Yu Xi
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Yu Xi has a total of 37 co-authors. Here are the top co-authors by number of shared publications:
- Lei Zang
Beijing Chao-Yang Hospital, Beijing, China
9 shared publications - Ruiyuan Chen
Department of Orthopedics Beijing Chao‐Yang Hospital, Capital Medical University Beijing China
8 shared publications - Aobo Wang
Department of Orthopedics, Beijing Chaoyang Hospital, Capital Medical University, Beijing, China
7 shared publications - Tianyi Wang
Department of Orthopedics, Beijing Chaoyang Hospital, Capital Medical University, Beijing, China
7 shared publications - Ning Fan
Department of Orthopedics, Beijing Chaoyang Hospital, Capital Medical University, Beijing, China
6 shared publications - Shuo Yuan
Department of Orthopedics, Beijing Chaoyang Hospital, Capital Medical University, Beijing, China
6 shared publications
Latest Publications
- Open Access
- 5 October 2026
- Published by JMIR Publications Inc. in Journal of Medical Internet Research
- Vol. 28, e92008
Machine Learning and Deep Learning for the Diagnosis of Cervical Degenerative Diseases: Systematic Review and Meta-Analysis
Background: Cervical degenerative diseases are a global public health issue, and their incidence is rising worldwide. Although an increasing number of studies on traditional machine learning (TML) and deep learning (DL) have been conducted in the detection and segmentation of cervical degenerative diseases and have reported promising task-specific results, the performance of these models has not yet been systematically analyzed. Objective: This systematic review and meta-analysis aimed to summarize and evaluate existing evidence on TML and DL approaches for diagnosing cervical degenerative diseases, thereby comprehensively guiding future research and clinical applications. Methods: This systematic review was conducted in accordance with the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines. A comprehensive literature search was conducted on PubMed, Embase, the Cochrane Library, Web of Science, Scopus, and the Institute of Electrical and Electronics Engineers (IEEE Xplore) from January 2000 to June 2026, supplemented by backward and forward citation searching in Scopus. Studies evaluating TML and DL algorithms for diagnosing cervical degenerative diseases using medical imaging were included. Methodological quality was assessed using the Quality Assessment of Diagnostic Accuracy Studies 2 (QUADAS-2) tool and the Quality Assessment of Diagnostic Accuracy Studies AI (QUADAS-AI) tool. For the primary diagnostic accuracy meta-analysis, data were synthesized using a bivariate mixed-effects logistic regression model. Sensitivity and specificity were summarized separately using random-effects meta-analysis with the Knapp-Hartung adjustment, and 95% prediction intervals (PIs) were reported. Certainty of evidence was assessed using the GRADE (Grading of Recommendations Assessment, Development and Evaluation) approach. Results: This systematic review included 30 studies, of which 21 involved a total of 25,301 patients included in the meta-analysis. The pooled sensitivity and specificity were 0.92 (95% CI 0.89‐0.96; 95% PI 0.80‐1.00) and 0.88 (95% CI 0.84‐0.91; 95% PI 0.72‐1.00), respectively. The positive likelihood ratio (LR) was 8.36 (95% CI 6.14‐11.36), and the negative LR was 0.07 (95% CI 0.04‐0.11). The area under the summary receiver operating characteristic (SROC) curve was 0.96 (95% CI 0.94‐0.97). Leave-one-out analyses did not materially alter the pooled estimates. High risk of bias was identified in 4 studies using QUADAS-2 and in 17 using QUADAS-AI. The overall certainty of evidence was rated as low according to the GRADE approach. Conclusions: TML and DL models demonstrated satisfactory diagnostic performance for cervical degenerative diseases, although external validation was limited. Unlike previously published reviews in this field, this study provides pooled estimates of the diagnostic performance of TML and DL for cervical degenerative diseases and indicates that, given between-study heterogeneity and low certainty of evidence, AI should currently be used as clinical decision support rather than an independent replacement for physicians.
- 5 October 2026
- Published by JMIR Publications Inc. in Journal of Medical Internet Research
- Vol. 28, e92008
Open AccessMachine Learning and Deep Learning for the Diagnosis of Cervical Degenerative Diseases: Systematic Review and Meta-Analysis
Background: Cervical degenerative diseases are a global public health issue, and their incidence is rising worldwide. Although an increasing number of studies on traditional machine learning (TML) and deep learning (DL) have been conducted in the detection and segmentation of cervical degenerative diseases and have reported promising task-specific results, the performance of these models has not yet been systematically analyzed. Objective: This systematic review and meta-analysis aimed to summarize and evaluate existing evidence on TML and DL approaches for diagnosing cervical degenerative diseases, thereby comprehensively guiding future research and clinical applications. Methods: This systematic review was conducted in accordance with the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines. A comprehensive literature search was conducted on PubMed, Embase, the Cochrane Library, Web of Science, Scopus, and the Institute of Electrical and Electronics Engineers (IEEE Xplore) from January 2000 to June 2026, supplemented by backward and forward citation searching in Scopus. Studies evaluating TML and DL algorithms for diagnosing cervical degenerative diseases using medical imaging were included. Methodological quality was assessed using the Quality Assessment of Diagnostic Accuracy Studies 2 (QUADAS-2) tool and the Quality Assessment of Diagnostic Accuracy Studies AI (QUADAS-AI) tool. For the primary diagnostic accuracy meta-analysis, data were synthesized using a bivariate mixed-effects logistic regression model. Sensitivity and specificity were summarized separately using random-effects meta-analysis with the Knapp-Hartung adjustment, and 95% prediction intervals (PIs) were reported. Certainty of evidence was assessed using the GRADE (Grading of Recommendations Assessment, Development and Evaluation) approach. Results: This systematic review included 30 studies, of which 21 involved a total of 25,301 patients included in the meta-analysis. The pooled sensitivity and specificity were 0.92 (95% CI 0.89‐0.96; 95% PI 0.80‐1.00) and 0.88 (95% CI 0.84‐0.91; 95% PI 0.72‐1.00), respectively. The positive likelihood ratio (LR) was 8.36 (95% CI 6.14‐11.36), and the negative LR was 0.07 (95% CI 0.04‐0.11). The area under the summary receiver operating characteristic (SROC) curve was 0.96 (95% CI 0.94‐0.97). Leave-one-out analyses did not materially alter the pooled estimates. High risk of bias was identified in 4 studies using QUADAS-2 and in 17 using QUADAS-AI. The overall certainty of evidence was rated as low according to the GRADE approach. Conclusions: TML and DL models demonstrated satisfactory diagnostic performance for cervical degenerative diseases, although external validation was limited. Unlike previously published reviews in this field, this study provides pooled estimates of the diagnostic performance of TML and DL for cervical degenerative diseases and indicates that, given between-study heterogeneity and low certainty of evidence, AI should currently be used as clinical decision support rather than an independent replacement for physicians.
