Pub Date : 2025-01-20DOI: 10.1016/j.cmpb.2025.108606
Xidong Wu, Mingke Yan, Renqiao Wang, Liping Xie
Background and Objective
Atrial fibrillation (AF) is a significant cause of life-threatening heart disease due to its potential to lead to stroke and heart failure. Although deep learning-assisted diagnosis of AF based on ECG holds significance in clinical settings, it remains unsatisfactory due to insufficient consideration of noise and redundant features. In this work, we propose a novel multiscale feature-enhanced gating network (MFEG Net) for AF diagnosis.
Method
The network integrates multiscale convolution, adaptive feature enhancement (FE), and dynamic temporal processing. The multiscale convolution helps capture global and local information. The FE module consists of a soft-threshold residual shrinkage component, a dilated convolution module, and a Squeeze-and-Excitation (SE) module, eliminating redundant features and emphasizing effective features. The design allows the network to focus on the most relevant AF features, thereby enhancing its robustness and accuracy in the presence of noise and irrelevant information. The dynamic temporal module helps the network learn and recognize the time dependence associated with AF. The novel design endows the model with excellent robustness to cope with random noise in real-world environments.
Result
Compared with the state-of-the-art methods, our model exhibits excellent classification performance with an accuracy of 0.930, an F1 score of 0.883, and remarkable resilience to noise interference on the PhysioNet Challenge 2017 dataset. Moreover, the model was trained on the CinC2017 database and validated on the CPSC2018 database and AFDB database, achieving accuracies of 0.908 and 0.938, respectively.
Conclusion
The excellent classification performance of MFEG Net, coupled with its robustness in processing noisy electrocardiogram signals, makes it a powerful method for automatic atrial fibrillation detection. This method has made significant progress over state-of-the-art methods and may alleviate the burden of manual diagnosis for clinical doctors.
{"title":"Multiscale feature enhanced gating network for atrial fibrillation detection","authors":"Xidong Wu, Mingke Yan, Renqiao Wang, Liping Xie","doi":"10.1016/j.cmpb.2025.108606","DOIUrl":"10.1016/j.cmpb.2025.108606","url":null,"abstract":"<div><h3>Background and Objective</h3><div>Atrial fibrillation (AF) is a significant cause of life-threatening heart disease due to its potential to lead to stroke and heart failure. Although deep learning-assisted diagnosis of AF based on ECG holds significance in clinical settings, it remains unsatisfactory due to insufficient consideration of noise and redundant features. In this work, we propose a novel multiscale feature-enhanced gating network (MFEG Net) for AF diagnosis.</div></div><div><h3>Method</h3><div>The network integrates multiscale convolution, adaptive feature enhancement (FE), and dynamic temporal processing. The multiscale convolution helps capture global and local information. The FE module consists of a soft-threshold residual shrinkage component, a dilated convolution module, and a Squeeze-and-Excitation (SE) module, eliminating redundant features and emphasizing effective features. The design allows the network to focus on the most relevant AF features, thereby enhancing its robustness and accuracy in the presence of noise and irrelevant information. The dynamic temporal module helps the network learn and recognize the time dependence associated with AF. The novel design endows the model with excellent robustness to cope with random noise in real-world environments.</div></div><div><h3>Result</h3><div>Compared with the state-of-the-art methods, our model exhibits excellent classification performance with an accuracy of 0.930, an F1 score of 0.883, and remarkable resilience to noise interference on the PhysioNet Challenge 2017 dataset. Moreover, the model was trained on the CinC2017 database and validated on the CPSC2018 database and AFDB database, achieving accuracies of 0.908 and 0.938, respectively.</div></div><div><h3>Conclusion</h3><div>The excellent classification performance of MFEG Net, coupled with its robustness in processing noisy electrocardiogram signals, makes it a powerful method for automatic atrial fibrillation detection. This method has made significant progress over state-of-the-art methods and may alleviate the burden of manual diagnosis for clinical doctors.</div></div>","PeriodicalId":10624,"journal":{"name":"Computer methods and programs in biomedicine","volume":"261 ","pages":"Article 108606"},"PeriodicalIF":4.9,"publicationDate":"2025-01-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143028018","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Several computational pipelines for biomedical data have been proposed to stratify patients and to predict their prognosis through survival analysis. However, these analyses are usually performed independently, without integrating the information derived from each of them. Clustering of survival data is an underexplored problem, and current approaches are limited for biomedical applications, whose data are usually heterogeneous and multimodal, with poor scalability for high-dimensionality.
