Yuning Pan
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Yuning Pan has a total of 78 co-authors. Here are the top co-authors by number of shared publications:
Yichuan Liang
Department of Radiology, The First Affiliated Hospital of Ningbo University, Ningbo, Zhejiang, China
3 shared publicationsZhehao Zhang
Department of Radiology, The First Affiliated Hospital of Ningbo University, Ningbo, China
3 shared publicationsAijing Li
Department of Radiology, Ningbo No. 2 Hospital, Ningbo, Zhejiang, China
2 shared publicationsHui Zhang
Department of Radiology, THE First Affiliated Hospital of Ningbo University, Ningbo, Zhejiang, 315000, People’s Republic of China
2 shared publicationsHuiying Wang
Department of Radiology, The First Affiliated Hospital of Ningbo University, Ningbo, Zhejiang, China
2 shared publicationsLi Wang
Department of Elderly Health Care and Management, Ningbo College of Health Sciences, Ningbo, Zhejiang, 315000, People’s Republic of China
2 shared publications
Latest Publications
- Open Access
- 1 August 2026
- Published by AME Publishing Company in Quantitative Imaging in Medicine and Surgery
- Vol. 16 (8) , 636
Wireless detector-based magnetic resonance imaging for grading assessment of the peridiscal attachments in temporomandibular joint internal derangement
Background: The relationship between the peridiscal attachments and temporomandibular joint (TMJ) internal derangement (ID) remains poorly understood. This study aimed to investigate the association between graded imaging features of the peridiscal attachments and disc displacement using wireless detector-based magnetic resonance imaging (WD-MRI). Methods: This retrospective study included 61 patients (122 TMJs) diagnosed with TMD according to the Diagnostic Criteria for Temporomandibular Disorders (DC/TMD). All patients underwent oblique sagittal proton density-weighted imaging in both closed- and open-mouth positions using WD-MRI. Two radiologists independently assessed disc position [normal, disc displacement with reduction (DDWR), or disc displacement without reduction (DDWOR)] and performed ordinal grading (1–3 points: clearly, indistinctly, or non-visualized) of four peridiscal attachments: the anterior attachment superior (AAS), anterior attachment inferior (AAI), posterior attachment superior (PAS), and posterior attachment inferior (PAI). Inter-observer reliability between the two radiologists was evaluated using the weighted kappa coefficient. Group differences and correlations between attachment scores and disc position were analyzed using the Kruskal-Wallis H test, Mann-Whitney U test, and Spearman correlation analysis. Results: A total of 122 TMJs (12 normal, 37 DDWR, and 73 DDWOR) were evaluated using WD-MRI. Peridiscal attachment scores differed significantly among the three groups (all PConclusions: This study demonstrated that structural alterations in the peridiscal attachments, particularly the posterior attachment, are significantly correlated with TMJ ID severity on WD-MRI, suggesting their potential value in TMD assessment.
- 1 August 2026
- Published by AME Publishing Company in Quantitative Imaging in Medicine and Surgery
- Vol. 16 (8) , 636
Open AccessWireless detector-based magnetic resonance imaging for grading assessment of the peridiscal attachments in temporomandibular joint internal derangement
Background: The relationship between the peridiscal attachments and temporomandibular joint (TMJ) internal derangement (ID) remains poorly understood. This study aimed to investigate the association between graded imaging features of the peridiscal attachments and disc displacement using wireless detector-based magnetic resonance imaging (WD-MRI). Methods: This retrospective study included 61 patients (122 TMJs) diagnosed with TMD according to the Diagnostic Criteria for Temporomandibular Disorders (DC/TMD). All patients underwent oblique sagittal proton density-weighted imaging in both closed- and open-mouth positions using WD-MRI. Two radiologists independently assessed disc position [normal, disc displacement with reduction (DDWR), or disc displacement without reduction (DDWOR)] and performed ordinal grading (1–3 points: clearly, indistinctly, or non-visualized) of four peridiscal attachments: the anterior attachment superior (AAS), anterior attachment inferior (AAI), posterior attachment superior (PAS), and posterior attachment inferior (PAI). Inter-observer reliability between the two radiologists was evaluated using the weighted kappa coefficient. Group differences and correlations between attachment scores and disc position were analyzed using the Kruskal-Wallis H test, Mann-Whitney U test, and Spearman correlation analysis. Results: A total of 122 TMJs (12 normal, 37 DDWR, and 73 DDWOR) were evaluated using WD-MRI. Peridiscal attachment scores differed significantly among the three groups (all PConclusions: This study demonstrated that structural alterations in the peridiscal attachments, particularly the posterior attachment, are significantly correlated with TMJ ID severity on WD-MRI, suggesting their potential value in TMD assessment.
