Yuanzhong Xie
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
Yuanzhong Xie has a total of 49 co-authors. Here are the top co-authors by number of shared publications:
- Shengdong Nie
School of Health Science and Engineering, University of Shanghai for Science and Technology, Shanghai, China
8 shared publications Yang Chen
School of Health Science and Engineering, University of Shanghai for Science and Technology, Shanghai, China
6 shared publicationsRong Sun
School of Health Science and Engineering, University of Shanghai for Science and Technology, No. 516 Jun-Gong Road, Shanghai 200093, China
3 shared publicationsXuewen Hou
School of Health Science and Engineering University of Shanghai for Science and Technology Shanghai China
3 shared publicationsYifeng Yang
Department of Medical Imaging, Huadong Hospital, Fudan University, Shanghai, China
3 shared publicationsLiangyun Hu
Center for Functional Neurosurgery, Ruijin Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai, China
2 shared publications
Latest Publications
- Open Access
- 4 February 2025
- Published by Frontiers Media SA in Frontiers in Aging Neuroscience
- Vol. 17, 1510192
Identification of Parkinson’s disease using MRI and genetic data from the PPMI cohort: an improved machine learning fusion approach
Objective: This study aim to leverage advanced machine learning techniques to develop and validate novel MRI imaging features and single nucleotide polymorphism (SNP) gene data fusion methodologies to enhance the early identification and diagnosis of Parkinson’s disease (PD).Methods: We leveraged a comprehensive dataset from the Parkinson’s Progression Markers Initiative (PPMI), which includes high-resolution neuroimaging data, genetic single-nucleotide polymorphism (SNP) profiles, and detailed clinical information from individuals with early-stage PD and healthy controls. Two multi-modal fusion strategies were used: feature-level fusion, where we employed a hybrid feature selection algorithm combining Fisher discriminant analysis, an ensemble Lasso (EnLasso) method, and partial least squares (PLS) regression to identify and integrate the most informative features from neuroimaging and genetic data; and decision-level fusion, where we developed an adaptive ensemble stacking (AE_Stacking) model to synergistically integrate the predictions from multiple base classifiers trained on individual modalities.Results: The AE_Stacking model achieving the highest average balanced accuracy of 95.36% and an area under the receiver operating characteristic curve (AUC) of 0.974, significantly outperforming feature-level fusion and single-modal models (p < 0.05). Furthermore, by analyzing the features selected across multiple iterations of our models, we identified stable brain region features [lh 6r (FD) and rh 46 (GI)] and key genetic markers (rs356181 and rs2736990 SNPs within the SNCA gene region; rs213202 SNP within the VPS52 gene region), highlighting their potential as reliable early diagnostic indicators for the disease.Conclusion: The AE_Stacking model, trained on MRI and genetic data, demonstrates potential in distinguishing individuals with PD. Our findings enhance understanding of the disease and advance us toward the goal of precision medicine for neurodegenerative disorder.
- 4 February 2025
- Published by Frontiers Media SA in Frontiers in Aging Neuroscience
- Vol. 17, 1510192
Open AccessIdentification of Parkinson’s disease using MRI and genetic data from the PPMI cohort: an improved machine learning fusion approach
Objective: This study aim to leverage advanced machine learning techniques to develop and validate novel MRI imaging features and single nucleotide polymorphism (SNP) gene data fusion methodologies to enhance the early identification and diagnosis of Parkinson’s disease (PD).Methods: We leveraged a comprehensive dataset from the Parkinson’s Progression Markers Initiative (PPMI), which includes high-resolution neuroimaging data, genetic single-nucleotide polymorphism (SNP) profiles, and detailed clinical information from individuals with early-stage PD and healthy controls. Two multi-modal fusion strategies were used: feature-level fusion, where we employed a hybrid feature selection algorithm combining Fisher discriminant analysis, an ensemble Lasso (EnLasso) method, and partial least squares (PLS) regression to identify and integrate the most informative features from neuroimaging and genetic data; and decision-level fusion, where we developed an adaptive ensemble stacking (AE_Stacking) model to synergistically integrate the predictions from multiple base classifiers trained on individual modalities.Results: The AE_Stacking model achieving the highest average balanced accuracy of 95.36% and an area under the receiver operating characteristic curve (AUC) of 0.974, significantly outperforming feature-level fusion and single-modal models (p < 0.05). Furthermore, by analyzing the features selected across multiple iterations of our models, we identified stable brain region features [lh 6r (FD) and rh 46 (GI)] and key genetic markers (rs356181 and rs2736990 SNPs within the SNCA gene region; rs213202 SNP within the VPS52 gene region), highlighting their potential as reliable early diagnostic indicators for the disease.Conclusion: The AE_Stacking model, trained on MRI and genetic data, demonstrates potential in distinguishing individuals with PD. Our findings enhance understanding of the disease and advance us toward the goal of precision medicine for neurodegenerative disorder.