Methods
We introduce VAE-Surv, a multimodal computational framework for patients’ stratification and prognosis prediction. VAE-Surv integrates a Variational Autoencoder (VAE), which reduces the high-dimensional space characterizing the molecular data, with a deep survival model, which combines the embedded information with the clinical features. The VAE embedding step prioritizes local coherence within the feature space to detect potential nonlinear relationships among the molecular markers. The latent representation is then exploited to perform K-means clustering. To test the clinical robustness of the algorithm, VAE-Surv was applied to the Genomed4all cohort of Myelodysplastic Syndromes (MDS), comparing the identified subtypes with the World Health Organization (WHO) classification. The survival outcome was compared with the state-of-the-art Cox model and its penalized versions. Finally, to assess the generalizability of the results, the method was also validated on an external MDS cohort.
Results
Tested on 2,043 patients in the GenomMed4All cohort, VAE-Surv achieved a median C-index of 0.78, outperforming classical approaches. In addition, the latent space enhanced the clustering performance compared to a traditional approach that applies the clustering directly to the input data. Compared to the WHO 2016 MDS subtypes, the analysis of the identified clusters showed that the proposed framework can capture existing clinical categorizations while also suggesting novel, data-driven patient groups. Even tested in an external MDS cohort of 2,384 patients, VAE-Surv achieved a good prediction performance (median C-index=0.74), preserving the interpretability of the main clinical and genetic features.
Conclusions
VAE-Surv enables automatic identification of patients’ clusters, while outperforming the traditional CoxPH model in survival prediction tasks at the same time. Applied to MDS use case, the obtained genetic-based clusters exhibit a clear survival stratification, and the application of the clinical information allowed high performance in prognosis prediction.
{"title":"VAE-Surv: A novel approach for genetic-based clustering and prognosis prediction in myelodysplastic syndromes","authors":"Cesare Rollo , Corrado Pancotti , Flavio Sartori , Isabella Caranzano , Saverio D’Amico , Luciana Carota , Francesco Casadei , Giovanni Birolo , Luca Lanino , Elisabetta Sauta , Gianluca Asti , Alessandro Buizza , Mattia Delleani , Elena Zazzetti , Marilena Bicchieri , Giulia Maggioni , Pierre Fenaux , Uwe Platzbecker , Maria Diez-Campelo , Torsten Haferlach , Tiziana Sanavia","doi":"10.1016/j.cmpb.2025.108605","DOIUrl":"10.1016/j.cmpb.2025.108605","url":null,"abstract":"<div><h3>Background and Objectives</h3><div>Several computational pipelines for biomedical data have been proposed to stratify patients and to predict their prognosis through survival analysis. However, these analyses are usually performed independently, without integrating the information derived from each of them. Clustering of survival data is an underexplored problem, and current approaches are limited for biomedical applications, whose data are usually heterogeneous and multimodal, with poor scalability for high-dimensionality.</div></div><div><h3>Methods</h3><div>We introduce VAE-Surv, a multimodal computational framework for patients’ stratification and prognosis prediction. VAE-Surv integrates a Variational Autoencoder (VAE), which reduces the high-dimensional space characterizing the molecular data, with a deep survival model, which combines the embedded information with the clinical features. The VAE embedding step prioritizes local coherence within the feature space to detect potential nonlinear relationships among the molecular markers. The latent representation is then exploited to perform K-means clustering. To test the clinical robustness of the algorithm, VAE-Surv was applied to the Genomed4all cohort of Myelodysplastic Syndromes (MDS), comparing the identified subtypes with the World Health Organization (WHO) classification. The survival outcome was compared with the state-of-the-art Cox model and its penalized versions. Finally, to assess the generalizability of the results, the method was also validated on an external MDS cohort.