Pub Date : 2024-06-24DOI: 10.1101/2024.06.24.24309298
Corey Ratcliffe, Christophe de Bezenac, Kumar Das, Shubhabrata Biswas, Anthony Marson, Simon S. Keller
Introduction: Structural neuroimaging analyses require 'research quality' images, acquired with costly MRI acquisitions. Isotropic (3D) T1 images are desirable for quantitative analyses, however a routine compromise in the clinical setting is to acquire anisotropic (2D) analogues for qualitative visual inspection. Machine learning-based software have shown promise in addressing some of the limitations of 2D scans in research applications, yet their efficacy in quantitative research is not well understood. To evaluate the applicability of image preprocessing methods, morphometry in idiopathic generalised epilepsy (IGE)--in which, pathology-related abnormalities of the subcortical structures have been reproducibly demonstrated--was investigated first in 3D scans, then in 2D scans, resampled images, and synthesised images. Methods: 2D and 3D T1 MRI were acquired during the same scanning session from 31 individuals (males = 14, mean age = 32.16) undergoing evaluation for IGE at the Walton Centre NHS Foundation Trust, Liverpool, as well as 39 healthy age- and sex-matched controls (males = 16, mean age = 32.13). The DL+DiReCT pipeline was used to provide segmentations of the 2D images, and estimates of regional volume and thickness. The 2D scans were also resampled into isotropic images using NiBabel, and preprocessed into synthetic isotropic images using SynthSR. For the 3D scans, untransformed 2D scans, resampled images, and synthesised images, FreeSurfer 7.2.0 was used to create parcellations of 178 anatomical regions (equivalent to the 178 parcellations provided as part of the DL+DiReCT pipeline), defined by the aseg and Destrieux atlases, and FSL FIRST was used to segment subcortical surface shapes. Spatial correspondence and intraclass correlations between the morphometrics of the five parcellations were first determined, then subcortical surface shape abnormalities associated with IGE were identified by comparing the FSL FIRST outputs of patients with controls. Results: When standardised to the metrics derived from the 3D scans, cortical volume and thickness estimates trended lower for the untransformed 2D, DL+DiReCT, resampled, and SynthSR images, whereas subcortical volume estimates did not differ. Dice coefficients revealed a low spatial similarity between the cortices of the 3D scans and the other images overall, which was higher in the subcortical structures. Intraclass correlation coefficients reiterated this disparity, with estimates of thickness being less similar than those of volume, and DL+DiReCT estimates trending less similar than the other images types. For the people with epilepsy, the 3D scans showed significant surface deflations across various subcortical structures when compared to healthy controls. Analysis of the untransformed 2D scans enabled the detection of a subset of subcortical abnormalities, whereas analyses of the resampled and synthetic images attenuated almost all significance. Conclusions: Generalised image synthe
{"title":"Machine Learning-Based Reconstruction of 2D MRI for Quantitative Morphometry in Epilepsy","authors":"Corey Ratcliffe, Christophe de Bezenac, Kumar Das, Shubhabrata Biswas, Anthony Marson, Simon S. Keller","doi":"10.1101/2024.06.24.24309298","DOIUrl":"https://doi.org/10.1101/2024.06.24.24309298","url":null,"abstract":"Introduction: Structural neuroimaging analyses require 'research quality' images, acquired with costly MRI acquisitions. Isotropic (3D) T1 images are desirable for quantitative analyses, however a routine compromise in the clinical setting is to acquire anisotropic (2D) analogues for qualitative visual inspection. Machine learning-based software have shown promise in addressing some of the limitations of 2D scans in research applications, yet their efficacy in quantitative research is not well understood. To evaluate the applicability of image preprocessing methods, morphometry in idiopathic generalised epilepsy (IGE)--in which, pathology-related abnormalities of the subcortical structures have been reproducibly demonstrated--was investigated first in 3D scans, then in 2D scans, resampled images, and synthesised images. Methods: 2D and 3D T1 MRI were acquired during the same scanning session from 31 individuals (males = 14, mean age = 32.16) undergoing evaluation for IGE at the Walton Centre NHS Foundation Trust, Liverpool, as well as 39 healthy age- and sex-matched controls (males = 16, mean age = 32.13). The DL+DiReCT pipeline was used to provide segmentations of the 2D images, and estimates of regional volume and thickness. The 2D scans were also resampled into isotropic images using NiBabel, and preprocessed into synthetic isotropic images using SynthSR. For the 3D scans, untransformed 2D scans, resampled images, and synthesised images, FreeSurfer 7.2.0 was used to create parcellations of 178 anatomical regions (equivalent to the 178 parcellations provided as part of the DL+DiReCT pipeline), defined by the aseg and Destrieux atlases, and FSL FIRST was used to segment subcortical surface shapes. Spatial correspondence and intraclass correlations between the morphometrics of the five parcellations were first determined, then subcortical surface shape abnormalities associated with IGE were identified by comparing the FSL FIRST outputs of patients with controls. Results: When standardised to the metrics derived from the 3D scans, cortical volume and thickness estimates trended lower for the untransformed 2D, DL+DiReCT, resampled, and SynthSR images, whereas subcortical volume estimates did not differ. Dice coefficients revealed a low spatial similarity between the cortices of the 3D scans and the other images overall, which was higher in the subcortical structures. Intraclass correlation coefficients reiterated this disparity, with estimates of thickness being less similar than those of volume, and DL+DiReCT estimates trending less similar than the other images types. For the people with epilepsy, the 3D scans showed significant surface deflations across various subcortical structures when compared to healthy controls. Analysis of the untransformed 2D scans enabled the detection of a subset of subcortical abnormalities, whereas analyses of the resampled and synthetic images attenuated almost all significance. Conclusions: Generalised image synthe","PeriodicalId":501358,"journal":{"name":"medRxiv - Radiology and Imaging","volume":"8 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-06-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141503254","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
This paper introduces a novel, entity-aware metric, termed as Radiological Report (Text) Evaluation (RaTEScore), to assess the quality of medical reports generated by AI models. RaTEScore emphasizes crucial medical entities such as diagnostic outcomes and anatomical details, and is robust against complex medical synonyms and sensitive to negation expressions. Technically, we developed a comprehensive medical NER dataset, RaTE-NER, and trained an NER model specifically for this purpose. This model enables the decomposition of complex radiological reports into constituent medical entities. The metric itself is derived by comparing the similarity of entity embeddings, obtained from a language model, based on their types and relevance to clinical significance. Our evaluations demonstrate that RaTEScore aligns more closely with human preference than existing metrics, validated both on established public benchmarks and our newly proposed RaTE-Eval benchmark.