- 20 February 2026
- Published by Springer Nature in Neuroradiology
- Vol. 68 (4) , 843-854
Comparison of AI-assisted quantitative collateral score and visual collateral score in stroke thrombectomy triage
- 20 February 2026
- Published by Springer Nature in Neuroradiology
- Vol. 68 (4) , 843-854
Comparison of AI-assisted quantitative collateral score and visual collateral score in stroke thrombectomy triage
- Open Access
- 13 January 2026
- Published by Springer Nature in BioMedical Engineering OnLine
- Vol. 25 (1)
Desargues cloud TPS: a cloud-based automatic radiation treatment planning system for IMRT
To develop a cloud-based automated treatment planning system for intensity-modulated radiation therapy and evaluate its efficacy and safety for tumors in v
- 13 January 2026
- Published by Springer Nature in BioMedical Engineering OnLine
- Vol. 25 (1)
Open AccessDesargues cloud TPS: a cloud-based automatic radiation treatment planning system for IMRT
To develop a cloud-based automated treatment planning system for intensity-modulated radiation therapy and evaluate its efficacy and safety for tumors in v
- Open Access
- 1 December 2025
- Published by Springer Nature in BMC Medical Imaging
- Vol. 26 (1)
Automated abdominal aortic calcification scoring via deep learning: a multi-center validation of LVLCRNet
Background: To develop and validate a deep learning model for automated quantification of abdominal aortic calcification scores (AACS) adhering to the Kauppila protocol, with multicenter clinical validation. Methods: This retrospective multicenter study analyzed 2,660 lateral lumbar/thoracoabdominal radiographs from four centers, partitioned into development (training: n = 1,478; validation: n = 423) and test cohorts (internal: n = 211; external: n = 157 from Center C and n = 391 from Center D). We proposed the Lumbar Vertebrae Localization-Contrastive Rank-Aware Network (LVLCRNet), incorporating automatic lumbar vertebrae localization, aortic region segmentation, and contrastive rank-aware network for ordinal classification. Comparative analyses against baseline network and Lumbar Vertebrae Localization Network were conducted using expert-annotated AACS as ground truth (GT), evaluated through Wilcoxon matched-paired signed-rank test, intraclass correlation coefficient (ICC), mean absolute error (MAE), coefficient of determination (R²), Bland-Altman analysis, and multiclass accuracy. Results: No significant difference was found between LVLCRNet and GT, whereas the baseline network showed significant deviations from GT across all cohorts ( p < 0.017, Bonferroni-corrected). LVLCRNet achieved superior agreement with GT, demonstrating R² of 0.858 (internal) and 0.842/0.837 (external), ICC of 0.916 (internal) and 0.904/0.899 (external), and MAE of 1.547 (internal) and 1.189/1.972 (external). Bland-Altman analysis showed minimal systemic bias. Classification accuracy reached 82.94% (internal) and 79.62%/81.59% (external), outperforming comparators by 5.10–9.98%. Conclusion: LVLCRNet provides reliable automated AACS through integrated anatomic localization and contrastive rank-aware learning. Its strong generalizability and precision in severity grading support clinical utility for cardiovascular risk stratification.