- 10 September 2024
- Published by Springer Nature in Journal of Imaging Informatics in Medicine
- Vol. 38 (2) , 1062-1075
Sex-Specific Imaging Biomarkers for Parkinson’s Disease Diagnosis: A Machine Learning Analysis
This study aimed to identify sex-specific imaging biomarkers for Parkinson’s disease (PD) based on multiple MRI morphological features by using machine learning methods. Participants were categorized into female and male subgroups, and various structural morphological features were extracted. An ensemble Lasso (EnLasso) method was employed to identify a stable optimal feature subset for each sex-based subgroup. Eight typical classifiers were adopted to construct classification models for PD and HC, respectively, to validate whether models specific to sex subgroups could bolster the precision of PD identification. Finally, statistical analysis and correlation tests were carried out on significant brain region features to identify potential sex-specific imaging biomarkers. The best model (MLP) based on the female subgroup and male subgroup achieved average classification accuracy of 92.83% and 92.11%, respectively, which were better than that of the model based on the overall samples (86.88%) and the overall model incorporating gender factor (87.52%). In addition, the most discriminative feature of PD among males was the lh 6r (FD), but among females, it was the lh PreS (GI). The findings indicate that the sex-specific PD diagnosis model yields a significantly higher classification performance compared to previous models that included all participants. Additionally, the male subgroup exhibited a greater number of brain region changes than the female subgroup, suggesting sex-specific differences in PD risk markers. This study underscore the importance of stratifying data by sex and offer insights into sex-specific variations in PD phenotypes, which could aid in the development of precise and personalized diagnostic approaches in the early stages of the disease.
- 10 September 2024
- Published by Springer Nature in Journal of Imaging Informatics in Medicine
- Vol. 38 (2) , 1062-1075
Sex-Specific Imaging Biomarkers for Parkinson’s Disease Diagnosis: A Machine Learning Analysis
This study aimed to identify sex-specific imaging biomarkers for Parkinson’s disease (PD) based on multiple MRI morphological features by using machine learning methods. Participants were categorized into female and male subgroups, and various structural morphological features were extracted. An ensemble Lasso (EnLasso) method was employed to identify a stable optimal feature subset for each sex-based subgroup. Eight typical classifiers were adopted to construct classification models for PD and HC, respectively, to validate whether models specific to sex subgroups could bolster the precision of PD identification. Finally, statistical analysis and correlation tests were carried out on significant brain region features to identify potential sex-specific imaging biomarkers. The best model (MLP) based on the female subgroup and male subgroup achieved average classification accuracy of 92.83% and 92.11%, respectively, which were better than that of the model based on the overall samples (86.88%) and the overall model incorporating gender factor (87.52%). In addition, the most discriminative feature of PD among males was the lh 6r (FD), but among females, it was the lh PreS (GI). The findings indicate that the sex-specific PD diagnosis model yields a significantly higher classification performance compared to previous models that included all participants. Additionally, the male subgroup exhibited a greater number of brain region changes than the female subgroup, suggesting sex-specific differences in PD risk markers. This study underscore the importance of stratifying data by sex and offer insights into sex-specific variations in PD phenotypes, which could aid in the development of precise and personalized diagnostic approaches in the early stages of the disease.