</div></div><div><h3>Results</h3><div>Tested on 2,043 patients in the GenomMed4All cohort, VAE-Surv achieved a median C-index of 0.78, outperforming classical approaches. In addition, the latent space enhanced the clustering performance compared to a traditional approach that applies the clustering directly to the input data. Compared to the WHO 2016 MDS subtypes, the analysis of the identified clusters showed that the proposed framework can capture existing clinical categorizations while also suggesting novel, data-driven patient groups. Even tested in an external MDS cohort of 2,384 patients, VAE-Surv achieved a good prediction performance (median C-index=0.74), preserving the interpretability of the main clinical and genetic features.</div></div><div><h3>Conclusions</h3><div>VAE-Surv enables automatic identification of patients’ clusters, while outperforming the traditional CoxPH model in survival prediction tasks at the same time. Applied to MDS use case, the obtained genetic-based clusters exhibit a clear survival stratification, and the application of the clinical information allowed high performance in prognosis prediction.</div></div>","PeriodicalId":10624,"journal":{"name":"Computer methods and programs in biomedicine","volume":"261 ","pages":"Article 108605"},"PeriodicalIF":4.9,"publicationDate":"2025-01-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143058374","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-01-17DOI: 10.1016/j.cmpb.2025.108609
Ali Ikhsanul Qauli , Nurul Qashri Mahardika T , Ulfa Latifa Hanum , Frederique Jos Vanheusden , Ki Moo Lim
Background and objective
Using electrophysiological simulations and machine learning to predict drug proarrhythmia risk has gained popularity due to its effectiveness. The leading in silico drug assessment system mainly uses a single biomarker (qNet) to predict proarrhythmia risk, offering good performance and straightforward interpretation. Other advanced classifiers incorporating additional physiological biomarkers provide better predictive capabilities but are less intuitive. Thus, a method that accommodates multiple biomarkers while maintaining interpretability is needed.
Methods
We enhance the current best ordinal logistic regression (OLR) model by adding more physiological biomarkers to overcome its limitations. We also introduce a general torsade metric score (TMS) for multi-biomarker approaches to facilitate easier interpretation. Additionally, a novel ranking algorithm based on a simple multi-criteria decision analysis method is employed to evaluate various classifiers against standard proarrhythmia risk criteria efficiently.
Results
Our proposed method demonstrates that using multiple well-known biomarkers yields better performance than using qNet alone. Some accepted multi-biomarker OLR models do not incorporate qNet yet outperform those that do. Moreover, some ill-performing biomarkers when utilized individually can show improved performance in combination with other biomarkers.
Conclusion
The proposed approach offers an effective way of utilizing multiple biomarkers, including well-known ones, providing practical alternatives for proarrhythmia risk assessment. The interpretability of the accepted models is straightforward, thanks to the TMS thresholds for multi-biomarker OLR models that allow direct evaluation of the classification prediction of individual drugs.