- Open Access
A Deep Learning–Based Multimodal Fusion Model for Predicting Bone Cement Leakage in Percutaneous Kyphoplasty: Development and Validation
Background: There is a lack of intelligent methodologies that effectively integrate multimodal information to predict bone cement leakage (BCL) during percutaneous kyphoplasty (PKP). This study aimed to develop and validate a deep learning (DL)–based multimodal fusion model that incorporates preoperative CT, MRI, and clinical variables to predict BCL subtypes during PKP. Methods: This study included a retrospective internal dataset for model training and validation, a prospective internal dataset, and an external dataset for independent testing. The fusion model incorporated preoperative spinal CT, MRI, and structured clinical baseline data within a two‐stage framework. The first stage consisted of target vertebra localization based on vertebral segmentation. The second stage comprised a classification module implemented using a multibranch 3D ResNet‐50 network. Performance was compared with image‐only models, single‐modality models, and spine surgeons using the area under the receiver operating characteristic curve (AUC), the area under the precision–recall curve, and other metrics. Results: The multimodal DL model achieved AUC values ranging from 0.795 to 0.861 in the internal test set and from 0.767 to 0.848 in the external test set for predicting BCL subtypes. Type III leakage demonstrated the highest predictive performance (internal AUC, 0.861; external AUC, 0.848). Overall, the fusion model achieved the highest AUC values and showed superior accuracy and agreement compared with spine surgeons, particularly for Type I ( κ : internal, 0.439 vs. 0.030–0.075; external, 0.298 vs. 0.019–0.066) and Type IV leakage ( κ : internal, 0.459 vs. 0.133–0.186; external, 0.545 vs. 0.002–0.095). Conclusion: A two‐stage multimodal fusion DL framework enables accurate, reliable, and promisingly generalizable prediction of BCL subtypes in PKP, outperforming spine surgeons and supporting individualized preoperative decision‐making.
Open AccessA Deep Learning–Based Multimodal Fusion Model for Predicting Bone Cement Leakage in Percutaneous Kyphoplasty: Development and Validation
Background: There is a lack of intelligent methodologies that effectively integrate multimodal information to predict bone cement leakage (BCL) during percutaneous kyphoplasty (PKP). This study aimed to develop and validate a deep learning (DL)–based multimodal fusion model that incorporates preoperative CT, MRI, and clinical variables to predict BCL subtypes during PKP. Methods: This study included a retrospective internal dataset for model training and validation, a prospective internal dataset, and an external dataset for independent testing. The fusion model incorporated preoperative spinal CT, MRI, and structured clinical baseline data within a two‐stage framework. The first stage consisted of target vertebra localization based on vertebral segmentation. The second stage comprised a classification module implemented using a multibranch 3D ResNet‐50 network. Performance was compared with image‐only models, single‐modality models, and spine surgeons using the area under the receiver operating characteristic curve (AUC), the area under the precision–recall curve, and other metrics. Results: The multimodal DL model achieved AUC values ranging from 0.795 to 0.861 in the internal test set and from 0.767 to 0.848 in the external test set for predicting BCL subtypes. Type III leakage demonstrated the highest predictive performance (internal AUC, 0.861; external AUC, 0.848). Overall, the fusion model achieved the highest AUC values and showed superior accuracy and agreement compared with spine surgeons, particularly for Type I ( κ : internal, 0.439 vs. 0.030–0.075; external, 0.298 vs. 0.019–0.066) and Type IV leakage ( κ : internal, 0.459 vs. 0.133–0.186; external, 0.545 vs. 0.002–0.095). Conclusion: A two‐stage multimodal fusion DL framework enables accurate, reliable, and promisingly generalizable prediction of BCL subtypes in PKP, outperforming spine surgeons and supporting individualized preoperative decision‐making.
- Open Access
- 10 July 2026
- Published by JMIR Publications Inc. in Journal of Medical Internet Research
- Vol. 28, e87816
Performance of DeepSeek-R1 and ChatGPT-5 in the Generation of North American Spine Society Clinical Guidelines for Adult Vertebral Compression Fractures: Comparative Study
Background: Vertebral compression fractures (VCFs) impose a substantial clinical and health care burden, and their management relies on timely access to evidence-based guidelines. Large language models (LLMs) may help clinicians rapidly obtain guideline-related information, but their performance on VCF guidelines remains unclear. Objective: This study aimed to evaluate the performance of LLMs, including DeepSeek-R1 and ChatGPT-5, in generating responses consistent with VCF clinical guidelines. Methods: Using the 2024 North American Spine Society VCF clinical guidelines as the reference standard, 34 open-ended and 87 closed-ended questions were submitted to DeepSeek-R1 and ChatGPT-5. Four senior spine surgeons independently rated responses to both closed-ended and open-ended questions using a 5-point Likert scale for accuracy, consistency, self-awareness, and fabrication/falsification. For open-ended questions, comprehensiveness, clarity, and trust and confidence were additionally assessed. Subgroup analyses were performed by question type, recommendation grade, and VCF subtype, with direct comparisons between models. Results: A total of 726 responses were generated for 121 questions. For closed-ended questions, ChatGPT-5 and DeepSeek-R1 showed comparable performance in accuracy ( P =.11), self-awareness ( P =.10), and fabrication/falsification ( P =.10). DeepSeek-R1 demonstrated better consistency than ChatGPT-5 for both closed-ended and open-ended questions ( P <.001 and P =.001, respectively). For open-ended questions, the models differed significantly in comprehensiveness ( P =.03) and trust and confidence ( P =.02), but not in accuracy ( P =.42), self-awareness ( P =.22), fabrication/falsification ( P =.64), or clarity ( P =.48). Closed-ended questions generally outperformed open-ended questions. Responses to grade A-C recommendations outperformed grade I recommendations in accuracy, consistency, and fabrication/falsification (all P ≤.001) but scored lower in self-awareness ( P <.001). No significant differences were observed across VCF subtypes. Conclusions: Under a standardized clinician-oriented prompting condition, ChatGPT-5 and DeepSeek-R1 showed generally high but variable scores across evaluation dimensions, with important deficiencies remaining, particularly in interventional and surgical treatment recommendations and in questions linked to recommendation grade I. Because these findings were obtained in a controlled prompting setting, caution is warranted when extrapolating them to other query styles, clinical scenarios, or LLMs.
