulation of TRAF2 inhibits chondrocytes apoptosis in lumbar facet joint osteoarthritis. Biochem Biophys Res Commun 503(3), 1659-1665 Chen C, Bao GF, Xu G, Sun Y, & Cui ZM (2018) Altered Wnt and NF-kappaB Signaling in Facet Joint Osteoarthritis: Insights from RNA Deep Sequencing. Tohoku J Exp Med 245(1), 69-77 Lakadamyali H, Tarhan NC, Ergun T, Cakir B, & Agildere AM (2008) STIR sequence for depiction of degenerative changes in posterior stabilizing elements in patients with lower back pain. AJR Am J Roentgenol 191(4), 973-979 Kwee RM, & Kwee TC (2021) Imaging of facet joint diseases. Clin Imaging 80, 167-179 Weishaupt D, Zanetti M, Boos N, & Hodler J (1999) MR imaging and CT in osteoarthritis of the lumbar facet joints. Skeletal Radiol 28(4), 215-219 Wang A, Wang T, Zang L, Yuan S, Fan N, Du P, & Wu Q (2022) Quantitative Radiological Characteristics of the Facet Joints in Patients with Lumbar Foraminal Stenosis. J Pain Res 15, 2363-2371 Gillies RJ, Kinahan PE, & Hricak H (2016) Radiomics: Images Are More than Pictures, They Are Data. Radiology 278(2), 563-577 Cheng L, Cai F, Xu M, Liu P, Liao J, & Zong S (2023) A diagnostic approach integrated multimodal radiomics with machine learning models based on lumbar spine CT and X-ray for osteoporosis. J Bone Miner Metab 41(6), 877-889 Magneli M, Axenhus M, Fagrell J, Ling P, Gislen J, Demir Y, Domeij-Arverud E, Hallberg K, Salomonsson B, & Gordon M (2024) Artificial intelligence can be used in the identification and classification of shoulder osteoarthritis and avascular necrosis on plain radiographs: a training study of 7,139 radiograph sets. Acta Orthop 95, 319-324 Hidaka R, Matsuda K, Igari T, Takeuchi S, Imoto Y, Yagi S, & Kawano H (2024) Development and accuracy of an artificial intelligence model for predicting the progression of hip osteoarthritis using plain radiographs and clinical data: a retrospective study. BMC Musculoskelet Disord 25(1), 893 Hermens F (2024) Automatic object detection for behavioural research using YOLOv8. Behav Res Methods 56(7), 7307-7330 Shia WC, & Ku TH (2024) Enhancing Microcalcification Detection in Mammography with YOLO-v8 Performance and Clinical Implications. Diagnostics (Basel) 14(24) Mao M, & Hong M (2025) YOLO Object Detection for Real-Time Fabric Defect Inspection in the Textile Industry: A Review of YOLOv1 to YOLOv11. Sensors (Basel) 25(7) Abdullah SS, & Rajasekaran MP (2022) Automatic detection and classification of knee osteoarthritis using deep learning approach. Radiol Med 127(4), 398-406 Anaya JEC, Coelho SRN, Taneja AK, Cardoso FN, Skaf AY, & Aihara AY (2021) Differential Diagnosis of Facet Joint Disorders. Radiographics 41(2), 543-558 Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7540599","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":542312342,"identity":"35c71f49-2a73-45fe-b5f7-96980ebe07ba","order_by":0,"name":"baisen chen","email":"","orcid":"","institution":"Nantong City No 1 People's Hospital and Second Affiliated Hospital of Nantong University","correspondingAuthor":false,"prefix":"","firstName":"baisen","middleName":"","lastName":"chen","suffix":""},{"id":542312343,"identity":"2476cb2a-e968-45e2-b80a-c123bdeb8bcc","order_by":1,"name":"yuyu sun","email":"","orcid":"","institution":"Nantong Third People's Hospital","correspondingAuthor":false,"prefix":"","firstName":"yuyu","middleName":"","lastName":"sun","suffix":""},{"id":542312345,"identity":"f5afcabc-5c6c-4bf7-b173-4cc3c63c5594","order_by":2,"name":"jiaming cui","email":"","orcid":"","institution":"Shenzhen University","correspondingAuthor":false,"prefix":"","firstName":"jiaming","middleName":"","lastName":"cui","suffix":""},{"id":542312346,"identity":"124427dc-f80b-468b-bbb7-60491c9c5e0b","order_by":3,"name":"tianqi wu","email":"","orcid":"","institution":"Nantong University School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"tianqi","middleName":"","lastName":"wu","suffix":""},{"id":542312347,"identity":"b974f6d1-c5ed-4f02-86dc-6a610f5c92fc","order_by":4,"name":"guanhua xu","email":"","orcid":"","institution":"Nantong City No 1 People's Hospital and Second Affiliated Hospital of Nantong University","correspondingAuthor":false,"prefix":"","firstName":"guanhua","middleName":"","lastName":"xu","suffix":""},{"id":542312348,"identity":"9ef2a0cc-3fd1-4152-8d44-ed74f1a60075","order_by":5,"name":"zhiming cui","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA0ElEQVRIiWNgGAWjYNACAwYGfgiLmQQtkg3MJGkB6TpArBb+/tPJnwsK7sgZ3+4/JsFQYZ3YwH72AF4tEgfObpOeYfDM2OzOYTYJhjPpiQ08eQn43cPYu42Zx+Bw4rYbyWwSjG2HExskeAzwa2Hm3fwZqKV+8wyQln/EaGHj3SAN1JJgIAHS0kCEFokzvCC/HDaccSPZ2CLhWLpxG08Ofi38/Wc3fy74c1ief0biwxsfaqxl+9nP4NcCAoi4SABiNoLqGUiL8VEwCkbBKBiJAAAuAj9DrrI2bwAAAABJRU5ErkJggg==","orcid":"","institution":"Nantong City No 1 People's Hospital and Second Affiliated Hospital of Nantong University","correspondingAuthor":true,"prefix":"","firstName":"zhiming","middleName":"","lastName":"cui","suffix":""}],"badges":[],"createdAt":"2025-09-05 