本文介绍了一种新颖的实体感知度量方法，称为放射报告（文本）评估（RaTEScore），用于评估人工智能模型生成的医疗报告的质量。RaTEScore 强调诊断结果和解剖细节等关键医学实体，对复杂的医学同义词具有鲁棒性，对否定表达也很敏感。在技术上，我们开发了一个全面的医学 NER 数据集 RaTE-NER，并专门为此训练了一个 NER 模型。该模型可将复杂的放射报告分解为组成医疗实体。该指标本身是通过比较从语言模型中获得的实体嵌入的相似性而得出的，其依据是实体的类型和与临床意义的相关性。我们的评估结果表明，与现有指标相比，RaTEScore 更符合人类的偏好，这一点在已有的公共基准和我们新提出的 RaTE-Eval 基准上都得到了验证。
{"title":"RaTEScore: A Metric for Radiology Report Generation","authors":"Weike Zhao, Chaoyi Wu, Xiaoman Zhang, Ya Zhang, Yanfeng Wang, Weidi Xie","doi":"10.1101/2024.06.24.24309405","DOIUrl":"https://doi.org/10.1101/2024.06.24.24309405","url":null,"abstract":"This paper introduces a novel, entity-aware metric, termed as Radiological Report (Text) Evaluation (RaTEScore), to assess the quality of medical reports generated by AI models. RaTEScore emphasizes crucial medical entities such as diagnostic outcomes and anatomical details, and is robust against complex medical synonyms and sensitive to negation expressions. Technically, we developed a comprehensive medical NER dataset, RaTE-NER, and trained an NER model specifically for this purpose. This model enables the decomposition of complex radiological reports into constituent medical entities. The metric itself is derived by comparing the similarity of entity embeddings, obtained from a language model, based on their types and relevance to clinical significance. Our evaluations demonstrate that RaTEScore aligns more closely with human preference than existing metrics, validated both on established public benchmarks and our newly proposed RaTE-Eval benchmark.","PeriodicalId":501358,"journal":{"name":"medRxiv - Radiology and Imaging","volume":"30 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-06-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141528929","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-06-24DOI: 10.1101/2024.06.24.24308597
Julian Klinger, Doris Leithner, Sungmin Woo, Michael Weber, Hebert Alberto Vargas, Marius E. Mayerhoefer
Objectives: To determine the impact of segmentation techniques on radiomic features extracted from ultrahigh-field (UHF) MRI of the brain. Materials and Methods: Twenty-one 7T MRI scans of the brain, including a 3D magnetization-prepared two rapid acquisition gradient echo (MP2RAGE) T1-weighted sequence with an isotropic 0.63 mm3 voxel size, were analyzed. Radiomic features (histogram, texture, and shape; total n=101) from six brain regions -cerebral gray and white matter, basal ganglia, ventricles, cerebellum, and brainstem- were extracted from segmentation masks constructed with four different techniques: the iGT (reference standard), based on a custom pipeline that combined automatic segmentation tools and expert reader correction; the deep-learning algorithm Cerebrum-7T; the Freesurfer-v7 software suite; and the Nighres algorithm. Principal components (PCs) were calculated for histogram and texture features. To test the reproducibility of radiomic features, intraclass correlation coefficients (ICC) were used to compare Cerebrum-7, Freesurfer-v7, and Nighres to the iGT, respectively. Results: For histogram PCs, median ICCs for Cerebrum-7T, Freesurfer-v7, and Nighres were 0.99, 0.42, and 0.11 for the gray matter; 0.84, 0.25, and 0.43 for the basal ganglia; 0.89, 0.063, and 0.036 for the white matter; 0.84, 0.21, and 0.33 for the ventricles; 0.94, 0.64, and 0.93 for the cerebellum; and 0.78, 0.21, and 0.53 for the brainstem. For texture PCs, median ICCs for Cerebrum-7T, Freesurfer-v7, and Nighres were 0.95, 0.21, and 0.15 for the gray matter; 0.70, 0.36, and 0.023 for the basal ganglia; 0.91, 0.25, and 0.023 for the white matter; 0.80, 0.75, and 0.59 for the ventricles; 0.95, 0.43, and 0.86 for the cerebellum; and 0.72, 0.39, and 0.46 for the brainstem. For shape features, median ICCs for Cerebrum-7T, FreeSurfer-v7, and Nighres were 0.99, 0.91, and 0.36 for the gray matter; 0.89, 0.90, and 0.13 for the basal ganglia; 0.98, 0.91, and 0.027 for the white matter; 0.91, 0.91, and 0.36 for the ventricles; 0.80, 0.68, and 0.47 for the cerebellum; and 0.79, 0.17, and 0.15 for the brainstem. Conclusions: Radiomic features in UHF MRI of the brain show substantial variability depending on the segmentation algorithm. The deep learning algorithm Cerebrum-7T enabled the highest reproducibility. Dedicated software tools for UHF MRI may be needed to achieve more stable results.