- 10 July 2026
- Published by JMIR Publications Inc. in Journal of Medical Internet Research
- Vol. 28, e87816
Open AccessPerformance of DeepSeek-R1 and ChatGPT-5 in the Generation of North American Spine Society Clinical Guidelines for Adult Vertebral Compression Fractures: Comparative Study
Background: Vertebral compression fractures (VCFs) impose a substantial clinical and health care burden, and their management relies on timely access to evidence-based guidelines. Large language models (LLMs) may help clinicians rapidly obtain guideline-related information, but their performance on VCF guidelines remains unclear. Objective: This study aimed to evaluate the performance of LLMs, including DeepSeek-R1 and ChatGPT-5, in generating responses consistent with VCF clinical guidelines. Methods: Using the 2024 North American Spine Society VCF clinical guidelines as the reference standard, 34 open-ended and 87 closed-ended questions were submitted to DeepSeek-R1 and ChatGPT-5. Four senior spine surgeons independently rated responses to both closed-ended and open-ended questions using a 5-point Likert scale for accuracy, consistency, self-awareness, and fabrication/falsification. For open-ended questions, comprehensiveness, clarity, and trust and confidence were additionally assessed. Subgroup analyses were performed by question type, recommendation grade, and VCF subtype, with direct comparisons between models. Results: A total of 726 responses were generated for 121 questions. For closed-ended questions, ChatGPT-5 and DeepSeek-R1 showed comparable performance in accuracy ( P =.11), self-awareness ( P =.10), and fabrication/falsification ( P =.10). DeepSeek-R1 demonstrated better consistency than ChatGPT-5 for both closed-ended and open-ended questions ( P <.001 and P =.001, respectively). For open-ended questions, the models differed significantly in comprehensiveness ( P =.03) and trust and confidence ( P =.02), but not in accuracy ( P =.42), self-awareness ( P =.22), fabrication/falsification ( P =.64), or clarity ( P =.48). Closed-ended questions generally outperformed open-ended questions. Responses to grade A-C recommendations outperformed grade I recommendations in accuracy, consistency, and fabrication/falsification (all P ≤.001) but scored lower in self-awareness ( P <.001). No significant differences were observed across VCF subtypes. Conclusions: Under a standardized clinician-oriented prompting condition, ChatGPT-5 and DeepSeek-R1 showed generally high but variable scores across evaluation dimensions, with important deficiencies remaining, particularly in interventional and surgical treatment recommendations and in questions linked to recommendation grade I. Because these findings were obtained in a controlled prompting setting, caution is warranted when extrapolating them to other query styles, clinical scenarios, or LLMs.
- Open Access
- 3 July 2026
- Published by JMIR Publications Inc. in Journal of Medical Internet Research
- Vol. 28, e92315
Performance of DeepSeek V3.2 and ChatGPT 5.1 in Musculoskeletal Triage and Differential Diagnosis of Outpatients With Low Back Pain: Multidimensional Comparative Study
Background: Outpatients presenting with low back pain (LBP) often require efficient preconsultation triage and early differential diagnostic support. Large language models may assist these text-based tasks, but their performance under different clinical information conditions remains unclear. Objective: This study aimed to compare the performance of ChatGPT (5.1; OpenAI) and DeepSeek (V3.2; DeepSeek AI) in musculoskeletal disorders (MSDs) triage and the differential diagnosis of outpatients with LBP using real-world outpatient records under 2 simulated information conditions. Methods: This retrospective comparative study was conducted at a tertiary academic teaching hospital in Beijing. A total of 160 cases were included using a balanced design across 8 diagnostic categories (20 per category); 6 MSDs and 2 non-MSDs. Evaluation was performed in 2 phases: Phase 1 (chief complaint) and Phase 2 (structured questionnaire with 7 domains or 33 items), both executed in a zero-shot setting using standardized prompts. Outcomes included (1) triage accuracy, (2) preliminary diagnosis accuracy, and (3) differential diagnosis agreement. In Phase 2, 3 senior orthopedic evaluators additionally rated model rationales across 5 domains using a 5-point Likert scale. Results: For triage accuracy across all 160 cases, DeepSeek V3.2 improved from 84.4% to 90.6% (risk difference [RD] 6.2%, 95% CI –0.7% to 13.3%), and ChatGPT 5.1 improved from 75.6% to 93.1% (RD 17.5%, 95% CI 10.2%-24.9%). For preliminary diagnosis accuracy across the 120 musculoskeletal cases, DeepSeek V3.2 improved from 48.3% to 76.7% (RD 28.3%, 95% CI 16.8%-38.8%), whereas ChatGPT 5.1 improved from 35.0% to 87.5% (RD 52.5%, 95% CI 42.8%-60.6%). The mean number of correct differential diagnoses increased from 1.27 (SD 0.71) to 2.02 (SD 0.74) for DeepSeek V3.2 and from 1.34 (SD 0.70) to 2.03 (SD 0.77) for ChatGPT 5.1. In Phase 2, rationale ratings were generally good for both models, with ChatGPT 5.1 scoring higher in understanding and reasoning. Recognition of multiple myeloma (MM) remained limited, improving only from 45% to 55% (DeepSeek V3.2) and 55% to 60% (ChatGPT 5.1). Structured input reduced safety-risk errors in both models, but residual errors remained, especially for MM and metastatic spinal tumor. Conclusions: Both ChatGPT 5.1 and DeepSeek V3.2 demonstrated potential in text-based triage and differential diagnosis of MSDs for LBP, with structured clinical information generally improving performance, particularly for preliminary diagnosis accuracy and differential diagnosis agreement. However, their suboptimal sensitivity for red-flag conditions such as MM highlights significant safety concerns, indicating that they should not be used as stand-alone triage tools without clinician oversight. ChatGPT 5.1 showed stronger reasoning with structured inputs based on rationale ratings, whereas DeepSeek V3.2 showed better performance under chief-complaint-only input, with significantly higher Phase 1 preliminary diagnostic accuracy and numerically higher Phase 1 triage accuracy. These findings underscore the need for further model refinement, rigorous prospective validation, and integration with clinician oversight before clinical implementation.