- 18 December 2023
- Published by Wiley in Medical Physics
- Vol. 51 (4) , 2759-2771
Segmentation of lung nodules based on a refined segmentation network
Background: Accurate segmentation of lung nodules is of great significance for early screening and diagnosis of lung cancer.Purpose: However, the heterogeneity of lung nodules and the similarities between them and other lung tissues make it difficult to accurately segment these nodules. As regards the use of deep learning to segment lung nodules, convolutional neural networks would gradually lead to errors accumulating at the network layer due to the presence of multiple upsampling and downsampling layers, resulting in poor segmentation results.Methods: In this study, we developed a refined segmentation network (RS‐Net) for lung nodule segmentation to solve this problem. Accordingly, the proposed RS‐Net was first used to locate the core region of the lung nodules and to gradually refine the segmentation results of the core region. In addition, to solve the problem of misdetection of small‐sized nodules owing to the imbalance of positive and negative samples, we devised an average dice‐loss function computed on nodule level. By calculating the loss of each nodule sample to measure the overall loss, the network can address the misdetection problem of lung nodules with smaller diameters more efficiently.Results: Our method was evaluated based on 1055 lung nodules from Lung Image Database Consortium data and a set of 120 lung nodules collected from Shanghai Chest Hospital for additional validation. The segmentation dice coefficients of RS‐Net on these two datasets were 85.90% and 81.13%, respectively. The analysis of the segmentation effect of different properties and sizes of nodules indicates that RS‐Net yields a stable segmentation effect.Conclusions: The results show that the segmentation strategy based on gradual refinement can considerably improve the segmentation of lung nodules.
- 18 December 2023
- Published by Wiley in Medical Physics
- Vol. 51 (4) , 2759-2771
Segmentation of lung nodules based on a refined segmentation network
Background: Accurate segmentation of lung nodules is of great significance for early screening and diagnosis of lung cancer.Purpose: However, the heterogeneity of lung nodules and the similarities between them and other lung tissues make it difficult to accurately segment these nodules. As regards the use of deep learning to segment lung nodules, convolutional neural networks would gradually lead to errors accumulating at the network layer due to the presence of multiple upsampling and downsampling layers, resulting in poor segmentation results.Methods: In this study, we developed a refined segmentation network (RS‐Net) for lung nodule segmentation to solve this problem. Accordingly, the proposed RS‐Net was first used to locate the core region of the lung nodules and to gradually refine the segmentation results of the core region. In addition, to solve the problem of misdetection of small‐sized nodules owing to the imbalance of positive and negative samples, we devised an average dice‐loss function computed on nodule level. By calculating the loss of each nodule sample to measure the overall loss, the network can address the misdetection problem of lung nodules with smaller diameters more efficiently.Results: Our method was evaluated based on 1055 lung nodules from Lung Image Database Consortium data and a set of 120 lung nodules collected from Shanghai Chest Hospital for additional validation. The segmentation dice coefficients of RS‐Net on these two datasets were 85.90% and 81.13%, respectively. The analysis of the segmentation effect of different properties and sizes of nodules indicates that RS‐Net yields a stable segmentation effect.Conclusions: The results show that the segmentation strategy based on gradual refinement can considerably improve the segmentation of lung nodules.