{"title":"Reproducibility of radiomic features of the brain on ultrahigh-resolution MRI at 7 Tesla: a comparison of different segmentation techniques","authors":"Julian Klinger, Doris Leithner, Sungmin Woo, Michael Weber, Hebert Alberto Vargas, Marius E. Mayerhoefer","doi":"10.1101/2024.06.24.24308597","DOIUrl":"https://doi.org/10.1101/2024.06.24.24308597","url":null,"abstract":"Objectives: To determine the impact of segmentation techniques on radiomic features extracted from ultrahigh-field (UHF) MRI of the brain. Materials and Methods: Twenty-one 7T MRI scans of the brain, including a 3D magnetization-prepared two rapid acquisition gradient echo (MP2RAGE) T1-weighted sequence with an isotropic 0.63 mm3 voxel size, were analyzed. Radiomic features (histogram, texture, and shape; total n=101) from six brain regions -cerebral gray and white matter, basal ganglia, ventricles, cerebellum, and brainstem- were extracted from segmentation masks constructed with four different techniques: the iGT (reference standard), based on a custom pipeline that combined automatic segmentation tools and expert reader correction; the deep-learning algorithm Cerebrum-7T; the Freesurfer-v7 software suite; and the Nighres algorithm. Principal components (PCs) were calculated for histogram and texture features. To test the reproducibility of radiomic features, intraclass correlation coefficients (ICC) were used to compare Cerebrum-7, Freesurfer-v7, and Nighres to the iGT, respectively. Results: For histogram PCs, median ICCs for Cerebrum-7T, Freesurfer-v7, and Nighres were 0.99, 0.42, and 0.11 for the gray matter; 0.84, 0.25, and 0.43 for the basal ganglia; 0.89, 0.063, and 0.036 for the white matter; 0.84, 0.21, and 0.33 for the ventricles; 0.94, 0.64, and 0.93 for the cerebellum; and 0.78, 0.21, and 0.53 for the brainstem. For texture PCs, median ICCs for Cerebrum-7T, Freesurfer-v7, and Nighres were 0.95, 0.21, and 0.15 for the gray matter; 0.70, 0.36, and 0.023 for the basal ganglia; 0.91, 0.25, and 0.023 for the white matter; 0.80, 0.75, and 0.59 for the ventricles; 0.95, 0.43, and 0.86 for the cerebellum; and 0.72, 0.39, and 0.46 for the brainstem. For shape features, median ICCs for Cerebrum-7T, FreeSurfer-v7, and Nighres were 0.99, 0.91, and 0.36 for the gray matter; 0.89, 0.90, and 0.13 for the basal ganglia; 0.98, 0.91, and 0.027 for the white matter; 0.91, 0.91, and 0.36 for the ventricles; 0.80, 0.68, and 0.47 for the cerebellum; and 0.79, 0.17, and 0.15 for the brainstem. Conclusions: Radiomic features in UHF MRI of the brain show substantial variability depending on the segmentation algorithm. The deep learning algorithm Cerebrum-7T enabled the highest reproducibility. Dedicated software tools for UHF MRI may be needed to achieve more stable results.","PeriodicalId":501358,"journal":{"name":"medRxiv - Radiology and Imaging","volume":"145 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-06-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141503255","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-06-23DOI: 10.1101/2024.06.21.24309261
Conan Chen, Ryan A Barnes, Katherine J Bangen, Fei Han, Josef Pfeuffer, Eric C Wong, Thomas T Liu, Divya S Bolar
Purpose: By leveraging the small-vessel specificity of velocity-selective arterial spin labeling (VSASL), we present a novel technique for measuring cerebral MicroVascular Pulsatility named MVP-VSASL. Theory and Methods: We present a theoretical model relating the pulsatile, cerebral blood flow-driven VSASL signal to the microvascular pulsatility index (PI), a widely used metric for quantifying cardiac-dependent fluctuations. The model describes the dependence of PI on bolus duration τ (an adjustable VSASL sequence parameter) and provides guidance for selecting a value of τ that maximizes the SNR of the PI measurement. The model predictions were assessed in humans using data acquired with retrospectively cardiac-gated VSASL sequences over a broad range of τ values. In vivo measurements were also used to demonstrate the feasibility of whole-brain voxel-wise PI mapping, assess intrasession repeatability of the PI measurement, and illustrate the potential of this method to explore an association with age. Results: The theoretical model showed excellent agreement to the empirical data in a gray matter region of interest (average R2 value of 0.898 ± 0.107 across six subjects). We further showed excellent intrasession repeatability of the pulsatility measurement (ICC = 0.960, p < 0.001) and the potential to characterize associations with age (r = 0.554, p = 0.021). Conclusion: We have introduced a novel, VSASL-based cerebral microvascular pulsatility technique, which may facilitate investigation of cognitive disorders where damage to the microvasculature has been implicated.