- 3 July 2026
- Published by JMIR Publications Inc. in Journal of Medical Internet Research
- Vol. 28, e92315
Open AccessPerformance of DeepSeek V3.2 and ChatGPT 5.1 in Musculoskeletal Triage and Differential Diagnosis of Outpatients With Low Back Pain: Multidimensional Comparative Study
Background: Outpatients presenting with low back pain (LBP) often require efficient preconsultation triage and early differential diagnostic support. Large language models may assist these text-based tasks, but their performance under different clinical information conditions remains unclear. Objective: This study aimed to compare the performance of ChatGPT (5.1; OpenAI) and DeepSeek (V3.2; DeepSeek AI) in musculoskeletal disorders (MSDs) triage and the differential diagnosis of outpatients with LBP using real-world outpatient records under 2 simulated information conditions. Methods: This retrospective comparative study was conducted at a tertiary academic teaching hospital in Beijing. A total of 160 cases were included using a balanced design across 8 diagnostic categories (20 per category); 6 MSDs and 2 non-MSDs. Evaluation was performed in 2 phases: Phase 1 (chief complaint) and Phase 2 (structured questionnaire with 7 domains or 33 items), both executed in a zero-shot setting using standardized prompts. Outcomes included (1) triage accuracy, (2) preliminary diagnosis accuracy, and (3) differential diagnosis agreement. In Phase 2, 3 senior orthopedic evaluators additionally rated model rationales across 5 domains using a 5-point Likert scale. Results: For triage accuracy across all 160 cases, DeepSeek V3.2 improved from 84.4% to 90.6% (risk difference [RD] 6.2%, 95% CI –0.7% to 13.3%), and ChatGPT 5.1 improved from 75.6% to 93.1% (RD 17.5%, 95% CI 10.2%-24.9%). For preliminary diagnosis accuracy across the 120 musculoskeletal cases, DeepSeek V3.2 improved from 48.3% to 76.7% (RD 28.3%, 95% CI 16.8%-38.8%), whereas ChatGPT 5.1 improved from 35.0% to 87.5% (RD 52.5%, 95% CI 42.8%-60.6%). The mean number of correct differential diagnoses increased from 1.27 (SD 0.71) to 2.02 (SD 0.74) for DeepSeek V3.2 and from 1.34 (SD 0.70) to 2.03 (SD 0.77) for ChatGPT 5.1. In Phase 2, rationale ratings were generally good for both models, with ChatGPT 5.1 scoring higher in understanding and reasoning. Recognition of multiple myeloma (MM) remained limited, improving only from 45% to 55% (DeepSeek V3.2) and 55% to 60% (ChatGPT 5.1). Structured input reduced safety-risk errors in both models, but residual errors remained, especially for MM and metastatic spinal tumor. Conclusions: Both ChatGPT 5.1 and DeepSeek V3.2 demonstrated potential in text-based triage and differential diagnosis of MSDs for LBP, with structured clinical information generally improving performance, particularly for preliminary diagnosis accuracy and differential diagnosis agreement. However, their suboptimal sensitivity for red-flag conditions such as MM highlights significant safety concerns, indicating that they should not be used as stand-alone triage tools without clinician oversight. ChatGPT 5.1 showed stronger reasoning with structured inputs based on rationale ratings, whereas DeepSeek V3.2 showed better performance under chief-complaint-only input, with significantly higher Phase 1 preliminary diagnostic accuracy and numerically higher Phase 1 triage accuracy. These findings underscore the need for further model refinement, rigorous prospective validation, and integration with clinician oversight before clinical implementation.
- Open Access
- 30 April 2026
- Published by The Korean Spinal Neurosurgery Society in Neurospine
- Vol. 23 (2) , 393-403
Deep Learning-Assisted Lumbar Degeneration Evaluation Using Plain Radiographs: Development and Validation of a Novel Dual-Mechanism Architecture
Objective: To develop and externally validate a dual-mechanism deep learning (DL) model that integrates vertebral segmentation and lesion detection for automated evaluation of lumbar degeneration and structured report generation on plain radiographs.Methods: In this retrospective study, 5,964 patients who underwent standing anteroposterior and lateral lumbar radiographs at a single institution and 600 patients from a public dataset (BUU-Spine) were included. Vertebral corners from T11–L5 (and S1 on lateral views) and 7 degenerative findings (scoliosis, straightened/preserved lordosis, spondylolisthesis, disc space narrowing, osteophytes, vertebral compression, and abdominal aortic calcification) were annotated by 3 spine surgeons. Two independently trained, parallel networks were developed, including a ResNet-based segmentation network and a YOLOv8-based detection network. A rule-based integration strategy reconciled both outputs and generated structured diagnostic reports. Segmentation accuracy, quantitative measurement agreement, diagnostic performance, and clinical acceptability of reports were evaluated.Results: Intra- and interobserver landmark distances within 3 mm reached 96% and >95%, respectively. On the internal test set, the percentage of correct keypoints within 3 mm was 95.7%–98.6%, with intraclass correlation coefficients of 0.84–0.89 and Pearson correlation coefficient (r) of 0.90–0.94 for key radiographic parameters. The segmentation- and detection-based models achieved precision of 92.2%–96.9% and 91.7%–95.5%, and recall of 91.6%–94.8% and 93.3%–95.2%, respectively. Under the dual-positive condition, the integrated model yielded the highest precision (93.8%–97.3%), whereas the any-positive condition achieved the highest recall (94.1%–97.6%). Of 596 automatically generated structured reports, 557 (93.4%) were deemed clinically acceptable.Conclusion: The proposed dual-mechanism DL framework enables accurate, multilesion assessment of lumbar degeneration and generation of clinically acceptable structured reports from plain radiographs, supporting workflow optimization in lumbar spine imaging.