- 6 September 2023
- Published by Elsevier in Computer Methods and Programs in Biomedicine
- Vol. 242, 107804
Molecular-subtype guided automatic invasive breast cancer grading using dynamic contrast-enhanced MRI
- 6 September 2023
- Published by Elsevier in Computer Methods and Programs in Biomedicine
- Vol. 242, 107804
Molecular-subtype guided automatic invasive breast cancer grading using dynamic contrast-enhanced MRI
- 30 May 2023
- Published by Springer Nature in Journal of Digital Imaging
- Vol. 36 (4) , 1553-1564
Weakly Supervised Breast Lesion Detection in Dynamic Contrast-Enhanced MRI
Currently, obtaining accurate medical annotations requires high labor and time effort, which largely limits the development of supervised learning-based tumor detection tasks. In this work, we investigated a weakly supervised learning model for detecting breast lesions in dynamic contrast-enhanced MRI (DCE-MRI) with only image-level labels. Two hundred fifty-four normal and 398 abnormal cases with pathologically confirmed lesions were retrospectively enrolled into the breast dataset, which was divided into the training set (80%), validation set (10%), and testing set (10%) at the patient level. First, the second image series S2 after the injection of a contrast agent was acquired from the 3.0-T, T1-weighted dynamic enhanced MR imaging sequences. Second, a feature pyramid network (FPN) with convolutional block attention module (CBAM) was proposed to extract multi-scale feature maps of the modified classification network VGG16. Then, initial location information was obtained from the heatmaps generated using the layer class activation mapping algorithm (Layer-CAM). Finally, the detection results of breast lesion were refined by the conditional random field (CRF). Accuracy, sensitivity, specificity, and area under the receiver operating characteristic (ROC) curve (AUC) were utilized for evaluation of image-level classification. Average precision (AP) was estimated for breast lesion localization. Delong's test was used to compare the AUCs of different models for significance. The proposed model was effective with accuracy of 95.2%, sensitivity of 91.6%, specificity of 99.2%, and AUC of 0.986. The AP for breast lesion detection was 84.1% using weakly supervised learning. Weakly supervised learning based on FPN combined with Layer-CAM facilitated automatic detection of breast lesion.
- 30 May 2023
- Published by Springer Nature in Journal of Digital Imaging
- Vol. 36 (4) , 1553-1564
Weakly Supervised Breast Lesion Detection in Dynamic Contrast-Enhanced MRI
Currently, obtaining accurate medical annotations requires high labor and time effort, which largely limits the development of supervised learning-based tumor detection tasks. In this work, we investigated a weakly supervised learning model for detecting breast lesions in dynamic contrast-enhanced MRI (DCE-MRI) with only image-level labels. Two hundred fifty-four normal and 398 abnormal cases with pathologically confirmed lesions were retrospectively enrolled into the breast dataset, which was divided into the training set (80%), validation set (10%), and testing set (10%) at the patient level. First, the second image series S2 after the injection of a contrast agent was acquired from the 3.0-T, T1-weighted dynamic enhanced MR imaging sequences. Second, a feature pyramid network (FPN) with convolutional block attention module (CBAM) was proposed to extract multi-scale feature maps of the modified classification network VGG16. Then, initial location information was obtained from the heatmaps generated using the layer class activation mapping algorithm (Layer-CAM). Finally, the detection results of breast lesion were refined by the conditional random field (CRF). Accuracy, sensitivity, specificity, and area under the receiver operating characteristic (ROC) curve (AUC) were utilized for evaluation of image-level classification. Average precision (AP) was estimated for breast lesion localization. Delong's test was used to compare the AUCs of different models for significance. The proposed model was effective with accuracy of 95.2%, sensitivity of 91.6%, specificity of 99.2%, and AUC of 0.986. The AP for breast lesion detection was 84.1% using weakly supervised learning. Weakly supervised learning based on FPN combined with Layer-CAM facilitated automatic detection of breast lesion.
- Open Access
- 24 December 2022
- Published by SAGE Publications in Journal of X-Ray Science and Technology
- Vol. 31 (2) , 223-235
Ischemic stroke subtyping method combining convolutional neural network and radiomics
BACKGROUND: Cardiogenic embolism (CE) and large-artery atherosclerosis embolism (LAA) are the two most common ischemic stroke (IS) subtypes. OBJECTIVE: In order to assist doctors in the precise diagnosis and treatment of patients, this study proposed an IS subtyping method combining convolutional neural networks (CNN) and radiomics. METHODS: Firstly, brain embolism regions were segmented from the computed tomography angiography (CTA) images, and radiomics features were extracted; Secondly, the extracted radiomics features were optimized with the L2 norm, and the feature selection was performed by combining random forest; then, the CNN Cap-UNet was built to extract the deep learning features of the last layer of the network; Finally, combining the selected radiomics features and deep learning features, 9 small-sample classifiers were trained respectively to build and select the optimal IS subtyping classification model. RESULTS: The experimental data include CTA images of 82 IS patients diagnosed and treated in Shanghai Sixth People’s Hospital. The AUC value and accuracy of the optimal subtyping model based on the Adaboost classifier are 0.9018 and 0.8929, respectively. CONCLUSION: The experimental results show that the proposed method can effectively predict the subtype of IS and has potential to assist doctors in making timely and accurate diagnoses of IS patients.