{"title":"MVP-VSASL: measuring MicroVascular Pulsatility using Velocity-Selective Arterial Spin Labeling","authors":"Conan Chen, Ryan A Barnes, Katherine J Bangen, Fei Han, Josef Pfeuffer, Eric C Wong, Thomas T Liu, Divya S Bolar","doi":"10.1101/2024.06.21.24309261","DOIUrl":"https://doi.org/10.1101/2024.06.21.24309261","url":null,"abstract":"<strong>Purpose:</strong> By leveraging the small-vessel specificity of velocity-selective arterial spin labeling (VSASL), we present a novel technique for measuring cerebral MicroVascular Pulsatility named MVP-VSASL.\u0000<strong>Theory and Methods:</strong> We present a theoretical model relating the pulsatile, cerebral blood flow-driven VSASL signal to the microvascular pulsatility index (PI), a widely used metric for quantifying cardiac-dependent fluctuations. The model describes the dependence of PI on bolus duration τ (an adjustable VSASL sequence parameter) and provides guidance for selecting a value of τ that maximizes the SNR of the PI measurement. The model predictions were assessed in humans using data acquired with retrospectively cardiac-gated VSASL sequences over a broad range of τ values. In vivo measurements were also used to demonstrate the feasibility of whole-brain voxel-wise PI mapping, assess intrasession repeatability of the PI measurement, and illustrate the potential of this method to explore an association with age.\u0000<strong>Results:</strong> The theoretical model showed excellent agreement to the empirical data in a gray matter region of interest (average R<sup>2</sup> value of 0.898 ± 0.107 across six subjects). We further showed excellent intrasession repeatability of the pulsatility measurement (ICC = 0.960, p < 0.001) and the potential to characterize associations with age (r = 0.554, p = 0.021).\u0000<strong>Conclusion:</strong> We have introduced a novel, VSASL-based cerebral microvascular pulsatility technique, which may facilitate investigation of cognitive disorders where damage to the microvasculature has been implicated.","PeriodicalId":501358,"journal":{"name":"medRxiv - Radiology and Imaging","volume":"22 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-06-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141503297","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-06-23DOI: 10.1101/2024.06.22.24309234
Steven Squires, Michelle Harvie, Anthony Howell, D Gareth Evans, Susan M Astley
Objectives: High mammographic density (MD) and excess weight are associated with increased risk of breast cancer. Weight loss interventions could reduce risk, but classically defined percentage density measures may not reflect this due to disproportionate loss of breast fat. We investigate an artificial intelligence-based density method, reporting density changes in 46 women enrolled in a weight-loss study in a family history breast cancer clinic, using a volumetric density method as a comparison. Methods: We analysed data from women who had weight recorded and mammograms taken at the start and end of the 12-month weight intervention study. MD was assessed at both time points using a deep learning model, pVAS, trained on expert estimates of percent density, and VolparaTM density software. Results: The Spearman rank correlation between reduction in weight and change in density was 0.17 (-0.13 to 0.43) for pVAS and 0.59 (0.36 to 0.75) for Volpara volumetric percent density. Conclusions: pVAS percent density measurements were not significantly affected by change in weight. Percent density measured with Volpara increased as weight decreased, driven by changes in fat volume. Advances in knowledge: The effect of weight change on pVAS mammographic density predictions has not previously been published.
目的：乳腺 X 线照相密度（MD）高和体重超重与乳腺癌风险增加有关。减肥干预措施可以降低风险，但由于乳房脂肪的不成比例损失，经典定义的百分比密度测量可能无法反映这一点。我们研究了一种基于人工智能的密度方法，报告了在家族史乳腺癌诊所参加减肥研究的 46 名妇女的密度变化，并使用体积密度方法作为对比。研究方法我们分析了在为期 12 个月的体重干预研究开始和结束时记录体重和进行乳房 X 光检查的妇女的数据。使用深度学习模型 pVAS 和 VolparaTM 密度软件对两个时间点的 MD 进行评估。结果显示体重下降与密度变化之间的斯皮尔曼等级相关性为：pVAS 为 0.17（-0.13 至 0.43），Volpara 容积百分比密度为 0.59（0.36 至 0.75）。结论：体重变化对 pVAS 百分密度测定的影响不大。Volpara 测量的百分比密度随着体重的下降而增加，这是脂肪体积变化的结果。知识进步：体重变化对 pVAS 乳房 X 线照相术密度预测的影响以前从未发表过。
{"title":"Mammographic density assessed using deep learning in women at high risk of developing breast cancer: the effect of weight change on density","authors":"Steven Squires, Michelle Harvie, Anthony Howell, D Gareth Evans, Susan M Astley","doi":"10.1101/2024.06.22.24309234","DOIUrl":"https://doi.org/10.1101/2024.06.22.24309234","url":null,"abstract":"Objectives: High mammographic density (MD) and excess weight are associated with increased risk of breast cancer. Weight loss interventions could reduce risk, but classically defined percentage density measures may not reflect this due to disproportionate loss of breast fat. We investigate an artificial intelligence-based density method, reporting density changes in 46 women enrolled in a weight-loss study in a family history breast cancer clinic, using a volumetric density method as a comparison. Methods: We analysed data from women who had weight recorded and mammograms taken at the start and end of the 12-month weight intervention study. MD was assessed at both time points using a deep learning model, pVAS, trained on expert estimates of percent density, and VolparaTM density software. Results: The Spearman rank correlation between reduction in weight and change in density was 0.17 (-0.13 to 0.43) for pVAS and 0.59 (0.36 to 0.75) for Volpara volumetric percent density. Conclusions: pVAS percent density measurements were not significantly affected by change in weight. Percent density measured with Volpara increased as weight decreased, driven by changes in fat volume. Advances in knowledge: The effect of weight change on pVAS mammographic density predictions has not previously been published.","PeriodicalId":501358,"journal":{"name":"medRxiv - Radiology and Imaging","volume":"22 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-06-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141503298","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-06-22DOI: 10.1101/2024.06.21.24309265
Ji Woong Kim, Aisha Urooj Khan, Imon Banerjee
Background: Vision Transformer (ViT) and Convolutional Neural Networks (CNNs) each possess distinct strengths in medical imaging: ViT excels in capturing long-range dependencies through self-attention, while CNNs are adept at extracting local features via spatial convolution filters. While ViTs might struggle with capturing detailed local spatial information critical for tasks like anomaly detection in medical imaging, shallow CNNs often fail to effectively abstract global context. Objective: This study aims to explore and evaluate hybrid architectures that integrate ViT and CNN to lever- age their complementary strengths for enhanced performance in medical vision tasks, such as segmentation, classification, and prediction. Methods: Following PRISMA guideline, a systematic review was conducted on 28 articles published between 2020 and 2023. These articles proposed hybrid ViT-CNN architectures specifically for medical imaging tasks in radiology. The review focused on analyzing architectural variations, merging strategies between ViT and CNN, innovative applications of ViT, and efficiency metrics including parameters, inference time (GFlops), and performance benchmarks. Results: The review identified that integrating ViT and CNN can help mitigate the limitations of each architecture, offering comprehensive solutions that combine global context understanding with precise local feature extraction. We benchmarked the articles based on architectural variations, merging strategies, innovative uses of ViT, and efficiency metrics (number of parameters, inference time (GFlops), performance). Conclusion: By synthesizing current literature, this review defines fundamental concepts of hybrid vision transformers and highlights emerging trends in the field. It provides a clear direction for future research aimed at optimizing the integration of ViT and CNN for effective utilization in medical imaging, contributing to advancements in diagnostic accuracy and image analysis.