{"title":"Elevating performance and interpretability of in silico classifiers for drug proarrhythmia risk evaluations using multi-biomarker approach with ranking algorithm","authors":"Ali Ikhsanul Qauli , Nurul Qashri Mahardika T , Ulfa Latifa Hanum , Frederique Jos Vanheusden , Ki Moo Lim","doi":"10.1016/j.cmpb.2025.108609","DOIUrl":"10.1016/j.cmpb.2025.108609","url":null,"abstract":"<div><h3>Background and objective</h3><div>Using electrophysiological simulations and machine learning to predict drug proarrhythmia risk has gained popularity due to its effectiveness. The leading <em>in silico</em> drug assessment system mainly uses a single biomarker (qNet) to predict proarrhythmia risk, offering good performance and straightforward interpretation. Other advanced classifiers incorporating additional physiological biomarkers provide better predictive capabilities but are less intuitive. Thus, a method that accommodates multiple biomarkers while maintaining interpretability is needed.</div></div><div><h3>Methods</h3><div>We enhance the current best ordinal logistic regression (OLR) model by adding more physiological biomarkers to overcome its limitations. We also introduce a general torsade metric score (TMS) for multi-biomarker approaches to facilitate easier interpretation. Additionally, a novel ranking algorithm based on a simple multi-criteria decision analysis method is employed to evaluate various classifiers against standard proarrhythmia risk criteria efficiently.</div></div><div><h3>Results</h3><div>Our proposed method demonstrates that using multiple well-known biomarkers yields better performance than using qNet alone. Some accepted multi-biomarker OLR models do not incorporate qNet yet outperform those that do. Moreover, some ill-performing biomarkers when utilized individually can show improved performance in combination with other biomarkers.</div></div><div><h3>Conclusion</h3><div>The proposed approach offers an effective way of utilizing multiple biomarkers, including well-known ones, providing practical alternatives for proarrhythmia risk assessment. The interpretability of the accepted models is straightforward, thanks to the TMS thresholds for multi-biomarker OLR models that allow direct evaluation of the classification prediction of individual drugs.</div></div>","PeriodicalId":10624,"journal":{"name":"Computer methods and programs in biomedicine","volume":"261 ","pages":"Article 108609"},"PeriodicalIF":4.9,"publicationDate":"2025-01-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143028016","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-01-16DOI: 10.1016/j.cmpb.2025.108608
Shidi Miao , Qifan Xuan , Wenjuan Huang , Yuyang Jiang , Mengzhuo Sun , Hongzhuo Qi , Ao Li , Zengyao Liu , Jing Li , Xuemei Ding , Ruitao Wang
Background and objective
Central lymph node metastasis (CLNM) is associated with high recurrence rate and low survival in patients with papillary thyroid carcinoma (PTC). However, there is no satisfactory model to predict CLNM in PTC. This study aimed to integrate PTC deep learning feature based on ultrasound (US) images, fat radiomics features based on computed tomography (CT) images and clinical characteristics to construct a multimodal and multi-region nomogram (MMRN) for predicting the CLNM in PTC.
Methods
We enrolled 661 patients diagnosed with PTC by thyroidectomy from two independent centers. Patients were divided into the primary cohort, internal test cohort (ITC), and external test cohort (ETC), and collected their US images and CT images. Resnet50 was employed to predict the CLNM status of PTC based on US images. Using radiomics feature extraction methods to extract fat radiomics features from CT images. Feature selection was conducted using the least absolute shrinkage and selection operator (LASSO) regression. The predictive performance of the MMRN was evaluated using five-fold cross-validation. We comprehensively evaluated the DLRCN and compared it with five radiologists.
Results
In the ITC and ETC, the area under the curves (AUCs) of MMRN were 0.829 (95 % CI: 0.822, 0.835) and 0.818 (95 % CI: 0.808, 0.828). The calibration curve revealed good predictive accuracy between the actual probability and predicted probability (P > 0.05). Decision curve analysis showed that the MMRN was clinically useful. Under equal specificity or sensitivity, the performance of MMRN increased by 6.5 % or 2.9 % compared to radiologist assessments. The incorporation of fat radiomics features led to significant net reclassification improvement (NRI) and integrated discrimination improvement (IDI) (NRI=0.174, P < 0.05, IDI=0.035, P < 0.05).
Conclusion
The MMRN demonstrated good performance in predicting the CLNM status of PTC, which was comparable to radiologist assessments. The fat radiomics features exhibited supplementary value for predicting CLNM in PTC.