- 30 April 2026
- Published by The Korean Spinal Neurosurgery Society in Neurospine
- Vol. 23 (2) , 393-403
Open AccessDeep Learning-Assisted Lumbar Degeneration Evaluation Using Plain Radiographs: Development and Validation of a Novel Dual-Mechanism Architecture
Objective: To develop and externally validate a dual-mechanism deep learning (DL) model that integrates vertebral segmentation and lesion detection for automated evaluation of lumbar degeneration and structured report generation on plain radiographs.Methods: In this retrospective study, 5,964 patients who underwent standing anteroposterior and lateral lumbar radiographs at a single institution and 600 patients from a public dataset (BUU-Spine) were included. Vertebral corners from T11–L5 (and S1 on lateral views) and 7 degenerative findings (scoliosis, straightened/preserved lordosis, spondylolisthesis, disc space narrowing, osteophytes, vertebral compression, and abdominal aortic calcification) were annotated by 3 spine surgeons. Two independently trained, parallel networks were developed, including a ResNet-based segmentation network and a YOLOv8-based detection network. A rule-based integration strategy reconciled both outputs and generated structured diagnostic reports. Segmentation accuracy, quantitative measurement agreement, diagnostic performance, and clinical acceptability of reports were evaluated.Results: Intra- and interobserver landmark distances within 3 mm reached 96% and >95%, respectively. On the internal test set, the percentage of correct keypoints within 3 mm was 95.7%–98.6%, with intraclass correlation coefficients of 0.84–0.89 and Pearson correlation coefficient (r) of 0.90–0.94 for key radiographic parameters. The segmentation- and detection-based models achieved precision of 92.2%–96.9% and 91.7%–95.5%, and recall of 91.6%–94.8% and 93.3%–95.2%, respectively. Under the dual-positive condition, the integrated model yielded the highest precision (93.8%–97.3%), whereas the any-positive condition achieved the highest recall (94.1%–97.6%). Of 596 automatically generated structured reports, 557 (93.4%) were deemed clinically acceptable.Conclusion: The proposed dual-mechanism DL framework enables accurate, multilesion assessment of lumbar degeneration and generation of clinically acceptable structured reports from plain radiographs, supporting workflow optimization in lumbar spine imaging.
- Open Access
- 11 March 2026
- Published by Wiley in Orthopaedic Surgery
- Vol. 18 (4) , 824-834
Efficacy of Percutaneous Kyphoplasty in the Treatment of Kümmell's Disease and Osteoporotic Vertebral Compression Fractures: A Clinical Comparative Study
Objective: Kümmell’s disease (KD) represents a delayed form of osteoporotic vertebral collapse and shares clinical features with osteoporotic vertebral compression fractures (OVCF). Percutaneous kyphoplasty (PKP) is commonly performed for both conditions, yet comparative evidence and predictors of 1‐year outcomes in KD remain limited. This study aimed to evaluate the efficacy and safety of PKP for the treatment of stage I and II KD and to identify factors associated with the outcomes at 1‐year follow‐up. Methods: We included 387 inpatients with KD or OVCF who underwent PKP from January 2016 to December 2022. All patients were assigned to the KD group ( n = 107) and the OVCF group ( n = 280). The difference of demographic data (age, gender, surgical segment, osteoporosis severity and disease duration), clinical efficacy (visual analog scale and Oswestry disability index of pre‐operation, 3‐day post‐operation, 3‐month post‐operation, and 1‐year post‐operation), complications (bone cement leakage during surgery and postoperative refractures), and radiographic parameters (anterior vertebral height and kyphotic angle of pre‐operation, post‐operation, and 1‐year follow‐up) was analyzed. Intergroup comparisons of continuous variables were performed using the Student's t‐test. Repeated measures ANOVA with Bonferroni post hoc correction was used to evaluate intra‐group differences of visual analog scale (VAS), Oswestry disability index (ODI), anterior vertebral height (AVH) and kyphotic angle (KA) across different time points. Multivariate logistic regression analysis was employed to identify the independent factors influencing VAS and ODI scores during the follow‐up period. Results: The disease duration of the KD group was much longer than that of the OVCF group. Significant improvements in VAS and ODI were observed at three‐day, three‐month, and one‐year after PKP. Multivariate regression analysis identified the blocky cement distribution pattern, higher preoperative VAS, and higher preoperative ODI as independent risk factors for suboptimal recovery during follow‐up. Besides, the KD group had lower AVH and larger KA than the OVCF group preoperatively. Both groups showed significant improvements in AVH and KA after PKP. However, the KD group had a higher rate of type II bone cement leakage (BCL) and more severe cemented vertebral collapse at the final follow‐up. The mean bone cement volume was significantly greater in the KD group. Refracture rates were similar between the two groups during follow‐up. Conclusions: PKP can effectively alleviate back pain, improve functional impairment, and correct local deformity in KD patients, with the risks of BCL and vertebral collapse or refractures during follow‐up. It is suitable for KD treatment without nerve injury symptoms. Appropriate measures should be taken to reduce the risk of complications.
- 11 March 2026
- Published by Wiley in Orthopaedic Surgery
- Vol. 18 (4) , 824-834
Open AccessEfficacy of Percutaneous Kyphoplasty in the Treatment of Kümmell's Disease and Osteoporotic Vertebral Compression Fractures: A Clinical Comparative Study
Objective: Kümmell’s disease (KD) represents a delayed form of osteoporotic vertebral collapse and shares clinical features with osteoporotic vertebral compression fractures (OVCF). Percutaneous kyphoplasty (PKP) is commonly performed for both conditions, yet comparative evidence and predictors of 1‐year outcomes in KD remain limited. This study aimed to evaluate the efficacy and safety of PKP for the treatment of stage I and II KD and to identify factors associated with the outcomes at 1‐year follow‐up. Methods: We included 387 inpatients with KD or OVCF who underwent PKP from January 2016 to December 2022. All patients were assigned to the KD group ( n = 107) and the OVCF group ( n = 280). The difference of demographic data (age, gender, surgical segment, osteoporosis severity and disease duration), clinical efficacy (visual analog scale and Oswestry disability index of pre‐operation, 3‐day post‐operation, 3‐month post‐operation, and 1‐year post‐operation), complications (bone cement leakage during surgery and postoperative refractures), and radiographic parameters (anterior vertebral height and kyphotic angle of pre‐operation, post‐operation, and 1‐year follow‐up) was analyzed. Intergroup comparisons of continuous variables were performed using the Student's t‐test. Repeated measures ANOVA with Bonferroni post hoc correction was used to evaluate intra‐group differences of visual analog scale (VAS), Oswestry disability index (ODI), anterior vertebral height (AVH) and kyphotic angle (KA) across different time points. Multivariate logistic regression analysis was employed to identify the independent factors influencing VAS and ODI scores during the follow‐up period. Results: The disease duration of the KD group was much longer than that of the OVCF group. Significant improvements in VAS and ODI were observed at three‐day, three‐month, and one‐year after PKP. Multivariate regression analysis identified the blocky cement distribution pattern, higher preoperative VAS, and higher preoperative ODI as independent risk factors for suboptimal recovery during follow‐up. Besides, the KD group had lower AVH and larger KA than the OVCF group preoperatively. Both groups showed significant improvements in AVH and KA after PKP. However, the KD group had a higher rate of type II bone cement leakage (BCL) and more severe cemented vertebral collapse at the final follow‐up. The mean bone cement volume was significantly greater in the KD group. Refracture rates were similar between the two groups during follow‐up. Conclusions: PKP can effectively alleviate back pain, improve functional impairment, and correct local deformity in KD patients, with the risks of BCL and vertebral collapse or refractures during follow‐up. It is suitable for KD treatment without nerve injury symptoms. Appropriate measures should be taken to reduce the risk of complications.