- 1 December 2025
- Published by Springer Nature in BMC Medical Imaging
- Vol. 26 (1)
Open AccessAutomated abdominal aortic calcification scoring via deep learning: a multi-center validation of LVLCRNet
Background: To develop and validate a deep learning model for automated quantification of abdominal aortic calcification scores (AACS) adhering to the Kauppila protocol, with multicenter clinical validation. Methods: This retrospective multicenter study analyzed 2,660 lateral lumbar/thoracoabdominal radiographs from four centers, partitioned into development (training: n = 1,478; validation: n = 423) and test cohorts (internal: n = 211; external: n = 157 from Center C and n = 391 from Center D). We proposed the Lumbar Vertebrae Localization-Contrastive Rank-Aware Network (LVLCRNet), incorporating automatic lumbar vertebrae localization, aortic region segmentation, and contrastive rank-aware network for ordinal classification. Comparative analyses against baseline network and Lumbar Vertebrae Localization Network were conducted using expert-annotated AACS as ground truth (GT), evaluated through Wilcoxon matched-paired signed-rank test, intraclass correlation coefficient (ICC), mean absolute error (MAE), coefficient of determination (R²), Bland-Altman analysis, and multiclass accuracy. Results: No significant difference was found between LVLCRNet and GT, whereas the baseline network showed significant deviations from GT across all cohorts ( p < 0.017, Bonferroni-corrected). LVLCRNet achieved superior agreement with GT, demonstrating R² of 0.858 (internal) and 0.842/0.837 (external), ICC of 0.916 (internal) and 0.904/0.899 (external), and MAE of 1.547 (internal) and 1.189/1.972 (external). Bland-Altman analysis showed minimal systemic bias. Classification accuracy reached 82.94% (internal) and 79.62%/81.59% (external), outperforming comparators by 5.10–9.98%. Conclusion: LVLCRNet provides reliable automated AACS through integrated anatomic localization and contrastive rank-aware learning. Its strong generalizability and precision in severity grading support clinical utility for cardiovascular risk stratification.
- Open Access
- 21 October 2025
- Published by Frontiers Media SA in Frontiers in Cardiovascular Medicine
- Vol. 12, 1647882
Deep learning-based automated quantification system for abdominal aortic calcification: multicenter cohort study for algorithm development and clinical validation
Objectives: To establish an automated scoring system for abdominal aortic calcification (AAC) to facilitate standardized quantitative imaging analysis in support of clinical decision-making in atherosclerosis management.Methods: x-ray images of the abdominal aorta were obtained for 2,941 individuals from five medical centers in Zhejiang Province. Calcification severity was graded manually using the Kauppila scoring system, and cases were stratified into three groups based on total calcification burden. The automated assessment framework comprised two sequential components: a lumbar spine segmentation model based on nnUnet and an AAC score regression model based on ResNet. Model development was conducted using 1,737 training cases, with internal validation in 471 cases and external validation in 733 cases from independent centers. A retrospective matched cohort study was conducted in 200 AAC patients from Center B (100 dialysis-dependent and 100 not dialysis-dependent cases), to investigate associations with major adverse cardiovascular events.Results: The developed automated quantification system demonstrated mean absolute errors of 1.686 (internal validation set) and 1.920 (external validation set), with strong correlation to expert ratings (Spearman's ρ = 0.923 and 0.888, respectively, both P < 0.001). Inter-rater reliability analysis revealed excellent agreement with manual scoring (intraclass correlation coefficients of 0.913 internally and 0.874 externally). Stratification based on calcification severity showed optimal sensitivity for the moderate calcification category (88.6%), with superior specificity for the non/mild (94.2%) and severe (91.5%) categories.Conclusion: The established automated quantification system for AAC exhibits good assessment efficiency and measurement accuracy, offering a standardized approach to refine cardiovascular risk stratification in clinical practice.