背景与目的：中央淋巴结转移（CLNM）与甲状腺乳头状癌（PTC）患者的高复发率和低生存率相关。然而，目前还没有令人满意的模型来预测PTC的CLNM。本研究旨在整合基于超声（US）图像的PTC深度学习特征、基于计算机断层扫描（CT）图像的脂肪放射组学特征和临床特征，构建预测PTC CLNM的多模态多区域nomogram （MMRN）。方法：我们从两个独立的中心纳入661例经甲状腺切除术诊断为PTC的患者。将患者分为初级队列、内部测试队列（ITC）和外部测试队列（ETC），收集患者的超声图像和CT图像。采用Resnet50基于US图像预测PTC的CLNM状态。利用放射组学特征提取方法从CT图像中提取脂肪放射组学特征。使用最小绝对收缩和选择算子（LASSO）回归进行特征选择。MMRN的预测性能采用五重交叉验证进行评估。我们对DLRCN进行了综合评价，并与5位放射科医生进行了比较。结果：在ITC和ETC中，MMRN曲线下面积（auc）分别为0.829 （95% CI: 0.822, 0.835）和0.818 （95% CI: 0.808, 0.828）。校正曲线显示实际概率与预测概率之间具有较好的预测精度（P < 0.05）。决策曲线分析表明MMRN在临床上是有用的。在相同的特异性或敏感性下，与放射科医生的评估相比，MMRN的表现提高了6.5%或2.9%。纳入脂肪放射组学特征后，净重分类改善（NRI）和综合识别改善（IDI）显著（NRI=0.174, P < 0.05， IDI=0.035, P < 0.05）。结论：MMRN在预测PTC的CLNM状态方面表现良好，与放射科医生的评估相当。脂肪放射组学特征对预测PTC的CLNM具有补充价值。
{"title":"Multi-region nomogram for predicting central lymph node metastasis in papillary thyroid carcinoma using multimodal imaging: A multicenter study","authors":"Shidi Miao , Qifan Xuan , Wenjuan Huang , Yuyang Jiang , Mengzhuo Sun , Hongzhuo Qi , Ao Li , Zengyao Liu , Jing Li , Xuemei Ding , Ruitao Wang","doi":"10.1016/j.cmpb.2025.108608","DOIUrl":"10.1016/j.cmpb.2025.108608","url":null,"abstract":"<div><h3>Background and objective</h3><div>Central lymph node metastasis (CLNM) is associated with high recurrence rate and low survival in patients with papillary thyroid carcinoma (PTC). However, there is no satisfactory model to predict CLNM in PTC. This study aimed to integrate PTC deep learning feature based on ultrasound (US) images, fat radiomics features based on computed tomography (CT) images and clinical characteristics to construct a multimodal and multi-region nomogram (MMRN) for predicting the CLNM in PTC.</div></div><div><h3>Methods</h3><div>We enrolled 661 patients diagnosed with PTC by thyroidectomy from two independent centers. Patients were divided into the primary cohort, internal test cohort (ITC), and external test cohort (ETC), and collected their US images and CT images. Resnet50 was employed to predict the CLNM status of PTC based on US images. Using radiomics feature extraction methods to extract fat radiomics features from CT images. Feature selection was conducted using the least absolute shrinkage and selection operator (LASSO) regression. The predictive performance of the MMRN was evaluated using five-fold cross-validation. We comprehensively evaluated the DLRCN and compared it with five radiologists.</div></div><div><h3>Results</h3><div>In the ITC and ETC, the area under the curves (AUCs) of MMRN were 0.829 (95 % CI: 0.822, 0.835) and 0.818 (95 % CI: 0.808, 0.828). The calibration curve revealed good predictive accuracy between the actual probability and predicted probability (<em>P</em> > 0.05). Decision curve analysis showed that the MMRN was clinically useful. Under equal specificity or sensitivity, the performance of MMRN increased by 6.5 % or 2.9 % compared to radiologist assessments. The incorporation of fat radiomics features led to significant net reclassification improvement (NRI) and integrated discrimination improvement (IDI) (NRI=0.174, <em>P</em> < 0.05, IDI=0.035, <em>P</em> < 0.05).