- Open Access
- 24 November 2025
- Published by JMIR Publications Inc. in Journal of Medical Internet Research
- Vol. 27, e77155
Deep Learning–Assisted Automated Diagnosis of Osteoporosis Based on Computed Tomography Scans: Systematic Review and Meta-Analysis
Background: Osteoporosis is a prevalent skeletal disorder characterized by decreased bone mass and increased fracture risk; however, it frequently remains underdiagnosed due to limited health care resources and its asymptomatic progression. Deep learning (DL) provides a promising solution for automated screening using computed tomography (CT) scans, enabling earlier detection and improved management. Objective: This systematic review and meta-analysis aimed to investigate the diagnostic performance of DL models in diagnosing osteoporosis based on CT scans. Methods: This study was conducted under the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines using articles extracted from PubMed, Scopus, Web of Science (Core), and Embase (Ovid). Studies involving adult participants who underwent CT and in which DL was applied for osteoporosis diagnosis were included. The QUADAS-2 (Quality Assessment of Diagnostic Accuracy Studies-2) tool was used to estimate the risk of bias in each study. The confusion matrices from the included studies were extracted to summarize the diagnostic performance of DL models for osteoporosis. Within a bivariate random-effects framework, sensitivity and specificity were jointly synthesized to yield the summary estimates. Heterogeneity was quantified with Higgins I² statistics. Subgroup analyses were performed to explore potential sources of heterogeneity among the included studies. Results: This review included 24 studies, encompassing CT images from 29,808 participants. All studies used conventional CT scans and used DL-based architectures. Fifteen, 6, and 3 studies were assessed as having a low, uncertain, and high risk of bias, respectively. The meta-analysis included 20 studies. The pooled sensitivity and specificity were 0.88 (95% CI 0.85‐0.91; I 2 =83.69%) and 0.94 (95% CI 0.91‐0.96; I 2 =95.07%) for osteoporosis diagnosis; 0.81 (95% CI 0.76‐0.85; I 2 =82.38%) and 0.92 (95% CI 0.90‐0.94; I 2 =79.05%) for osteopenia identification; and 0.95 (95% CI 0.92‐0.97; I 2 =98.28%) and 0.93 (95% CI 0.91‐0.95; I 2 =94.93%) for normal case identification. The area under the curve of the DL models for identifying osteoporosis, osteopenia, and normal cases was 0.96 (95% CI 0.93‐0.97), 0.94 (95% CI 0.92‐0.96), and 0.98 (95% CI 0.96‐0.99), respectively. Subgroup analyses revealed that models based on DenseNet variants ( P <.01), multislice input ( P <.01), 3D architecture ( P <.01), and CT as the reference standard ( P <.01) demonstrated superior diagnostic performance. Conclusions: This study indicated that CT-based DL models achieve promising diagnostic performance for osteoporosis. However, substantial heterogeneity among the included studies, limited external validation, and incomplete end-to-end pipelines constrain the generalizability of the proposed models. Further research is warranted to support their clinical translation and standardized application.
- 24 November 2025
- Published by JMIR Publications Inc. in Journal of Medical Internet Research
- Vol. 27, e77155
Open AccessDeep Learning–Assisted Automated Diagnosis of Osteoporosis Based on Computed Tomography Scans: Systematic Review and Meta-Analysis
Background: Osteoporosis is a prevalent skeletal disorder characterized by decreased bone mass and increased fracture risk; however, it frequently remains underdiagnosed due to limited health care resources and its asymptomatic progression. Deep learning (DL) provides a promising solution for automated screening using computed tomography (CT) scans, enabling earlier detection and improved management. Objective: This systematic review and meta-analysis aimed to investigate the diagnostic performance of DL models in diagnosing osteoporosis based on CT scans. Methods: This study was conducted under the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines using articles extracted from PubMed, Scopus, Web of Science (Core), and Embase (Ovid). Studies involving adult participants who underwent CT and in which DL was applied for osteoporosis diagnosis were included. The QUADAS-2 (Quality Assessment of Diagnostic Accuracy Studies-2) tool was used to estimate the risk of bias in each study. The confusion matrices from the included studies were extracted to summarize the diagnostic performance of DL models for osteoporosis. Within a bivariate random-effects framework, sensitivity and specificity were jointly synthesized to yield the summary estimates. Heterogeneity was quantified with Higgins I² statistics. Subgroup analyses were performed to explore potential sources of heterogeneity among the included studies. Results: This review included 24 studies, encompassing CT images from 29,808 participants. All studies used conventional CT scans and used DL-based architectures. Fifteen, 6, and 3 studies were assessed as having a low, uncertain, and high risk of bias, respectively. The meta-analysis included 20 studies. The pooled sensitivity and specificity were 0.88 (95% CI 0.85‐0.91; I 2 =83.69%) and 0.94 (95% CI 0.91‐0.96; I 2 =95.07%) for osteoporosis diagnosis; 0.81 (95% CI 0.76‐0.85; I 2 =82.38%) and 0.92 (95% CI 0.90‐0.94; I 2 =79.05%) for osteopenia identification; and 0.95 (95% CI 0.92‐0.97; I 2 =98.28%) and 0.93 (95% CI 0.91‐0.95; I 2 =94.93%) for normal case identification. The area under the curve of the DL models for identifying osteoporosis, osteopenia, and normal cases was 0.96 (95% CI 0.93‐0.97), 0.94 (95% CI 0.92‐0.96), and 0.98 (95% CI 0.96‐0.99), respectively. Subgroup analyses revealed that models based on DenseNet variants ( P <.01), multislice input ( P <.01), 3D architecture ( P <.01), and CT as the reference standard ( P <.01) demonstrated superior diagnostic performance. Conclusions: This study indicated that CT-based DL models achieve promising diagnostic performance for osteoporosis. However, substantial heterogeneity among the included studies, limited external validation, and incomplete end-to-end pipelines constrain the generalizability of the proposed models. Further research is warranted to support their clinical translation and standardized application.