- 24 December 2022
- Published by SAGE Publications in Journal of X-Ray Science and Technology
- Vol. 31 (2) , 223-235
Open AccessIschemic stroke subtyping method combining convolutional neural network and radiomics
BACKGROUND: Cardiogenic embolism (CE) and large-artery atherosclerosis embolism (LAA) are the two most common ischemic stroke (IS) subtypes. OBJECTIVE: In order to assist doctors in the precise diagnosis and treatment of patients, this study proposed an IS subtyping method combining convolutional neural networks (CNN) and radiomics. METHODS: Firstly, brain embolism regions were segmented from the computed tomography angiography (CTA) images, and radiomics features were extracted; Secondly, the extracted radiomics features were optimized with the L2 norm, and the feature selection was performed by combining random forest; then, the CNN Cap-UNet was built to extract the deep learning features of the last layer of the network; Finally, combining the selected radiomics features and deep learning features, 9 small-sample classifiers were trained respectively to build and select the optimal IS subtyping classification model. RESULTS: The experimental data include CTA images of 82 IS patients diagnosed and treated in Shanghai Sixth People’s Hospital. The AUC value and accuracy of the optimal subtyping model based on the Adaboost classifier are 0.9018 and 0.8929, respectively. CONCLUSION: The experimental results show that the proposed method can effectively predict the subtype of IS and has potential to assist doctors in making timely and accurate diagnoses of IS patients.
- 21 November 2022
- Published by Zhejiang University Press in Journal of Zhejiang University-SCIENCE B
- Vol. 23 (11) , 957-967
A method for distinguishing benign and malignant pulmonary nodules based on 3D dual path network aided by K-means clustering analysis
概要 目的 为了提高肺癌早期诊断的准确性,本文使用机器学习,可以有效地帮助放射科医生区分肺结节的良恶性。 创新点 基于三维双路径网络(3DDPN)辅助K均值聚类分析区分良恶性肺结节,类别分析可以有效地表示良恶性肺结节的多种潜在亚型。 方法 在这项研究中,我们提出了一种基于3DDPN并辅以聚类分析来识别良恶性肺结节的新分类方案。首先,根据四位放射科医生的标注结果,从计算机断层扫描(CT)图像中截取以肺结节为中心,尺寸为64×64×64的像素单元;并训练pre-3D DPN模型提取卷积神经网络(CNN)特征。随后,采用随机森林特征选择算法滤除不相关的特征,并采用K均值聚类算法生成聚类标签。最后,使用具有新聚类标签的数据训练3D DPN对肺结节进行良恶性分类。 结果 使用肺影像数据联盟-影像数据库资源计划(LIDC-IDRI)数据库中的966个肺结节进行实验验证,最终所提方法的分类准确率、敏感度、特异度及接受者操作特性曲线(ROC)下面积(AUC)分别达92.86%、94.44%、91.94%及96.43%。此外,从上海胸科医院(SCH)收集了67个结节进行临床验证,获得的准确率为86.57%。 结论 本文所提出的方法可以准确地区分良恶性结节,可作为肺结节良恶性诊断的计算机辅助方法。
- 21 November 2022
- Published by Zhejiang University Press in Journal of Zhejiang University-SCIENCE B
- Vol. 23 (11) , 957-967
A method for distinguishing benign and malignant pulmonary nodules based on 3D dual path network aided by K-means clustering analysis
概要 目的 为了提高肺癌早期诊断的准确性,本文使用机器学习,可以有效地帮助放射科医生区分肺结节的良恶性。 创新点 基于三维双路径网络(3DDPN)辅助K均值聚类分析区分良恶性肺结节,类别分析可以有效地表示良恶性肺结节的多种潜在亚型。 方法 在这项研究中,我们提出了一种基于3DDPN并辅以聚类分析来识别良恶性肺结节的新分类方案。首先,根据四位放射科医生的标注结果,从计算机断层扫描(CT)图像中截取以肺结节为中心,尺寸为64×64×64的像素单元;并训练pre-3D DPN模型提取卷积神经网络(CNN)特征。随后,采用随机森林特征选择算法滤除不相关的特征,并采用K均值聚类算法生成聚类标签。最后,使用具有新聚类标签的数据训练3D DPN对肺结节进行良恶性分类。 结果 使用肺影像数据联盟-影像数据库资源计划(LIDC-IDRI)数据库中的966个肺结节进行实验验证,最终所提方法的分类准确率、敏感度、特异度及接受者操作特性曲线(ROC)下面积(AUC)分别达92.86%、94.44%、91.94%及96.43%。此外,从上海胸科医院(SCH)收集了67个结节进行临床验证,获得的准确率为86.57%。 结论 本文所提出的方法可以准确地区分良恶性结节,可作为肺结节良恶性诊断的计算机辅助方法。
- Open Access
- 10 March 2022
- Published by Hindawi Limited in Concepts in Magnetic Resonance Part B
- Vol. 2022, 1-10
Study of Correlation between MRI Morphology of Primary Tumor and Extramural Vascular Invasion in Rectal Cancer