</div></div><div><h3>Conclusion</h3><div>The MMRN demonstrated good performance in predicting the CLNM status of PTC, which was comparable to radiologist assessments. The fat radiomics features exhibited supplementary value for predicting CLNM in PTC.</div></div>","PeriodicalId":10624,"journal":{"name":"Computer methods and programs in biomedicine","volume":"261 ","pages":"Article 108608"},"PeriodicalIF":4.9,"publicationDate":"2025-01-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143001424","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Hypertensive Retinopathy (HR) is a retinal manifestation resulting from persistently elevated blood pressure. Severity grading of HR is essential for patient risk stratification, effective management, progression monitoring, timely intervention, and minimizing the risk of vision impairment. Computer-aided diagnosis and artificial intelligence (AI) systems play vital roles in the diagnosis and grading of HR. Over the years, very limited research has been conducted for the grading of HR. Nevertheless, there are no publicly available datasets for HR grading. Moreover, one of the key challenges observed is high-class imbalance.
Methods:
To address these issues, in this paper, we develop “HRSG: Expert-Annotated Hypertensive Retinopathy Severity Grading” dataset, classifying HR severity into four distinct classes: normal, mild, moderate, and severe. Further, to enhance the grading performance on limited datasets, this paper introduces a novel hybrid architecture that combines the strengths of pretrained ResNet-50 via transfer learning, and a modified Vision Transformer (ViT) architecture enhanced with a combination of global self-attention and locality self-attention mechanisms. The locality self-attention addresses the common issue of a lack of inductive bias in ViT architecture. This architecture effectively captures both local and global contextual information, resulting in a robust and resilient classification model. To overcome class imbalance, Decouple Representation and Classifier (DRC) - based training approach is proposed. This method improves the model’s ability to learn effective features while preserving the original dataset’s distribution, leading to better diagnostic accuracy.
Results:
Performance evaluation results show the competence of the proposed method in accurately grading the severity of HR. The proposed method achieved an average accuracy of 0.9688, sensitivity of 0.9435, specificity of 0.9766, F1-score of 0.9442, and precision of 0.9474. The comparative results indicate that the proposed method outperforms existing HR methods, state-of-the-art CNN models, and baseline pretrained ViT models. Additionally, we compared our method with a CNNViT model, which combines a shallow CNN architecture with 3 convolution blocks consisting of a convolution layer, a batch normalization layer, a max pooling layer, and lightweight ViT architecture, due to limited datasets. In comparison with the CNNViT, the proposed method achieved superior performance, demonstrating its effectiveness.
Conclusion:
The experimental results demonstrate the efficacy of the proposed method in accurately grading HR severity.