04:23:10","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7540599/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7540599/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":95807833,"identity":"177d21d8-d0af-4597-9f0b-93ae4bc2da74","added_by":"auto","created_at":"2025-11-13 08:49:11","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":3443045,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.docx","url":"https://assets-eu.researchsquare.com/files/rs-7540599/v1/a61d12865dd854baefb43830.docx"},{"id":95807805,"identity":"5b5c89b0-1829-47b4-9230-e0e53ee5b915","added_by":"auto","created_at":"2025-11-13 08:49:10","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":7389,"visible":true,"origin":"","legend":"","description":"","filename":"a3523198549e405087d4e3d251aaa41d.json","url":"https://assets-eu.researchsquare.com/files/rs-7540599/v1/a0e8e65852361a2f3679617e.json"},{"id":95807949,"identity":"8ed30482-191c-42b7-9401-e0940bb882df","added_by":"auto","created_at":"2025-11-13 08:49:15","extension":"xml","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":55099,"visible":true,"origin":"","legend":"","description":"","filename":"a3523198549e405087d4e3d251aaa41d1enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7540599/v1/bdd7ee75e8a7d92fefba43ea.xml"},{"id":95808471,"identity":"b1329a25-8fa0-4976-9102-797ff75f6b55","added_by":"auto","created_at":"2025-11-13 08:49:29","extension":"png","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":49901,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7540599/v1/140f5b0079c6c016d9174887.png"},{"id":95807851,"identity":"c69289de-2fcb-47b5-9105-e77727e25758","added_by":"auto","created_at":"2025-11-13 08:49:12","extension":"png","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":127404,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7540599/v1/354b4aae76fc1185c30d4212.png"},{"id":95808128,"identity":"7cefe502-021e-4399-a07c-a49a05a44856","added_by":"auto","created_at":"2025-11-13 08:49:20","extension":"png","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":30432,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7540599/v1/5b0ba5cb9a31fd5a61ba3d7f.png"},{"id":95807609,"identity":"33a23b5d-45d7-4fe2-b1d4-642044689eaa","added_by":"auto","created_at":"2025-11-13 08:49:01","extension":"png","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":2854991,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7540599/v1/a030093fdc855d73ecc8f4bc.png"},{"id":95808339,"identity":"89a9a161-8b9b-4a0f-8e48-3c6443a06791","added_by":"auto","created_at":"2025-11-13 08:49:25","extension":"png","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":83179,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-7540599/v1/c129d9b773a6c11d4836f35a.png"},{"id":95808221,"identity":"66ed9456-f993-44c6-b95c-f5ea7fb2037a","added_by":"auto","created_at":"2025-11-13 08:49:23","extension":"png","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":253058,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-7540599/v1/5f0e936d5072eab90047e950.png"},{"id":95807944,"identity":"41300a08-00dc-41a5-881c-f94f4d4f08e5","added_by":"auto","created_at":"2025-11-13 08:49:15","extension":"png","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":45172,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7540599/v1/b5855b2ac05ee79caba19fb9.png"},{"id":95807658,"identity":"a70e6c65-e601-4adf-b74a-91363ac6cd8a","added_by":"auto","created_at":"2025-11-13 08:49:03","extension":"png","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":28771,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7540599/v1/9746d4a09b480f7e0ee0650d.png"},{"id":95807832,"identity":"e3521e19-24e0-4168-b646-5c452b78b89e","added_by":"auto","created_at":"2025-11-13 08:49:11","extension":"png","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":29267,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7540599/v1/a6d69f1e97423ca33b210549.png"},{"id":95807837,"identity":"49890e2d-959a-4c18-9437-1f8fd94ec69f","added_by":"auto","created_at":"2025-11-13 08:49:11","extension":"png","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":351737,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7540599/v1/ef10fdc2f18ea6af5a3a5e6b.png"},{"id":95807947,"identity":"cd5197ab-5497-49a0-9d65-2a1df60f1503","added_by":"auto","created_at":"2025-11-13 08:49:15","extension":"png","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":25872,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-7540599/v1/843272b1ce3af592c4b8b8b0.png"},{"id":95807945,"identity":"026bc6e2-7e37-4d60-8045-2948a9da4687","added_by":"auto","created_at":"2025-11-13 08:49:15","extension":"png","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":92791,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-7540599/v1/645156d00452f27c6d5d0b89.png"},{"id":95807595,"identity":"e0fc7c5f-ea69-4a57-946f-026fb0d0dd20","added_by":"auto","created_at":"2025-11-13 08:48:59","extension":"xml","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":53043,"visible":true,"origin":"","legend":"","description":"","filename":"a3523198549e405087d4e3d251aaa41d1structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7540599/v1/d626bca9c423109f7b1756ad.xml"},{"id":95808054,"identity":"14f872ee-43a3-4e16-a652-3d0e8c00367a","added_by":"auto","created_at":"2025-11-13 