- 21 October 2025
- Published by Frontiers Media SA in Frontiers in Cardiovascular Medicine
- Vol. 12, 1647882
Open AccessDeep learning-based automated quantification system for abdominal aortic calcification: multicenter cohort study for algorithm development and clinical validation
Objectives: To establish an automated scoring system for abdominal aortic calcification (AAC) to facilitate standardized quantitative imaging analysis in support of clinical decision-making in atherosclerosis management.Methods: x-ray images of the abdominal aorta were obtained for 2,941 individuals from five medical centers in Zhejiang Province. Calcification severity was graded manually using the Kauppila scoring system, and cases were stratified into three groups based on total calcification burden. The automated assessment framework comprised two sequential components: a lumbar spine segmentation model based on nnUnet and an AAC score regression model based on ResNet. Model development was conducted using 1,737 training cases, with internal validation in 471 cases and external validation in 733 cases from independent centers. A retrospective matched cohort study was conducted in 200 AAC patients from Center B (100 dialysis-dependent and 100 not dialysis-dependent cases), to investigate associations with major adverse cardiovascular events.Results: The developed automated quantification system demonstrated mean absolute errors of 1.686 (internal validation set) and 1.920 (external validation set), with strong correlation to expert ratings (Spearman's ρ = 0.923 and 0.888, respectively, both P < 0.001). Inter-rater reliability analysis revealed excellent agreement with manual scoring (intraclass correlation coefficients of 0.913 internally and 0.874 externally). Stratification based on calcification severity showed optimal sensitivity for the moderate calcification category (88.6%), with superior specificity for the non/mild (94.2%) and severe (91.5%) categories.Conclusion: The established automated quantification system for AAC exhibits good assessment efficiency and measurement accuracy, offering a standardized approach to refine cardiovascular risk stratification in clinical practice.
- Open Access
- 30 July 2025
- Published by Frontiers Media SA in Frontiers in Oncology
- Vol. 15, 1457238
Assessment of the imaging quality and heterogeneous predictive value of readout-segmented echo planar imaging for diffusion-weighted rectal MRI
Objectives: To qualitatively and quantitatively compare the image quality of readout-segmented echo planar imaging (rs-EPI) and single-shot echo planar imaging (ss-EPI) for diffusion-weighted (DWI) rectal MRI, as well as the heterogeneous predictive value of the apparent diffusion coefficient (ADC) values obtained by the two DWI techniques.Methods: The rs-EPI and ss-EPI images were subjectively assessed for lesion sharpness, display of normal structure, overall image quality, geometric distortion, and anatomical differences. The signal-to-noise ratio (SNR), contrast ratio (CR), contrast-to-noise ratio (CNR), and ADC values were objectively compared. Pearson’s correlations and ROC analysis were used to explore the relationships of ADC values obtained by the two techniques and nucleus related antigen (Ki-67) and hypoxia inducible factor-1α (HIF-1α).Results: Eighty patients with rectal cancer (RC) were included. Lesion sharpness, normal structure display, overall image quality, geometric distortion and anatomical structure differences in the rs-EPI DWI group were higher than in the ss-EPI DWI group (PPP=0.23, 0.75).Conclusion: rs-EPI DWI can improve image quality and the ADC value is associated with pathologic markers of tumor aggression.
- 30 July 2025
- Published by Frontiers Media SA in Frontiers in Oncology
- Vol. 15, 1457238
Open AccessAssessment of the imaging quality and heterogeneous predictive value of readout-segmented echo planar imaging for diffusion-weighted rectal MRI
Objectives: To qualitatively and quantitatively compare the image quality of readout-segmented echo planar imaging (rs-EPI) and single-shot echo planar imaging (ss-EPI) for diffusion-weighted (DWI) rectal MRI, as well as the heterogeneous predictive value of the apparent diffusion coefficient (ADC) values obtained by the two DWI techniques.Methods: The rs-EPI and ss-EPI images were subjectively assessed for lesion sharpness, display of normal structure, overall image quality, geometric distortion, and anatomical differences. The signal-to-noise ratio (SNR), contrast ratio (CR), contrast-to-noise ratio (CNR), and ADC values were objectively compared. Pearson’s correlations and ROC analysis were used to explore the relationships of ADC values obtained by the two techniques and nucleus related antigen (Ki-67) and hypoxia inducible factor-1α (HIF-1α).Results: Eighty patients with rectal cancer (RC) were included. Lesion sharpness, normal structure display, overall image quality, geometric distortion and anatomical structure differences in the rs-EPI DWI group were higher than in the ss-EPI DWI group (PPP=0.23, 0.75).Conclusion: rs-EPI DWI can improve image quality and the ADC value is associated with pathologic markers of tumor aggression.