{"title":"Systematic Review of Hybrid Vision Transformer Architectures for Radiological Image Analysis","authors":"Ji Woong Kim, Aisha Urooj Khan, Imon Banerjee","doi":"10.1101/2024.06.21.24309265","DOIUrl":"https://doi.org/10.1101/2024.06.21.24309265","url":null,"abstract":"Background: Vision Transformer (ViT) and Convolutional Neural Networks (CNNs) each possess distinct strengths in medical imaging: ViT excels in capturing long-range dependencies through self-attention, while CNNs are adept at extracting local features via spatial convolution filters. While ViTs might struggle with capturing detailed local spatial information critical for tasks like anomaly detection in medical imaging, shallow CNNs often fail to effectively abstract global context.\u0000Objective: This study aims to explore and evaluate hybrid architectures that integrate ViT and CNN to lever- age their complementary strengths for enhanced performance in medical vision tasks, such as segmentation, classification, and prediction.\u0000Methods: Following PRISMA guideline, a systematic review was conducted on 28 articles published between 2020 and 2023. These articles proposed hybrid ViT-CNN architectures specifically for medical imaging tasks in radiology. The review focused on analyzing architectural variations, merging strategies between ViT and CNN, innovative applications of ViT, and efficiency metrics including parameters, inference time (GFlops), and performance benchmarks.\u0000Results: The review identified that integrating ViT and CNN can help mitigate the limitations of each architecture, offering comprehensive solutions that combine global context understanding with precise local feature extraction. We benchmarked the articles based on architectural variations, merging strategies, innovative uses of ViT, and efficiency metrics (number of parameters, inference time (GFlops), performance).\u0000Conclusion: By synthesizing current literature, this review defines fundamental concepts of hybrid vision transformers and highlights emerging trends in the field. It provides a clear direction for future research aimed at optimizing the integration of ViT and CNN for effective utilization in medical imaging, contributing to advancements in diagnostic accuracy and image analysis.","PeriodicalId":501358,"journal":{"name":"medRxiv - Radiology and Imaging","volume":"81 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-06-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141503299","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-06-20DOI: 10.1101/2024.06.19.24309026
Ines Horvat-Menih, Jonathan R Birchall, Maria J Zamora-Morales, Alice Bebb, Joshua Kaggie, Frank Riemer, Andrew B Gill, Andrew N Priest, Marta Wylot, Iosif A Mendichovszky, Anne Y Warren, James O Jones, James N Armitage, Thomas J Mitchell, Grant D Stewart, Mary A McLean, Ferdia A Gallagher
Background: Accurate non-invasive subtyping of localised kidney tumours is an unmet clinical question in uro-oncology. Differentiation of benign renal oncocytomas (RO) from malignant chromophobe renal cell carcinomas (chRCC) is not possible using conventional imaging. Despite the importance of renal function for sodium regulation, little is known about sodium handling in kidney tumours. Purpose: Here we used non-invasive sodium MRI (23Na-MRI) to quantify sodium concentration and relaxation dynamics across a range of different kidney tumour subtypes and have correlated these findings with imaging surrogates for perfusion, hypoxia, and cellularity. Materials and Methods: Between January and April 2023, patients with localised renal masses were prospectively recruited and underwent 23Na and proton (1H) MRI at 3T to acquire 3D maps of B1, total sodium concentration (TSC), proton and sodium relaxation rates (R2*), and diffusion weighted imaging (DWI). Statistical analysis included comparison and correlation of quantified imaging parameters across kidney tumour subtypes. Results: Ten patients were included in the final analysis (mean age ± S.D. = 64 ± 8 years; 7:3 male:female ratio) encompassing seven ROs, two chRCCs, two clear cell RCCs (ccRCC), and one papillary RCC (pRCC). The TSC was significantly higher in the ROs compared to the chRCCs: 162 ± 58 mM vs. 71 ± 2 mM (P < 0.05). The mean TSC in ccRCC was 135 ± 59 mM, and 81 mM in pRCC. The 23Na-derived and 1H-derived R2* values showed a weak correlation (Spearman r = 0.17; P = 0.50). There was a significant inverse correlation between TSC and 1H-R2* (Spearman r = -0.39, P < 0.05), but TSC was independent of the DWI-derived imaging parameters. Conclusion: 23Na-MRI detected markedly different sodium concentrations within benign ROs and malignant chRCCs. In addition, the sodium signal inversely correlated with 1H-R2* as a surrogate for hypoxia. Therefore we have shown the feasibility and potential of 23Na-MRI for future research in renal tumours.