08:49:18","extension":"html","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":59238,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7540599/v1/4c42744ddf0bb22847f495ae.html"},{"id":95808078,"identity":"8a13fffd-3356-4f2f-ae0b-fd286c2dc124","added_by":"auto","created_at":"2025-11-13 08:49:18","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":257329,"visible":true,"origin":"","legend":"\u003cp\u003eResearch flow chart\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7540599/v1/ab3535af839c3e24323af4f7.png"},{"id":95807779,"identity":"af35fa70-29ce-4583-ace0-037b2cfb1eec","added_by":"auto","created_at":"2025-11-13 08:49:09","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":179505,"visible":true,"origin":"","legend":"\u003cp\u003eNetwork structure diagram of YOLOv8 Seg\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7540599/v1/d9a25d2458127884704623c9.png"},{"id":95808283,"identity":"ac20bb5c-5f28-4d92-a897-bdda1df36e03","added_by":"auto","created_at":"2025-11-13 08:49:24","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":182612,"visible":true,"origin":"","legend":"\u003cp\u003ePatient screening chart\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7540599/v1/3930b3c1b77c15d2db601bed.png"},{"id":95808182,"identity":"a3a792f0-ae31-4181-aae9-741711575915","added_by":"auto","created_at":"2025-11-13 08:49:22","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":810445,"visible":true,"origin":"","legend":"\u003cp\u003eExample of test results\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7540599/v1/e7991275d9c711b101e757c5.png"},{"id":95807829,"identity":"690b02b5-1eec-4177-9d1c-823d6fa7cdf7","added_by":"auto","created_at":"2025-11-13 08:49:11","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":88853,"visible":true,"origin":"","legend":"\u003cp\u003e(a) precision–confidence curve and (b) precision–recall curve of model\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-7540599/v1/9ceb67b9f592b3d28a622b6d.png"},{"id":95807921,"identity":"4956428d-8413-47dc-8626-79c671e27599","added_by":"auto","created_at":"2025-11-13 08:49:15","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":436857,"visible":true,"origin":"","legend":"\u003cp\u003ePerformance evaluation curve of the model. It includes bounding box loss, object presence loss, classification loss, accuracy, recall rate, and map during training periods\u003c/p\u003e","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-7540599/v1/d0c3e6c0f3942357da9ea096.png"},{"id":100767048,"identity":"b0d3e2a9-af5c-4463-b222-3b054f5a4a1a","added_by":"auto","created_at":"2026-01-21 09:06:47","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2423616,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7540599/v1/a596afbc-7dce-49d0-a9bf-79b1fe32218c.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Application of YOLO-v8 model based on lumbar X-ray in grading diagnosis of lumbar facet joint of osteoarthritis","fulltext":[{"header":"1 Introduction","content":"\u003cp\u003eLumbar facet joint of osteoarthritis (LFJOA) also known as lumbar facet osteoarthritis, involves intervertebral joints located at the back of the spine. Its pathological characteristics are cartilage degeneration, subchondral bone sclerosis and reactive proliferation of joint edges [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. The condition is prevalent in middle-aged and senior populations, often leading to low back pain and movement difficulties [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. In recent years, the incidence of LFJOA has been on the rise, seriously reducing the daily life and work status of patients [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e], so the diagnosis and treatment of disease has been one of the research hotspots.\u003c/p\u003e\u003cp\u003eMagnetic resonance imaging (MRI) can image articular cartilage, synovium, joint cavity effusion and other soft tissues, but its sensitivity to show the edge of bone cortex is low, and often underestimates the severity of LFJOA [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. For diagnosing LFJOA, computed tomography (CT) is the most reliable imaging method, which can clearly show the changes of bone tissue and surrounding soft tissue [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. Weishaupt proposed a system for diagnosing and grading LFJOA in CT images [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. Indicators of the system encompass narrowing of the joint space, formation of osteophytes, hypertrophy of the articular process, erosion of subarticular bone, subchondral cysts, and the vacuum joint phenomenon [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. However, the classification of LFJOA needs to be judged by clinicians and radiologists with rich clinical experience. X-ray is a commonly used imaging examination method, which is used to evaluate the lumbar bone structure [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. Lumbar X-ray has the advantages of fast and convenient, small radiation, wide application range, etc. Therefore, this paper proposes a conjecture, whether to find an effective means to automatically identify the facet joint position and assess the severity in the lateral position of lumbar X-ray, in order to develop a treatment plan promptly and avert complications.