- Open Access
- 2 July 2025
- Published by Frontiers Media SA in Frontiers in Nutrition
- Vol. 12, 1582527
Sex-specific effect of dietary fatty acids on non-alcoholic fatty liver disease
Background: Non-alcoholic fatty liver disease is the most common chronic hepatic disease worldwide. Dietary fatty acid is tightly associated with the development of non-alcoholic fatty liver disease but few large-scale and in-depth clinical researches have focused on the issue.Methods: We conducted a retrospective case-control study based on the data from the 2017–2018 cycle of the National Health and Nutrition Examination Survey.Results: A total of 2,470 adult participants were included in this study. Logistic regression analysis showed that dietary fatty acids were positively associated with non-alcoholic fatty liver disease (odds ratio and 95% confidence interval >1 and P < 0.05) except for polyunsaturated fatty acid. Subgroup analysis stratified by age stage and weight grade revealed that aforementioned association was significant only in individuals aged group 37–55 and those classified as obesity. In addition, all the fatty acid related ratios (the ratio of unsaturated fatty acids to saturated fatty acids, the ratio of polyunsaturated fatty acid to saturated fatty acid, the ratio of monounsaturated fatty acids to saturated fatty acids, the ratio of polyunsaturated fatty acids to monounsaturated fatty acids) showed protective effects against the onset and steatosis severity of non-alcoholic fatty liver disease in men, as evidenced by stratified logistic regression analysis (all the odds ratio [95% confidence interval] < 1 and P < 0.05) and smooth curve fittings.Conclusion: These findings suggest that dietary fatty acids modification could serve as a preventive strategy for male non-alcoholic fatty liver disease. Increasing the proportion of unsaturated fatty acids in the diet, especially polyunsaturated fatty acids, is promising to prevent non-alcoholic fatty liver disease in middle-aged obese men.
- 2 July 2025
- Published by Frontiers Media SA in Frontiers in Nutrition
- Vol. 12, 1582527
Open AccessSex-specific effect of dietary fatty acids on non-alcoholic fatty liver disease
Background: Non-alcoholic fatty liver disease is the most common chronic hepatic disease worldwide. Dietary fatty acid is tightly associated with the development of non-alcoholic fatty liver disease but few large-scale and in-depth clinical researches have focused on the issue.Methods: We conducted a retrospective case-control study based on the data from the 2017–2018 cycle of the National Health and Nutrition Examination Survey.Results: A total of 2,470 adult participants were included in this study. Logistic regression analysis showed that dietary fatty acids were positively associated with non-alcoholic fatty liver disease (odds ratio and 95% confidence interval >1 and P < 0.05) except for polyunsaturated fatty acid. Subgroup analysis stratified by age stage and weight grade revealed that aforementioned association was significant only in individuals aged group 37–55 and those classified as obesity. In addition, all the fatty acid related ratios (the ratio of unsaturated fatty acids to saturated fatty acids, the ratio of polyunsaturated fatty acid to saturated fatty acid, the ratio of monounsaturated fatty acids to saturated fatty acids, the ratio of polyunsaturated fatty acids to monounsaturated fatty acids) showed protective effects against the onset and steatosis severity of non-alcoholic fatty liver disease in men, as evidenced by stratified logistic regression analysis (all the odds ratio [95% confidence interval] < 1 and P < 0.05) and smooth curve fittings.Conclusion: These findings suggest that dietary fatty acids modification could serve as a preventive strategy for male non-alcoholic fatty liver disease. Increasing the proportion of unsaturated fatty acids in the diet, especially polyunsaturated fatty acids, is promising to prevent non-alcoholic fatty liver disease in middle-aged obese men.