Introduction. In clinical settings, nodular projection (NP) and cord sign (CS) at the tumor edge and irregular nodules (IN) in the mesorectum often appeared together with extramural vascular invasion (EMVI). We aim to evaluate the diagnostic efficiency of the MRI morphology of primary tumor in predicting EMVI in patients with rectal cancer (RC). Methods. This retrospective study included 156 patients with RC. Clinical and imaging factors including NP at the primary tumor's edge, CS at the primary tumor's edge, maximal extramural depth (EMD), IN in the mesorectum, growth pattern, tumor length, range of rectal wall invaded (RRWI) by tumor, peritoneal reflex invasion by surgery, and pathology-proven local node involvement (PLN) were evaluated. Then, ROC curve was drawn to depict the meaningful indicators in multivariate analyses. Results. There were 53 (34%) patients with pathological extramural venous invasion (pEMVI). Among the clinical and imaging factors evaluated, NP, CS, IN, EMD, PLN, differentiation, and peritoneal reflex invasion were significantly associated with pEMVI. NP and PLN were independent predictors of EMVI. Areas under the ROC curve (AUC) of NP for prediction of EMVI was 0.82 (95% CI, 0.74-0.90), with a sensitivity of 73.58%, a specificity of 90.29%, a positive predictive value (PPV) of 75.59%, a negative predictive value (NPV) of 86.92%, and an accuracy of 84.62%, respectively. Conclusions. Patients with primary tumor with EMVI usually showed NP and CS. NP was an independent predictor of EMVI and helpful for the diagnosis of EMVI in RC patients.
- 10 March 2022
- Published by Hindawi Limited in Concepts in Magnetic Resonance Part B
- Vol. 2022, 1-10
Open AccessStudy of Correlation between MRI Morphology of Primary Tumor and Extramural Vascular Invasion in Rectal Cancer
Introduction. In clinical settings, nodular projection (NP) and cord sign (CS) at the tumor edge and irregular nodules (IN) in the mesorectum often appeared together with extramural vascular invasion (EMVI). We aim to evaluate the diagnostic efficiency of the MRI morphology of primary tumor in predicting EMVI in patients with rectal cancer (RC). Methods. This retrospective study included 156 patients with RC. Clinical and imaging factors including NP at the primary tumor's edge, CS at the primary tumor's edge, maximal extramural depth (EMD), IN in the mesorectum, growth pattern, tumor length, range of rectal wall invaded (RRWI) by tumor, peritoneal reflex invasion by surgery, and pathology-proven local node involvement (PLN) were evaluated. Then, ROC curve was drawn to depict the meaningful indicators in multivariate analyses. Results. There were 53 (34%) patients with pathological extramural venous invasion (pEMVI). Among the clinical and imaging factors evaluated, NP, CS, IN, EMD, PLN, differentiation, and peritoneal reflex invasion were significantly associated with pEMVI. NP and PLN were independent predictors of EMVI. Areas under the ROC curve (AUC) of NP for prediction of EMVI was 0.82 (95% CI, 0.74-0.90), with a sensitivity of 73.58%, a specificity of 90.29%, a positive predictive value (PPV) of 75.59%, a negative predictive value (NPV) of 86.92%, and an accuracy of 84.62%, respectively. Conclusions. Patients with primary tumor with EMVI usually showed NP and CS. NP was an independent predictor of EMVI and helpful for the diagnosis of EMVI in RC patients.