\u003c/p\u003e\u003cp\u003eAt present, the application of artificial intelligence algorithm combined with imaging radiomics in osteoarthritis mainly focuses on shoulder joint, hip and knee joint and other large joints [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. Martin Magn\u0026eacute;li et al. can recognize and grade humeral head glenohumeral arthritis (GHOA) on X-ray by collecting 7139 shoulder X-ray to train the neural network model [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. Another study proposed a neural network model to predict the progress of hip OA and determine its precision based on hip X-ray plain film and patient data [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. These results indicate that the deep learning model has great potential in the use of X-ray to identify and assess the severity of inflammation.\u003c/p\u003e\u003cp\u003eYou only look once (YOLO)-v8 is the latest target detection model developed by Ultralytics, which provides model variants dedicated to detection, segmentation, key point estimation and other tasks [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. Currently, research has gathered breast X-rays from female participants, employed the YOLO-v8 target detection algorithm to identify breast microcalcifications, and examined its performance benefits and clinical applications [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. This shows that YOLO-v8 has great potential in identifying lesions on X-ray images. Therefore, we focus on facet arthritis disease. Utilizing the lateral X-ray of lumbar spine, we use the YOLO-v8 model to automatically identify and assess the severity of lumbar facet joints. Through the hierarchical diagnosis of disease by this model, doctors can customize personalized treatment plans for each patient. This means that the treatment method can be adjusted according to the classification of patients, so as to optimize the treatment effect and reduce unnecessary treatment and its related risks and costs.\u003c/p\u003e"},{"header":"2 Research methods","content":"\u003cp\u003e Approved by the ethics review committee of Nantong First People's Hospital (No. 2024KT411), this study did not require informed consent due to its retrospective design. As shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, the research process for this study involves collecting images, preprocessing them, training the model, and evaluating its performance. The following paragraphs will elaborate on these processes (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e flow chart).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003e2.1 Study cohort\u003c/h2\u003e\u003cp\u003eThis research was conducted retrospectively. The subjects were selected from patients treated in our hospital from May 2023 to June 2025. The inclusion criteria were as follows: (1) patients who had received lateral lumbar X-ray examination and lumbar CT in our hospital at the same time within one week; (2) The age of patients was 18\u0026ndash;90 years old; Exclusion criteria included: (1) the patient had previously received lumbar surgery; (2) The image is unclear or poor quality; (3) The region of interest (ROI) is difficult to cut.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e\u003ch2\u003e2.2 Data acquisition\u003c/h2\u003e\u003cp\u003eThe X-ray examination was performed by UDR 780i pro of Shanghai Lianying company. The CT examination utilized the Philips Ingenuity Core CT. Experienced spinal surgeons conducted the image acquisition to maintain process consistency. All obtained images are in the form of gray-scale digital imaging and JPG files with original resolution and color depth. In addition to the images, the corresponding clinical data, age, gender and body mineral index (BMI) of the patients were also obtained.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e\u003ch2\u003e2.3 Image processing\u003c/h2\u003e\u003cp\u003eIn the preprocessing stage, the image is enhanced by panning, zooming, and flipping the image. Two spine surgeons with clinical experience recognize and segment the facet joint area on lateral lumbar spine X-ray. Combined with the patient's lumbar CT image, the patient is graded according to the weishaupt standard [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. The weishaupt classification of 0\u0026ndash;1 is defined as the nosevere group, and the classification of 2\u0026ndash;3 is defined as the severe group. Finally, after reviewing and discussing the circled results and grading results, the ROI area was determined as the area designated as lumbar facet joint in the training set.