- Open Access
- 1 June 2025
- Published by Taylor & Francis in Neuropsychiatric Disease and Treatment
- Vol. ume 21, 1161-1172
Dynamic Changes in ALFF and DC in Subthreshold Depression Before and After ICBT Intervention
Purpose: Subthreshold depression (SD) is a highly prevalent mental health condition and a key precursor in major depressive disorder. Internet-based cognitive behavioral therapy (ICBT) effectively treats SD in clinical settings, but the neuromechanisms behind its success are not fully understood. This study aimed to examine changes in intrinsic brain activity in SD and their further alterations after ICBT intervention. Patients and Methods: Twenty-five individuals with SD and 25 matched healthy controls (HC) were enrolled. SD underwent the same MRI scans and clinical assessments before and after a five-week ICBT intervention. HC underwent a single MRI scan during enrollment. The amplitude of low-frequency fluctuation (ALFF) and degree centrality (DC) are established analytical techniques in the field of neuroimaging to assess intrinsic brain activity. Two-sample and paired-sample t-tests were employed to examine changes in whole-brain ALFF and DC. Pearson or Spearman correlation analysis was performed to evaluate associations between altered ALFF/DC values and changes in clinical assessment scores in SD. Results: Compared with HC, decreased DC in the right precuneus (PCUN) and increased ALFF in the right insula (INS) were found in SD before ICBT intervention. After ICBT intervention, ALFF in the left hippocampus (HIP) was increased, and DC in the right medial and paracingulate gyri (DCG) was reduced in SD. Altered ALFF values in the left HIP showed a negative correlation with changes in PHQ-9 scores, and a positive correlation with changes in FATS-ReB scores. Altered DC values in the right DCG were positively correlated with changes in CCTS-TR-BhA scores. Conclusion: These findings provide new insights into the altered intrinsic brain activities in SD. More importantly, this study found potential changes in intrinsic activity in brain regions associated with ICBT response to SD, which may help further explore the neuropsychological mechanism underlying the efficacy of ICBT in the future.
- 1 June 2025
- Published by Taylor & Francis in Neuropsychiatric Disease and Treatment
- Vol. ume 21, 1161-1172
Open AccessDynamic Changes in ALFF and DC in Subthreshold Depression Before and After ICBT Intervention
Purpose: Subthreshold depression (SD) is a highly prevalent mental health condition and a key precursor in major depressive disorder. Internet-based cognitive behavioral therapy (ICBT) effectively treats SD in clinical settings, but the neuromechanisms behind its success are not fully understood. This study aimed to examine changes in intrinsic brain activity in SD and their further alterations after ICBT intervention. Patients and Methods: Twenty-five individuals with SD and 25 matched healthy controls (HC) were enrolled. SD underwent the same MRI scans and clinical assessments before and after a five-week ICBT intervention. HC underwent a single MRI scan during enrollment. The amplitude of low-frequency fluctuation (ALFF) and degree centrality (DC) are established analytical techniques in the field of neuroimaging to assess intrinsic brain activity. Two-sample and paired-sample t-tests were employed to examine changes in whole-brain ALFF and DC. Pearson or Spearman correlation analysis was performed to evaluate associations between altered ALFF/DC values and changes in clinical assessment scores in SD. Results: Compared with HC, decreased DC in the right precuneus (PCUN) and increased ALFF in the right insula (INS) were found in SD before ICBT intervention. After ICBT intervention, ALFF in the left hippocampus (HIP) was increased, and DC in the right medial and paracingulate gyri (DCG) was reduced in SD. Altered ALFF values in the left HIP showed a negative correlation with changes in PHQ-9 scores, and a positive correlation with changes in FATS-ReB scores. Altered DC values in the right DCG were positively correlated with changes in CCTS-TR-BhA scores. Conclusion: These findings provide new insights into the altered intrinsic brain activities in SD. More importantly, this study found potential changes in intrinsic activity in brain regions associated with ICBT response to SD, which may help further explore the neuropsychological mechanism underlying the efficacy of ICBT in the future.