- Open Access
- 29 August 2025
- Published by SAGE Publications in Global Spine Journal
- Vol. 16 (2) , 1075-1084
Comparison of the Predictive Roles of CT- and MRI-Based Pedicle Regional Osteoporosis Status Measurements for Pedicle Screw Loosening After Posterior Lumbar Interbody Fusion
Study Design: Retrospective Study. Objectives: To compare the efficacy of pedicle Hounsfield unit (HU) values and pedicle bone quality (PBQ) scores in predicting pedicle screw loosening (PSL) after posterior lumbar interbody fusion (PLIF) in elderly patients and to identify the most discriminative bone mineral density (BMD) assessment indicator. Methods: This retrospective analysis included 275 consecutive patients who underwent PLIF. L 1 and pedicle HU values were obtained from CT scans, whereas vertebral bone quality (VBQ) and PBQ scores were obtained from MRI. Logistic regression analysis determined factors associated with PSL. Receiver-operating characteristic curve analysis was conducted to assess the predictive value of pedicle HU and PBQ score for PSL and to additionally compare their predictive value with that of L 1 HU value and VBQ score. Results: The PSL rate was 20.36% (56/275). The loosening group demonstrated a higher postoperative low-back pain visual analog scale score ( P < 0.05), lower L 1 ( P < 0.001) and pedicle HU values ( P < 0.001) and higher VBQ ( P < 0.001) and PBQ scores ( P < 0.001) than the non-loosening group. The logistic regression analysis identified L 1 (OR = 0.98, 95% CI = 0.97-0.99, P < 0.001) and pedicle HU values (OR = 0.99, 95% CI = 0.98-0.99, P < 0.001) and VBQ (OR = 2.81, 95% CI = 1.43-5.52, P = 0.003) and PBQ scores (OR = 3.85, 95% CI = 2.03-7.32, P < 0.001) as independent predictors of PSL. The area under the curve for L 1 and pedicle HU values and VBQ and PBQ scores were 0.751, 0.766, 0.684, and 0.702, respectively. The optimal pedicle HU cut-off for predicting PSL was 106.32 (sensitivity: 78.49%; specificity: 75.00%). Conclusions: Pedicle HU values exhibited a relatively higher predictive performance for PSL compared with the PBQ score and served as the most discriminative BMD indicator in patients who underwent PLIF. Measuring the pedicle HU value preoperatively help surgeons to select a more appropriate surgical plan and is expected to improve the patient outcomes.
- 29 August 2025
- Published by SAGE Publications in Global Spine Journal
- Vol. 16 (2) , 1075-1084
Open AccessComparison of the Predictive Roles of CT- and MRI-Based Pedicle Regional Osteoporosis Status Measurements for Pedicle Screw Loosening After Posterior Lumbar Interbody Fusion
Study Design: Retrospective Study. Objectives: To compare the efficacy of pedicle Hounsfield unit (HU) values and pedicle bone quality (PBQ) scores in predicting pedicle screw loosening (PSL) after posterior lumbar interbody fusion (PLIF) in elderly patients and to identify the most discriminative bone mineral density (BMD) assessment indicator. Methods: This retrospective analysis included 275 consecutive patients who underwent PLIF. L 1 and pedicle HU values were obtained from CT scans, whereas vertebral bone quality (VBQ) and PBQ scores were obtained from MRI. Logistic regression analysis determined factors associated with PSL. Receiver-operating characteristic curve analysis was conducted to assess the predictive value of pedicle HU and PBQ score for PSL and to additionally compare their predictive value with that of L 1 HU value and VBQ score. Results: The PSL rate was 20.36% (56/275). The loosening group demonstrated a higher postoperative low-back pain visual analog scale score ( P < 0.05), lower L 1 ( P < 0.001) and pedicle HU values ( P < 0.001) and higher VBQ ( P < 0.001) and PBQ scores ( P < 0.001) than the non-loosening group. The logistic regression analysis identified L 1 (OR = 0.98, 95% CI = 0.97-0.99, P < 0.001) and pedicle HU values (OR = 0.99, 95% CI = 0.98-0.99, P < 0.001) and VBQ (OR = 2.81, 95% CI = 1.43-5.52, P = 0.003) and PBQ scores (OR = 3.85, 95% CI = 2.03-7.32, P < 0.001) as independent predictors of PSL. The area under the curve for L 1 and pedicle HU values and VBQ and PBQ scores were 0.751, 0.766, 0.684, and 0.702, respectively. The optimal pedicle HU cut-off for predicting PSL was 106.32 (sensitivity: 78.49%; specificity: 75.00%). Conclusions: Pedicle HU values exhibited a relatively higher predictive performance for PSL compared with the PBQ score and served as the most discriminative BMD indicator in patients who underwent PLIF. Measuring the pedicle HU value preoperatively help surgeons to select a more appropriate surgical plan and is expected to improve the patient outcomes.