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec6\" class=\"Section2\"\u003e\u003ch2\u003e2.4 YOLO-v8 network and training parameters\u003c/h2\u003e\u003cp\u003eThis paper selects YOLO-v8 algorithm for image segmentation and processing to build a model (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e shows the YOLO-v8 framework)\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eThe balance between performance and precision is considered in the design of YOLO-v8. During training, pre-trained weight models of various dimensions can be chosen as the initial model based on requirements. In this study, the initial model selected is YOLO-v8n, known for being the most 'lightweight' model, and its precision will be inferior to other weight models. Considering the amount of data we have, we can still choose YOLO-v8n, which can accept 640\u0026times;640 pixels of image input.\u003c/p\u003e\u003cp\u003eDuring training, the protocol involves setting the initial learning rate to 0.001 and using the SGD optimizer. Using image enhancement technology to enhance the adaptability of the model; It includes random scaling (ranging from 0.8x to 1.2x), rotation (between \u0026minus;\u0026thinsp;90\u0026deg;and +\u0026thinsp;90\u0026deg;), cropping, vertical and horizontal flipping, and mosaic. To align with the YOLO-v8n model's input requirements, the image is resized to 640\u0026times;640 pixels before the input model is trained and assessed. The training also involves a batch size of 32 and 50 epochs. After 40 epochs, the training duration generally stabilized as precision stopped improving. These parameters are chosen based on YOLO-v8's default settings and are determined after evaluating the GPU environment's capacity for images and computing power, in order to achieve the best possible results and maintain a smooth training process.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec7\" class=\"Section2\"\u003e\u003ch2\u003e2.5 Performance evaluation\u003c/h2\u003e\u003cp\u003ePerformance was evaluated by calculating precision, recall and mean precision (map). The map is employed to evaluate how accurately the model detects facet joint areas and to assess its classification capabilities. Here, two thresholds are used to determine the 50% mean precision of evaluation (map 50) and the average precision (50%-95%) (map50-95). Precision measures the number of samples accurately identified as positive, while recall measures the number of samples correctly identified as positive. The map is applied to gauge the model's average detection precision across different intersection over union (IOU) thresholds, divided into map 50 (IOU threshold fixed at 0.5) and map 50\u0026ndash;95 (the average value calculated in increments of 0.05 from IOU threshold 0.5 to 0.95). Compared with map 50, map 50\u0026ndash;95 is a more stringent evaluation index. It can more comprehensively evaluate the positioning ability of the model, and is suitable for tasks requiring high-precision positioning.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\u003ch2\u003e2.6 Computer parameters\u003c/h2\u003e\u003cp\u003eThe computing resources allocated by the virtual machine used in this research are completed by the 13th Gen Intel (R) core (TM) i9-13980hx (2.20 GHz) processor. The operating system used is windows 11 home edition. Utilizing the NVIDIA GEFORCE RTX 4060 laptop GPU, the accelerated computing environment for graphics processing implements deep learning processes and performance evaluations through PyTorch 2.0.\u003c/p\u003e\u003c/div\u003e"},{"header":"3 Results","content":"\u003cp\u003eThe imaging data and clinical data of 987 participants who received lumbar lateral and lumbar CT in Nantong first people's Hospital from May 2023 to June 2025 were collected. We finally determined the clinical data of a total of 987 patients. Figure\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e illustrates the process of including and excluding all participants.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e\u003ch2\u003e3.1 Patient baseline data sheet\u003c/h2\u003e\u003cp\u003eOn average, the patients were 61.96\u0026thinsp;\u0026plusmn;\u0026thinsp;12.26 years old with a BMI of 24.96\u0026thinsp;\u0026plusmn;\u0026thinsp;3.35 (kg/m\u003csup\u003e2\u003c/sup\u003e) (Relevant parameters are shown in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). There were 4133 segments with nosevere LFJOA and 802 segments with severe LFJOA.