{"title":"Quantitative sodium-MRI detects differential sodium content in benign vs. malignant oncocytic renal tumours","authors":"Ines Horvat-Menih, Jonathan R Birchall, Maria J Zamora-Morales, Alice Bebb, Joshua Kaggie, Frank Riemer, Andrew B Gill, Andrew N Priest, Marta Wylot, Iosif A Mendichovszky, Anne Y Warren, James O Jones, James N Armitage, Thomas J Mitchell, Grant D Stewart, Mary A McLean, Ferdia A Gallagher","doi":"10.1101/2024.06.19.24309026","DOIUrl":"https://doi.org/10.1101/2024.06.19.24309026","url":null,"abstract":"Background: Accurate non-invasive subtyping of localised kidney tumours is an unmet clinical question in uro-oncology. Differentiation of benign renal oncocytomas (RO) from malignant chromophobe renal cell carcinomas (chRCC) is not possible using conventional imaging. Despite the importance of renal function for sodium regulation, little is known about sodium handling in kidney tumours. Purpose: Here we used non-invasive sodium MRI (23Na-MRI) to quantify sodium concentration and relaxation dynamics across a range of different kidney tumour subtypes and have correlated these findings with imaging surrogates for perfusion, hypoxia, and cellularity. Materials and Methods: Between January and April 2023, patients with localised renal masses were prospectively recruited and underwent 23Na and proton (1H) MRI at 3T to acquire 3D maps of B1, total sodium concentration (TSC), proton and sodium relaxation rates (R2*), and diffusion weighted imaging (DWI). Statistical analysis included comparison and correlation of quantified imaging parameters across kidney tumour subtypes. Results: Ten patients were included in the final analysis (mean age ± S.D. = 64 ± 8 years; 7:3 male:female ratio) encompassing seven ROs, two chRCCs, two clear cell RCCs (ccRCC), and one papillary RCC (pRCC). The TSC was significantly higher in the ROs compared to the chRCCs: 162 ± 58 mM vs. 71 ± 2 mM (P < 0.05). The mean TSC in ccRCC was 135 ± 59 mM, and 81 mM in pRCC. The 23Na-derived and 1H-derived R2* values showed a weak correlation (Spearman r = 0.17; P = 0.50). There was a significant inverse correlation between TSC and 1H-R2* (Spearman r = -0.39, P < 0.05), but TSC was independent of the DWI-derived imaging parameters. Conclusion: 23Na-MRI detected markedly different sodium concentrations within benign ROs and malignant chRCCs. In addition, the sodium signal inversely correlated with 1H-R2* as a surrogate for hypoxia. Therefore we have shown the feasibility and potential of 23Na-MRI for future research in renal tumours.","PeriodicalId":501358,"journal":{"name":"medRxiv - Radiology and Imaging","volume":"31 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-06-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141503300","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-06-20DOI: 10.1101/2024.06.19.24309154
Mario Tranfa, Maria Petracca, Marcello Moccia, Alessandra Scaravilli, Frederik Barkhof, Vincenzo Brescia Morra, Antonio Carotenuto, Sara Collorone, Andrea Elefante, Fabrizia Falco, Roberta Lanzillo, Luigi Lorenzini, Menno Schoonheim, Ahmed Toosy, Arturo Brunetti, Sirio Cocozza, Mario Quarantelli, Giuseppe Pontillo
Whilst multiple sclerosis (MS) can be conceptualized as a network disorder, brain network analyses are typically dependent on advanced MRI sequences not commonly acquired in clinical practice. Here, we used conventional MRI to assess cross-sectional and longitudinal modifications of structural disconnection and morphometric similarity networks in people with MS (pwMS), along with their relationship with clinical disability. In this longitudinal monocentric study, 3T structural MRI scans of pwMS and healthy controls (HC) were retrospectively analysed. Physical and cognitive disabilities were assessed with the expanded disability status scale (EDSS) and the symbol digit modalities test (SDMT), respectively. Demyelinating lesions were automatically segmented on 3D-T1w and FLAIR images and, based on normative tractography data, the corresponding masks were used to compute pairwise structural disconnection between atlas-defined brain regions (100 cortical and 14 subcortical). Using the Morphometric Inverse Divergence (MIND) method, we built matrices of morphometric similarity between cortical regions based on FreeSurfer surface reconstruction. Using network-based statistics (NBS) and its prediction-based extension NBS-predict, we tested whether subject-level connectomes were associated with disease status, progression, clinical disability, and long-term confirmed disability progression (CDP), independently from global lesion burden and atrophy. The coupling between structural disconnection and morphometric similarity was assessed at different scales. We studied 461 pwMS (age=37.2±10.6 years, F/M=324/137), corresponding to 1235 visits (mean follow-up time=1.9±2.0 years, range=0.1-13.3 years), and 55 HC (age=42.4±15.7 years; F/M=25/30). Long-term clinical follow-up was available for 285 pwMS (mean follow-up time=12.4±2.8 years), 127 of whom (44.6%) exhibited CDP. At baseline, structural disconnection in pwMS was mostly centered around the thalami and cortical sensory and association hubs, while morphometric similarity was extensively disrupted (pFWE<0.01). EDSS was related to fronto-thalamic disconnection (pFWE<0.01) and disrupted morphometric similarity around the left perisylvian cortex (pFWE=0.02), whilst SDMT was associated with cortico-subcortical disconnection in the left hemisphere (pFWE<0.01). Longitudinally, both structural disconnection and morphometric similarity disruption significantly progressed (pFWE=0.04 and pFWE<0.01), correlating with EDSS increase (rho=0.07, p=0.02 and rho=0.11, p<0.001), whilst baseline disconnection predicted long-term CDP with nearly 60% accuracy (p=0.03). On average, structural disconnection and morphometric similarity were positively associated at both the edge (rho=0.18, p<0.001) and node (rho=0.16, p<0.001) levels. Structural disconnection and morphometric similarity networks, as assessed through conventional MRI, are sensitive to MS-related brain damage and its progression. They explain di