- Open Access
- 1 January 2022
- Published by Wiley in Contrast Media & Molecular Imaging
- Vol. 2022 (1) , 4224701
Comparison of Contrast‐Enhanced Spectral Mammography and Contrast‐Enhanced MRI in Screening Multifocal and Multicentric Lesions in Breast Cancer Patients
Objectives. We aimed to determine the difference between contrast‐enhanced spectral mammography (CESM) and contrast‐enhanced magnetic resonance imaging (CE‐MRI) in detecting multifocal and multicentric breast cancer (MMBC). Methods: This study was conducted among breast cancer patients between July 1, 2017, and May 30, 2021. The sensitivity, specificity, and accuracy of CESM and CE‐MRI in the diagnosis of MMBC were evaluated with pathological results as the gold standard. Results. A total of 188 lesions were detected in 54 patients with MMBC, including 177 breast cancer and 11 benign lesions. Based on CESM and CE‐MRI, 4 false‐positive cases and 3 false‐negative cases and 7 false‐positive cases and 1 false‐negative case, respectively, were found. The accuracy of CESM was higher than that of MRI (96.3% vs 95.7%), and the specificity was higher than that of MRI (63.6% vs 36.4%). There were no significant differences in the sensitivity, specificity, and accuracy for the detection of MMBC between CESM and CE‐MRI (p = 0.500; p = 0.250; p = 0.792). Conclusion. CESM is an effective method for the detection of MMBC, which is consistent with the sensitivity and accuracy of CE‐MRI.
- 1 January 2022
- Published by Wiley in Contrast Media & Molecular Imaging
- Vol. 2022 (1) , 4224701
Open AccessComparison of Contrast‐Enhanced Spectral Mammography and Contrast‐Enhanced MRI in Screening Multifocal and Multicentric Lesions in Breast Cancer Patients
Objectives. We aimed to determine the difference between contrast‐enhanced spectral mammography (CESM) and contrast‐enhanced magnetic resonance imaging (CE‐MRI) in detecting multifocal and multicentric breast cancer (MMBC). Methods: This study was conducted among breast cancer patients between July 1, 2017, and May 30, 2021. The sensitivity, specificity, and accuracy of CESM and CE‐MRI in the diagnosis of MMBC were evaluated with pathological results as the gold standard. Results. A total of 188 lesions were detected in 54 patients with MMBC, including 177 breast cancer and 11 benign lesions. Based on CESM and CE‐MRI, 4 false‐positive cases and 3 false‐negative cases and 7 false‐positive cases and 1 false‐negative case, respectively, were found. The accuracy of CESM was higher than that of MRI (96.3% vs 95.7%), and the specificity was higher than that of MRI (63.6% vs 36.4%). There were no significant differences in the sensitivity, specificity, and accuracy for the detection of MMBC between CESM and CE‐MRI (p = 0.500; p = 0.250; p = 0.792). Conclusion. CESM is an effective method for the detection of MMBC, which is consistent with the sensitivity and accuracy of CE‐MRI.