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003ePatient's baseline data sheet\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"2\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eVariables\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eGender\u003c/p\u003e\u003cp\u003eMale\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e488(49.4%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eFemale\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e499(50.6%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAge(years)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e61.96\u0026thinsp;\u0026plusmn;\u0026thinsp;12.26\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eBMI\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e24.96\u0026thinsp;\u0026plusmn;\u0026thinsp;3.35\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003ctfoot\u003e\u003ctr\u003e\u003ctd colspan=\"2\"\u003e(BMI: body mass index)\u003c/td\u003e\u003c/tr\u003e\u003c/tfoot\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\u003ch2\u003e3.2 Facet joint grading visualization\u003c/h2\u003e\u003cp\u003eA number of images were chosen at random from the dataset, and the training weights were applied for inference testing. Figure\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e displays example of test results. The blue box represents the ultimate convergence result of the generated prediction box. White text on a blue background indicates the category name.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\u003ch2\u003e3.3. Performance of the model\u003c/h2\u003e\u003cp\u003eThe dataset consists of 987 X-rays, including 4935 facet joint regions. Figure\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e shows the Precision/Confidence (PC) curve and Precision/Recall (PR) curve of the model, which are used to help observe the ratio of correct recognition of lumbar facet joint area and the confidence level of classification prediction of the model. When the PR curve approaches the upper right corner, it means the model's prediction precision is high. In the PC curve analysis, the right end reflects the model's high-confidence predictions. Precision is nearly 1 when confidence reaches 0.67. In contrast, on the left side of the figure, precision decreases significantly when confidence falls below around 0.2. The evaluation of the model's performance on the test data set resulted in a precision of 0.71, a recall of 0.689, and an F1 score of 0.64. Figure\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e presents additional data, including frame loss and class loss. The train/box_loss curve demonstrates a decreasing pattern, suggesting that the model's boundary box prediction error reduces progressively during training. The metrics/map50 curve quickly ascends and then stabilizes, suggesting that the model is effective at object detection, particularly with an IOU threshold of 0.5. The metrics/map50-95 curve serves as a stricter evaluation metric, with its stability indicating the model's overall accuracy. Both the box_loss and DFL loss curves demonstrate a similar downward trend during training and eventually stabilize, which indicates that the training and verification performance of the model are consistent, and there is no obvious sign of over fitting. The model's map50 and map50-95 scores were 0.694 and 0.286, respectively.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e"},{"header":"4 Discussion","content":"\u003cp\u003eWith the advantages of single-stage design structure and multi-scale detection of YOLO-v8 model [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e], this study successfully realized the accurate identification and grading evaluation of lumbar facet joints. The experimental data confirmed that yolo-v8 is expected to become an effective tool for large-scale clinical screening work, and solve the important deficiencies in the current technical scheme.\u003c/p\u003e\u003cp\u003eAt present, the research of artificial intelligence combined with osteoarthritis mainly focuses on the automatic detection and classification of knee osteoarthritis. The research of s Sheik Abdullah et al. Accurately identify and assess the severity of knee OA in digital X-ray images [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. At present, there is no research on judging the severity of LFJOA combined with automatic detection and identification. Early evaluation of the extent of lumbar facet joint injuries aids in developing treatment plans for patients [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. Therefore, YOLO-v8 is chosen as the detection model in this study. The model used in this study takes the efficient feature extraction network as the backbone, it realizes the efficient capture of image spatial features and context information, and significantly improves the detection precision. The model adopts the multi-level feature fusion strategy, which not only effectively improves the recognition rate of different size targets, but also greatly enhances the model\u0026rsquo;s stability and generalization. The multi task joint loss function can perform cooperative optimization on three core tasks: target position, classification and recognition, and confidence prediction. This improvement makes the adjustment of model parameters more accurate, so as to comprehensively improve the detection performance. It is particularly worth noting that the excellent small target detection performance provided by the single-level detection architecture adopted by YOLO-v8 provides reliable technical support for accurately identifying the lumbar facet joint region.