{"title":"Mapping Structural Disconnection and Morphometric Similarity Alterations in Multiple Sclerosis","authors":"Mario Tranfa, Maria Petracca, Marcello Moccia, Alessandra Scaravilli, Frederik Barkhof, Vincenzo Brescia Morra, Antonio Carotenuto, Sara Collorone, Andrea Elefante, Fabrizia Falco, Roberta Lanzillo, Luigi Lorenzini, Menno Schoonheim, Ahmed Toosy, Arturo Brunetti, Sirio Cocozza, Mario Quarantelli, Giuseppe Pontillo","doi":"10.1101/2024.06.19.24309154","DOIUrl":"https://doi.org/10.1101/2024.06.19.24309154","url":null,"abstract":"Whilst multiple sclerosis (MS) can be conceptualized as a network disorder, brain network analyses are typically dependent on advanced MRI sequences not commonly acquired in clinical practice. Here, we used conventional MRI to assess cross-sectional and longitudinal modifications of structural disconnection and morphometric similarity networks in people with MS (pwMS), along with their relationship with clinical disability.\u0000In this longitudinal monocentric study, 3T structural MRI scans of pwMS and healthy controls (HC) were retrospectively analysed. Physical and cognitive disabilities were assessed with the expanded disability status scale (EDSS) and the symbol digit modalities test (SDMT), respectively. Demyelinating lesions were automatically segmented on 3D-T1w and FLAIR images and, based on normative tractography data, the corresponding masks were used to compute pairwise structural disconnection between atlas-defined brain regions (100 cortical and 14 subcortical). Using the Morphometric Inverse Divergence (MIND) method, we built matrices of morphometric similarity between cortical regions based on FreeSurfer surface reconstruction. Using network-based statistics (NBS) and its prediction-based extension NBS-predict, we tested whether subject-level connectomes were associated with disease status, progression, clinical disability, and long-term confirmed disability progression (CDP), independently from global lesion burden and atrophy. The coupling between structural disconnection and morphometric similarity was assessed at different scales.\u0000We studied 461 pwMS (age=37.2±10.6 years, F/M=324/137), corresponding to 1235 visits (mean follow-up time=1.9±2.0 years, range=0.1-13.3 years), and 55 HC (age=42.4±15.7 years; F/M=25/30). Long-term clinical follow-up was available for 285 pwMS (mean follow-up time=12.4±2.8 years), 127 of whom (44.6%) exhibited CDP. At baseline, structural disconnection in pwMS was mostly centered around the thalami and cortical sensory and association hubs, while morphometric similarity was extensively disrupted (pFWE<0.01). EDSS was related to fronto-thalamic disconnection (pFWE<0.01) and disrupted morphometric similarity around the left perisylvian cortex (pFWE=0.02), whilst SDMT was associated with cortico-subcortical disconnection in the left hemisphere (pFWE<0.01). Longitudinally, both structural disconnection and morphometric similarity disruption significantly progressed (pFWE=0.04 and pFWE<0.01), correlating with EDSS increase (rho=0.07, p=0.02 and rho=0.11, p<0.001), whilst baseline disconnection predicted long-term CDP with nearly 60% accuracy (p=0.03). On average, structural disconnection and morphometric similarity were positively associated at both the edge (rho=0.18, p<0.001) and node (rho=0.16, p<0.001) levels.\u0000Structural disconnection and morphometric similarity networks, as assessed through conventional MRI, are sensitive to MS-related brain damage and its progression. They explain di","PeriodicalId":501358,"journal":{"name":"medRxiv - Radiology and Imaging","volume":"71 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-06-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141503301","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-06-10DOI: 10.1101/2024.06.09.24308649
KUNAAL DHAWAN, Siddharth S. Nijhawan
Brain MRI scans and chest X-ray imaging are pivotal in diagnosing and managing neurological and respiratory diseases, respectively. Given their importance in diagnosis, the datasets to train the artificial intelligence (AI) models for automated diagnosis remain scarce. As an example, annotated chest X-ray datasets, especially those containing rare or abnormal cases like bacterial pneumonia, are scarce. Conventional dataset collection methods are labor-intensive and costly, exacerbating the data scarcity issue. To overcome these challenges, we propose a specialized Generative Adversarial Network (GAN) architecture for generating synthetic chest X-ray data representing healthy lungs and various pneumonia conditions, including viral and bacterial pneumonia. Additionally, we extended our experiments to brain MRI scans by simply swapping the training dataset and demonstrating the power of our GAN approach across different medical imaging contexts. Our method aims to streamline data collection and labeling processes while addressing privacy concerns associated with patient data. We demonstrate the effectiveness of synthetic data in facilitating the development and evaluation of machine learning algorithms, particularly leveraging an EfficientNet v2 model. Through comprehensive experimentation, we evaluate our approach on both real and synthetic datasets, showcasing the potential of synthetic data augmentation in improving disease classification accuracy across diverse pathological conditions. Indeed, the classifier performance when trained with fake + real data on brain MRI classification task shows highest accuracy at 85.9%. Our findings underscore the promising role of synthetic data in advancing automated diagnosis and treatment planning for pneumonia, other respiratory conditions, and brain pathologies.