- 20 October 2021
- Published by Wiley in Journal of Magnetic Resonance Imaging
- Vol. 55 (5) , 1518-1534
Transfer Learning Strategy Based on Unsupervised Learning and Ensemble Learning for Breast Cancer Molecular Subtype Prediction Using Dynamic Contrast‐Enhanced MRI
Background: Imaging‐driven deep learning strategies focus on training from scratch and transfer learning. However, the performance of training from scratch is often impeded by the lack of large‐scale labeled training data. Additionally, owing to the differences between source and target domains, analyzing medical image tasks satisfactorily via transfer learning based on ImageNet is difficult.Purpose: To investigate two transfer learning algorithms for breast cancer molecular subtype prediction (luminal and non‐luminal) based on unsupervised pre‐training and ensemble learning: M_EL and B_EL, using malignant and benign datasets as the source domain, respectively.Study Type: Retrospective.Population: Eight hundred and thirty‐three female patients with histologically confirmed breast lesions (567 benign and 266 malignant cases) were selected. In the 5‐fold cross‐validation, the malignant cohort was randomly divided into 5 subsets to form a training set (80%) and a validation set (20%).Field Strength/Sequence: 3.0 T, dynamic contrast‐enhanced magnetic resonance imaging (DCE‐MRI) using T1‐weighted high‐resolution isotropic volume examination.Assessment: First, three datasets acquired at different times post‐contrast were preprocessed as unlabeled source domains. Second, three baseline networks corresponding to the different MRI post‐contrast phases were built, optimized by a combination of mutual information maximization between high‐ and low‐level representations and prior distribution constraints. Next, the pre‐trained networks were fine‐tuned on the labeled target domain. Finally, prediction results were integrated using weighted voting‐based ensemble learning.Statistical Tests: Mean accuracy, precision, specificity, and area under receiver operating characteristic curve (AUC) were obtained with 5‐fold cross‐validation. P < 0.05 was considered to be statistically significant.Results: Compared with a convolutional long short‐term memory network, pre‐trained VGG‐16, VGG‐19, and DenseNet‐121 from ImageNet, M_EL and B_EL exhibited significantly more optimized prediction performance (specificity: 90.5% and 89.9%; accuracy: 82.6% and 81.1%; precision: 91.2% and 90.9%; AUC: 0.836 and 0.823, respectively).Data Conclusion: Transfer learning based on unsupervised pre‐training may facilitate automatic prediction of breast cancer molecular subtypes.Level of Evidence: 3Technical Efficacy: Stage 2
- 20 October 2021
- Published by Wiley in Journal of Magnetic Resonance Imaging
- Vol. 55 (5) , 1518-1534
Transfer Learning Strategy Based on Unsupervised Learning and Ensemble Learning for Breast Cancer Molecular Subtype Prediction Using Dynamic Contrast‐Enhanced MRI
Background: Imaging‐driven deep learning strategies focus on training from scratch and transfer learning. However, the performance of training from scratch is often impeded by the lack of large‐scale labeled training data. Additionally, owing to the differences between source and target domains, analyzing medical image tasks satisfactorily via transfer learning based on ImageNet is difficult.Purpose: To investigate two transfer learning algorithms for breast cancer molecular subtype prediction (luminal and non‐luminal) based on unsupervised pre‐training and ensemble learning: M_EL and B_EL, using malignant and benign datasets as the source domain, respectively.Study Type: Retrospective.Population: Eight hundred and thirty‐three female patients with histologically confirmed breast lesions (567 benign and 266 malignant cases) were selected. In the 5‐fold cross‐validation, the malignant cohort was randomly divided into 5 subsets to form a training set (80%) and a validation set (20%).Field Strength/Sequence: 3.0 T, dynamic contrast‐enhanced magnetic resonance imaging (DCE‐MRI) using T1‐weighted high‐resolution isotropic volume examination.Assessment: First, three datasets acquired at different times post‐contrast were preprocessed as unlabeled source domains. Second, three baseline networks corresponding to the different MRI post‐contrast phases were built, optimized by a combination of mutual information maximization between high‐ and low‐level representations and prior distribution constraints. Next, the pre‐trained networks were fine‐tuned on the labeled target domain. Finally, prediction results were integrated using weighted voting‐based ensemble learning.Statistical Tests: Mean accuracy, precision, specificity, and area under receiver operating characteristic curve (AUC) were obtained with 5‐fold cross‐validation. P < 0.05 was considered to be statistically significant.Results: Compared with a convolutional long short‐term memory network, pre‐trained VGG‐16, VGG‐19, and DenseNet‐121 from ImageNet, M_EL and B_EL exhibited significantly more optimized prediction performance (specificity: 90.5% and 89.9%; accuracy: 82.6% and 81.1%; precision: 91.2% and 90.9%; AUC: 0.836 and 0.823, respectively).Data Conclusion: Transfer learning based on unsupervised pre‐training may facilitate automatic prediction of breast cancer molecular subtypes.Level of Evidence: 3Technical Efficacy: Stage 2