\u003c/p\u003e\u003cp\u003eA common inquiry is if the detection outcomes from the deep learning approach align perfectly with human judgment. Based on the evaluation, the map50 is noted to be 0.694, which can basically meet the requirements of image classification. The map50-95 value is only 0.286, suggesting that the model does not perform well with high IOU thresholds. Consequently, doctors still need to make the final judgment when using this auxiliary detection technology.\u003c/p\u003e\u003cp\u003eThere are several limitations to our study. First, the data set is limited to 987 patients in one institution, which may limit the applicability of our results; Future validation should include a larger multi-center queue. Second, although the map50 of the model is about 0.7, the precision is still not enough to completely replace the standard clinical diagnosis. It is emphasized that the model needs to be improved and integrated with other imaging modes. Multimodal data (such as clinical data and MRI images) should be considered to enhance the performance of the model. The study tackled these gaps, laying the groundwork for the integration of more efficient and precise automated detection systems into clinical workflows, and opened new avenues for diagnosing and treating LFJOA.\u003c/p\u003e\u003cp\u003eIn short, this study solved the key problem of automatic detection and classification of LFJOA by integrating large-scale medical image data sets and deep learning target detection technology. Based on the YOLO-v8 algorithm framework, the precise positioning and grading evaluation of lumbar facet joints are realized, which provides a theoretical basis for clinical auxiliary diagnosis. After the system is integrated into the existing diagnosis and treatment process, it can assist physicians to quickly identify the lesion areas that need to be focused on in lumbar lateral X-ray film, so as to improve the precision of early diagnosis of LFJOA. The subsequent research will aim to broaden the multi-center image data collection and investigate the clinical utility of the multimodal model. By improving these key links, this study will promote the development of LFJOA screening technology in a more efficient and accurate direction, and ultimately improve the diagnosis and treatment effect of patients.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eBSC, JMC and ZMC contributed to the study design, participated in the review process and prepared the manuscript. YYS and JMC contributed to collecting the relevant literature, generating figures and critical interpretation. TQW and GHX processed and analyzed data. BSC, ZMC and GHX conceived the paper and modified the manuscript. All authors read and approved the final manuscript.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eWang J, Yang Z, He X, Wang Y, Luo D, Xu W, Zhang H, \u0026amp; Zhou X (2024) DNM3OS/miR-127-5p/CDH11, activates Wnt3a/beta-catenin/LEF-1 pathway to form a positive feedback and aggravate spine facet joint osteoarthritis. Noncoding RNA Res 9(2), 294-306\u003c/li\u003e\n\u003cli\u003eWang J, Lu Q, Mackay MJ, Liu X, Feng Y, Burton DC, \u0026amp; Asher MA (2022) Spontaneous Facet Joint Osteoarthritis in NFAT1-Mutant Mice: Age-Dependent Histopathologic Characteristics and Molecular Mechanisms. J Bone Joint Surg Am 104(10), 928-940\u003c/li\u003e\n\u003cli\u003ePang H, Chen S, Klyne DM, Harrich D, Ding W, Yang S, \u0026amp; Han FY (2023) Low back pain and osteoarthritis pain: a perspective of estrogen. Bone Res 11(1), 42\u003c/li\u003e\n\u003cli\u003ePerolat R, Kastler A, Nicot B, Pellat JM, Tahon F, Attye A, Heck O, Boubagra K, Grand S, \u0026amp; Krainik A (2018) Facet joint syndrome: from diagnosis to interventional management. Insights Imaging 9(5), 773-789\u003c/li\u003e\n\u003cli\u003eKalichman L, \u0026amp; Hunter DJ (2007) Lumbar facet joint osteoarthritis: a review. Semin Arthritis Rheum 37(2), 69-80\u003c/li\u003e\n\u003cli\u003eTan L, Du X, Tang R, Rong L, \u0026amp; Zhang L (2024) Preoperative Adjacent Facet Joint Osteoarthritis Is Associated with the Incidence of Adjacent Segment Degeneration and Low Back Pain after Lumbar Interbody Fusion. Asian Spine J 18(1), 21-31\u003c/li\u003e\n\u003cli\u003eJiang J, Zhang J, Wu C, Guo X, Chen C, Bao G, Sun Y, Chen J, Xue P, Xu G, \u0026amp; Cui Z (2018) Up-regulation of TRAF2 inhibits chondrocytes apoptosis in lumbar facet joint osteoarthritis. Biochem Biophys Res Commun 503(3), 1659-1665\u003c/li\u003e\n\u003cli\u003eChen C, Bao GF, Xu G, Sun Y, \u0026amp; Cui ZM (2018) Altered Wnt and NF-kappaB Signaling in Facet Joint Osteoarthritis: Insights from RNA Deep Sequencing. Tohoku J Exp Med 245(1), 69-77 \u003c/li\u003e\n\u003cli\u003eLakadamyali H, Tarhan NC, Ergun T, Cakir B, \u0026amp; Agildere AM (2008) STIR sequence for depiction of degenerative changes in posterior stabilizing elements in patients with lower back pain. AJR Am J Roentgenol 191(4), 973-979\u003c/li\u003e\n\u003cli\u003eKwee RM, \u0026amp; Kwee TC (2021) Imaging of facet joint diseases. Clin Imaging 80, 167-179\u003c/li\u003e\n\u003cli\u003eWeishaupt D, Zanetti M, Boos