脑磁共振成像扫描和胸部 X 射线成像分别是诊断和管理神经系统疾病和呼吸系统疾病的关键。鉴于它们在诊断中的重要性，用于训练自动诊断人工智能（AI）模型的数据集仍然稀缺。举例来说，带注释的胸部 X 光数据集，尤其是包含细菌性肺炎等罕见或异常病例的数据集非常稀缺。传统的数据集收集方法劳动密集且成本高昂，加剧了数据稀缺问题。为了克服这些挑战，我们提出了一种专门的生成对抗网络（GAN）架构，用于生成代表健康肺部和各种肺炎病症（包括病毒性和细菌性肺炎）的合成胸部 X 光数据。此外，我们通过简单地交换训练数据集，将实验扩展到了脑部核磁共振成像扫描，并展示了我们的 GAN 方法在不同医学成像环境下的强大功能。我们的方法旨在简化数据收集和标记过程，同时解决与患者数据相关的隐私问题。我们展示了合成数据在促进机器学习算法的开发和评估方面的有效性，特别是利用 EfficientNet v2 模型。通过全面的实验，我们在真实数据集和合成数据集上评估了我们的方法，展示了合成数据增强在提高不同病理条件下疾病分类准确性方面的潜力。事实上，在脑核磁共振成像分类任务中，使用虚假数据和真实数据训练的分类器准确率最高，达到 85.9%。我们的研究结果凸显了合成数据在推进肺炎、其他呼吸系统疾病和脑部病变的自动诊断和治疗规划方面的巨大潜力。
{"title":"Cross-Modality Synthetic Data Augmentation using GANs: Enhancing Brain MRI and Chest X-ray Classification","authors":"KUNAAL DHAWAN, Siddharth S. Nijhawan","doi":"10.1101/2024.06.09.24308649","DOIUrl":"https://doi.org/10.1101/2024.06.09.24308649","url":null,"abstract":"Brain MRI scans and chest X-ray imaging are pivotal in diagnosing and managing neurological and respiratory diseases, respectively. Given their importance in diagnosis, the datasets to train the artificial intelligence (AI) models for automated diagnosis remain scarce. As an example, annotated chest X-ray datasets, especially those containing rare or abnormal cases like bacterial pneumonia, are scarce. Conventional dataset collection methods are labor-intensive and costly, exacerbating the data scarcity issue. To overcome these challenges, we propose a specialized Generative Adversarial Network (GAN) architecture for generating synthetic chest X-ray data representing healthy lungs and various pneumonia conditions, including viral and bacterial pneumonia. Additionally, we extended our experiments to brain MRI scans by simply swapping the training dataset and demonstrating the power of our GAN approach across different medical imaging contexts. Our method aims to streamline data collection and labeling processes while addressing privacy concerns associated with patient data. We demonstrate the effectiveness of synthetic data in facilitating the development and evaluation of machine learning algorithms, particularly leveraging an EfficientNet v2 model. Through comprehensive experimentation, we evaluate our approach on both real and synthetic datasets, showcasing the potential of synthetic data augmentation in improving disease classification accuracy across diverse pathological conditions. Indeed, the classifier performance when trained with fake + real data on brain MRI classification task shows highest accuracy at 85.9%. Our findings underscore the promising role of synthetic data in advancing automated diagnosis and treatment planning for pneumonia, other respiratory conditions, and brain pathologies.","PeriodicalId":501358,"journal":{"name":"medRxiv - Radiology and Imaging","volume":"142 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-06-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141503330","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-05-31DOI: 10.1101/2024.05.31.24308261
Almudena Fuster-Matanzo, Alfonso Picó Peris, Fuensanta Bellvís Bataller, Ana Jimenez-Pastor, Glen J. Weiss, Luis Martí-Bonmatí, Antonio Lázaro Sánchez, Giuseppe L. Banna, Alfredo Addeo, Ángel Alberich-Bayarri
Background In non-small cell lung cancer (NSCLC), alternative strategies to determine patient oncogene mutation status are essential to overcome some of the drawbacks associated with current methods. We aimed to review the use of radiomics alone or in combination with clinical data and to evaluate the performance of artificial intelligence (AI)-based models on the prediction of oncogene mutation status.
{"title":"Prediction of oncogene mutation status in non-small cell lung cancer: A systematic review and meta-analysis with a special focus on artificial-intelligence-based methods","authors":"Almudena Fuster-Matanzo, Alfonso Picó Peris, Fuensanta Bellvís Bataller, Ana Jimenez-Pastor, Glen J. Weiss, Luis Martí-Bonmatí, Antonio Lázaro Sánchez, Giuseppe L. Banna, Alfredo Addeo, Ángel Alberich-Bayarri","doi":"10.1101/2024.05.31.24308261","DOIUrl":"https://doi.org/10.1101/2024.05.31.24308261","url":null,"abstract":"<strong>Background</strong> In non-small cell lung cancer (NSCLC), alternative strategies to determine patient oncogene mutation status are essential to overcome some of the drawbacks associated with current methods. We aimed to review the use of radiomics alone or in combination with clinical data and to evaluate the performance of artificial intelligence (AI)-based models on the prediction of oncogene mutation status.","PeriodicalId":501358,"journal":{"name":"medRxiv - Radiology and Imaging","volume":"41 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-05-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141253550","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：