- Open Access
- 12 May 2025
- Published by Frontiers Media SA in Frontiers in Neurology
- Vol. 16, 1502382
Automatic collateral quantification in acute ischemic stroke using U2-net
Objectives: To harness the U2-Net deep learning framework for automated quantification of collateral circulation in acute ischemic stroke (AIS) via computed tomography angiography (CTA) images, comparing its performance against traditional visual collateral scores (vCS).Methods: A cohort of 118 confirmed AIS cases was assembled and stratified into 94 development and 24 test cases. CTA images underwent preprocessing and annotation. The U2-Net was trained to segment collateral vessels, yielding a quantitative collateral score (qCS) based on vessel volume ratios between affected and healthy hemispheres. Performance was assessed via Dice Similarity Coefficient (DSC), Spearman correlation, Intraclass Correlation Coefficient (ICC), and accuracy, with comparisons to vCS (Tan and Menon score) and ground truth.Result: The U2-Net demonstrated robust segmentation capabilities, achieving a mean DSC of 0.75 in the test set. The qCS showed a strong correlation with vCS with ρ ranging from 0.78 to 0.92. When compared to the more refined six-class Menon score, the qCS exhibited stronger consistency (development set: ICC = 0.83, test set: ICC = 0.93) than when compared to the four-class Tan score (development set: ICC = 0.76, test set: ICC = 0.79). In terms of classification accuracy, the AI model achieved 0.83 and 0.71 against ground truth and vCS, respectively, for four-class classification. This accuracy escalated to 0.88 and 0.83 for binary classification, emphasizing its proficiency in differentiating collateral status.Conclusion: Our U2-Net AI model offers a reliable, objective tool for quantifying collateral circulation in AIS. The qCS aligns well with vCS and demonstrates the feasibility of automated collateral assessment, which may enhance diagnostic accuracy and therapeutic decision-making.
- 12 May 2025
- Published by Frontiers Media SA in Frontiers in Neurology
- Vol. 16, 1502382
Open AccessAutomatic collateral quantification in acute ischemic stroke using U2-net
Objectives: To harness the U2-Net deep learning framework for automated quantification of collateral circulation in acute ischemic stroke (AIS) via computed tomography angiography (CTA) images, comparing its performance against traditional visual collateral scores (vCS).Methods: A cohort of 118 confirmed AIS cases was assembled and stratified into 94 development and 24 test cases. CTA images underwent preprocessing and annotation. The U2-Net was trained to segment collateral vessels, yielding a quantitative collateral score (qCS) based on vessel volume ratios between affected and healthy hemispheres. Performance was assessed via Dice Similarity Coefficient (DSC), Spearman correlation, Intraclass Correlation Coefficient (ICC), and accuracy, with comparisons to vCS (Tan and Menon score) and ground truth.Result: The U2-Net demonstrated robust segmentation capabilities, achieving a mean DSC of 0.75 in the test set. The qCS showed a strong correlation with vCS with ρ ranging from 0.78 to 0.92. When compared to the more refined six-class Menon score, the qCS exhibited stronger consistency (development set: ICC = 0.83, test set: ICC = 0.93) than when compared to the four-class Tan score (development set: ICC = 0.76, test set: ICC = 0.79). In terms of classification accuracy, the AI model achieved 0.83 and 0.71 against ground truth and vCS, respectively, for four-class classification. This accuracy escalated to 0.88 and 0.83 for binary classification, emphasizing its proficiency in differentiating collateral status.Conclusion: Our U2-Net AI model offers a reliable, objective tool for quantifying collateral circulation in AIS. The qCS aligns well with vCS and demonstrates the feasibility of automated collateral assessment, which may enhance diagnostic accuracy and therapeutic decision-making.