N, \u0026amp; Hodler J (1999) MR imaging and CT in osteoarthritis of the lumbar facet joints. Skeletal Radiol 28(4), 215-219\u003c/li\u003e\n\u003cli\u003eWang A, Wang T, Zang L, Yuan S, Fan N, Du P, \u0026amp; Wu Q (2022) Quantitative Radiological Characteristics of the Facet Joints in Patients with Lumbar Foraminal Stenosis. J Pain Res 15, 2363-2371\u003c/li\u003e\n\u003cli\u003eGillies RJ, Kinahan PE, \u0026amp; Hricak H (2016) Radiomics: Images Are More than Pictures, They Are Data. Radiology 278(2), 563-577\u003c/li\u003e\n\u003cli\u003eCheng L, Cai F, Xu M, Liu P, Liao J, \u0026amp; Zong S (2023) A diagnostic approach integrated multimodal radiomics with machine learning models based on lumbar spine CT and X-ray for osteoporosis. J Bone Miner Metab 41(6), 877-889\u003c/li\u003e\n\u003cli\u003eMagneli M, Axenhus M, Fagrell J, Ling P, Gislen J, Demir Y, Domeij-Arverud E, Hallberg K, Salomonsson B, \u0026amp; Gordon M (2024) Artificial intelligence can be used in the identification and classification of shoulder osteoarthritis and avascular necrosis on plain radiographs: a training study of 7,139 radiograph sets. Acta Orthop 95, 319-324\u003c/li\u003e\n\u003cli\u003eHidaka R, Matsuda K, Igari T, Takeuchi S, Imoto Y, Yagi S, \u0026amp; Kawano H (2024) Development and accuracy of an artificial intelligence model for predicting the progression of hip osteoarthritis using plain radiographs and clinical data: a retrospective study. BMC Musculoskelet Disord 25(1), 893\u003c/li\u003e\n\u003cli\u003eHermens F (2024) Automatic object detection for behavioural research using YOLOv8. Behav Res Methods 56(7), 7307-7330\u003c/li\u003e\n\u003cli\u003eShia WC, \u0026amp; Ku TH (2024) Enhancing Microcalcification Detection in Mammography with YOLO-v8 Performance and Clinical Implications. Diagnostics (Basel) 14(24)\u003c/li\u003e\n\u003cli\u003eMao M, \u0026amp; Hong M (2025) YOLO Object Detection for Real-Time Fabric Defect Inspection in the Textile Industry: A Review of YOLOv1 to YOLOv11. Sensors (Basel) 25(7)\u003c/li\u003e\n\u003cli\u003eAbdullah SS, \u0026amp; Rajasekaran MP (2022) Automatic detection and classification of knee osteoarthritis using deep learning approach. Radiol Med 127(4), 398-406\u003c/li\u003e\n\u003cli\u003eAnaya JEC, Coelho SRN, Taneja AK, Cardoso FN, Skaf AY, \u0026amp; Aihara AY (2021) Differential Diagnosis of Facet Joint Disorders. Radiographics 41(2), 543-558\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-7540599/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7540599/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003ePurpose Lumbar facet joint of osteoarthritis (LFJOA) can cause intractable low back pain in patients. Early evaluation of the status of LFJOA is very important for subsequent treatment. This paper discusses the automatic segmentation and detection of LFJOA by studying the characteristics of artificial intelligence technology and its potential application in medical image analysis.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eMethods The ability to detect inflammation has been significantly enhanced in recent years due to deep learning technology, especially models based on object detection. This study collected 987 lateral lumbar X-ray from 987 patients, each of which was manually divided into five lumbar facet joint segments. According to the computed tomography (CT) image of each patient, the classification annotation was carried out based on weishaupt standard. Then, the you only look once (YOLO)-v8 model was used for hierarchical diagnosis. Precision, recall, f1 score, mean average precision (map)50, and map50-95 were used to evaluate the model's performance. Additionally, the research examined how this technology could be applied in clinical settings.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eResults In detecting facet arthritis, the YOLO-v8 model reached a map50 of 0.694, a map50-95 of 0.286, an F1 score of 0.64, a precision rate of 0.71, and a recall of 0.689.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eConclusion YOLO-v8 has diagnostic value in detecting the severity of LFJOA. Future research should the model’s classification potential to enhance its clinical application settings, and help spinal surgeons more effectively diagnose the severity of lumbar facet arthritis, so as to formulate accurate treatment plans.\u003c/p\u003e","manuscriptTitle":"Application of YOLO-v8 model based on lumbar X-ray in grading diagnosis of lumbar facet joint of osteoarthritis","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-11-13 08:16:02","doi":"10.21203/rs.3.rs-7540599/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"9b0a9fce-f61e-4bd1-9166-078e253415d2","owner":[],"postedDate":"November 13th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-01-21T08:28:28+00:00","versionOfRecord":[],"versionCreatedAt":"2025-11-13 08:16:02","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7540599","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7540599","identity":"rs-7540599","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}