- Open Access
- 31 July 2025
- Published by JMIR Publications Inc. in Journal of Medical Internet Research
- Vol. 27 (1) , e75567
Evaluating the Performance of State-of-the-Art Artificial Intelligence Chatbots Based on the WHO Global Guidelines for the Prevention of Surgical Site Infection: Cross-Sectional Study
Background: Surgical site infection (SSI) is the most prevalent type of health care–associated infection that leads to increased morbidity and mortality and a significant economic burden. Effective prevention of SSI relies on surgeons strictly following the latest clinical guidelines and implementing standardized and multilevel intervention strategies. However, the frequent updates to clinical guidelines render the processes of acquisition and interpretation quite time-consuming and intricate. The emergence of artificial intelligence (AI) chatbots offers both possibilities and challenges to address these issues in the surgical field. Objective: This study aimed to test the multidimensional capability of state-of-the-art AI chatbots for generating proper recommendations and corresponding rationales concordant with the global guideline for the prevention of SSI. Methods: Referred by other authoritative guidelines, recommendations and corresponding rationales from the 2018 World Health Organization global guidelines were refined and selected as benchmarks. Then, they were rephrased into a combined format of closed-ended queries for recommendations and open-ended queries for corresponding rationales, whereafter input into ChatGPT-4o (OpenAI), OpenAI-o1 (OpenAI), Claude 3.5 Sonnet (Anthropic), and Gemini 1.5 Pro (Google) 3 times. All responses were individually evaluated in 10 evaluation metrics based on the QUEST dimensions by 4 multidisciplinary senior surgeons using a 5-point Likert scale. The multidimensional performances among chatbots were compared, and the interrater agreements were calculated. Results: A total of 300 responses to 25 queries were generated by the 4 chatbots. The interrater agreements of the evaluators ranged from moderate to good (0.54‐0.87). In response to recommendations, the average accuracy, consistency, and harm scores for all chatbots were 4.03 (SD 1.09), 4.07 (SD 0.88), and 4.29 (SD 1.01), respectively. In responses for rationales, 4 subdimensions, including harm (mean 4.22, SD 0.97), relevance (mean 4.15, SD 0.83), fabrication and falsification (mean 4.12, SD 1.02), and understanding and reasoning (mean 4.04, SD 0.92) averagely scored ≥4. In contrast, consistency (mean 3.94, SD 0.72), clarity (mean 3.94, SD 0.89), comprehensiveness (mean 3.85, SD 0.83), and accuracy (mean 3.74, SD 0.91) performed at a moderate level. For the whole responses, the average self-awareness and trust and confidence scores for all chatbots were 3.84 (SD 0.89) and 3.88 (SD 0.91), respectively. Based on the average scores of the subdimensions, Claude 3.5 Sonnet and ChatGPT-4o were the top 2 outperformed models. Conclusions: The performance of AI chatbots in providing responses regarding well-established global guidelines in the prevention of SSI was acceptable, demonstrating immense potential in clinical applications. Nonetheless, a critical issue is the necessity of enhancing the stability of chatbots, as inaccurate responses can lead to severe consequences for SSI. Despite its limitations, it is anticipated that AI will trigger far-reaching changes in how clinicians access and use medical information.
- 31 July 2025
- Published by JMIR Publications Inc. in Journal of Medical Internet Research
- Vol. 27 (1) , e75567
Open AccessEvaluating the Performance of State-of-the-Art Artificial Intelligence Chatbots Based on the WHO Global Guidelines for the Prevention of Surgical Site Infection: Cross-Sectional Study
Background: Surgical site infection (SSI) is the most prevalent type of health care–associated infection that leads to increased morbidity and mortality and a significant economic burden. Effective prevention of SSI relies on surgeons strictly following the latest clinical guidelines and implementing standardized and multilevel intervention strategies. However, the frequent updates to clinical guidelines render the processes of acquisition and interpretation quite time-consuming and intricate. The emergence of artificial intelligence (AI) chatbots offers both possibilities and challenges to address these issues in the surgical field. Objective: This study aimed to test the multidimensional capability of state-of-the-art AI chatbots for generating proper recommendations and corresponding rationales concordant with the global guideline for the prevention of SSI. Methods: Referred by other authoritative guidelines, recommendations and corresponding rationales from the 2018 World Health Organization global guidelines were refined and selected as benchmarks. Then, they were rephrased into a combined format of closed-ended queries for recommendations and open-ended queries for corresponding rationales, whereafter input into ChatGPT-4o (OpenAI), OpenAI-o1 (OpenAI), Claude 3.5 Sonnet (Anthropic), and Gemini 1.5 Pro (Google) 3 times. All responses were individually evaluated in 10 evaluation metrics based on the QUEST dimensions by 4 multidisciplinary senior surgeons using a 5-point Likert scale. The multidimensional performances among chatbots were compared, and the interrater agreements were calculated. Results: A total of 300 responses to 25 queries were generated by the 4 chatbots. The interrater agreements of the evaluators ranged from moderate to good (0.54‐0.87). In response to recommendations, the average accuracy, consistency, and harm scores for all chatbots were 4.03 (SD 1.09), 4.07 (SD 0.88), and 4.29 (SD 1.01), respectively. In responses for rationales, 4 subdimensions, including harm (mean 4.22, SD 0.97), relevance (mean 4.15, SD 0.83), fabrication and falsification (mean 4.12, SD 1.02), and understanding and reasoning (mean 4.04, SD 0.92) averagely scored ≥4. In contrast, consistency (mean 3.94, SD 0.72), clarity (mean 3.94, SD 0.89), comprehensiveness (mean 3.85, SD 0.83), and accuracy (mean 3.74, SD 0.91) performed at a moderate level. For the whole responses, the average self-awareness and trust and confidence scores for all chatbots were 3.84 (SD 0.89) and 3.88 (SD 0.91), respectively. Based on the average scores of the subdimensions, Claude 3.5 Sonnet and ChatGPT-4o were the top 2 outperformed models. Conclusions: The performance of AI chatbots in providing responses regarding well-established global guidelines in the prevention of SSI was acceptable, demonstrating immense potential in clinical applications. Nonetheless, a critical issue is the necessity of enhancing the stability of chatbots, as inaccurate responses can lead to severe consequences for SSI. Despite its limitations, it is anticipated that AI will trigger far-reaching changes in how clinicians access and use medical information.
- 9 July 2024
- Published by Elsevier in Biochimica et Biophysica Acta (BBA) - Reviews on Cancer
- Vol. 1879 (5) , 189152
New horizons in the mechanisms and therapeutic strategies for PD-L1 protein degradation in cancer
- 9 July 2024
- Published by Elsevier in Biochimica et Biophysica Acta (BBA) - Reviews on Cancer
- Vol. 1879 (5) , 189152
New horizons in the mechanisms and therapeutic strategies for PD-L1 protein degradation in cancer