{"title":"Severity grading of hypertensive retinopathy using hybrid deep learning architecture","authors":"Supriya Suman , Anil Kumar Tiwari , Shreya Sachan , Kuldeep Singh , Seema Meena , Sakshi Kumar","doi":"10.1016/j.cmpb.2025.108585","DOIUrl":"10.1016/j.cmpb.2025.108585","url":null,"abstract":"<div><h3>Background and Objectives:</h3><div>Hypertensive Retinopathy (HR) is a retinal manifestation resulting from persistently elevated blood pressure. Severity grading of HR is essential for patient risk stratification, effective management, progression monitoring, timely intervention, and minimizing the risk of vision impairment. Computer-aided diagnosis and artificial intelligence (AI) systems play vital roles in the diagnosis and grading of HR. Over the years, very limited research has been conducted for the grading of HR. Nevertheless, there are no publicly available datasets for HR grading. Moreover, one of the key challenges observed is high-class imbalance.</div></div><div><h3>Methods:</h3><div>To address these issues, in this paper, we develop “HRSG: Expert-Annotated Hypertensive Retinopathy Severity Grading” dataset, classifying HR severity into four distinct classes: normal, mild, moderate, and severe. Further, to enhance the grading performance on limited datasets, this paper introduces a novel hybrid architecture that combines the strengths of pretrained ResNet-50 via transfer learning, and a modified Vision Transformer (ViT) architecture enhanced with a combination of global self-attention and locality self-attention mechanisms. The locality self-attention addresses the common issue of a lack of inductive bias in ViT architecture. This architecture effectively captures both local and global contextual information, resulting in a robust and resilient classification model. To overcome class imbalance, Decouple Representation and Classifier (DRC) - based training approach is proposed. This method improves the model’s ability to learn effective features while preserving the original dataset’s distribution, leading to better diagnostic accuracy.</div></div><div><h3>Results:</h3><div>Performance evaluation results show the competence of the proposed method in accurately grading the severity of HR. The proposed method achieved an average accuracy of 0.9688, sensitivity of 0.9435, specificity of 0.9766, F1-score of 0.9442, and precision of 0.9474. The comparative results indicate that the proposed method outperforms existing HR methods, state-of-the-art CNN models, and baseline pretrained ViT models. Additionally, we compared our method with a CNNViT model, which combines a shallow CNN architecture with 3 convolution blocks consisting of a convolution layer, a batch normalization layer, a max pooling layer, and lightweight ViT architecture, due to limited datasets. In comparison with the CNNViT, the proposed method achieved superior performance, demonstrating its effectiveness.</div></div><div><h3>Conclusion:</h3><div>The experimental results demonstrate the efficacy of the proposed method in accurately grading HR severity.</div></div>","PeriodicalId":10624,"journal":{"name":"Computer methods and programs in biomedicine","volume":"261 ","pages":"Article 108585"},"PeriodicalIF":4.9,"publicationDate":"2025-01-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143037476","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-01-14DOI: 10.1016/j.cmpb.2025.108602
Luyuan Chen , Haoyao Cao , Yiming Li , Mao Chen , Tinghui Zheng
Objectives
As is well known, plaque morphology plays an important role in the hemodynamics of stenotic coronary arteries, thus their clinic outcomes. However, so far, there has been no research on how the cross-sectional shape of a stenotic lumen affects its hemodynamics. Therefore, this study aims to explore the impact of plaque cross-sectional shape on coronary hemodynamics under mild or moderate stenosis conditions (diameter stenosis degree ≤50 %).
Methods
A three-dimensional model of the coronary tree was established using CT images of a subject without coronary stenosis. Based on real CT images of patients, six types of plaque cross-sectional morphologies were created at the same location in one main left coronary artery model, controlling for 50 % and 25 % diameter stenosis, respectively. Computational fluid dynamics (CFD) simulations were performed on the six stenosed coronary models and one normal control model under the same boundary conditions. The differences in hemodynamic results among the models were compared.
Results
(1) Type III plaque caused the largest disturbance in the flow field. (2) In type IV plaque, the area with an oscillatory shear index (OSI) >0.1 accounted for 11.18 %. (3) Type V plaque exhibited the most prominent vortex flow lines. (4) Hemodynamic parameters within type VI plaques were most similar to those of normal coronary arteries. (5) Area stenosis better reflects the severity of coronary stenosis.
Conclusion
Different cross-sectional morphologies can lead to abnormalities in different hemodynamic parameters, leading to different clinical outcomes. Especially, type III plaques are most likely to cause vascular wall damage, while type V plaques warrant caution due to the risk of complications such as thrombosis. Considering plaque cross-sectional morphology can provide doctors with more information and theoretical support for diagnosis.
{"title":"Analysis of the hemodynamic impact of coronary plaque morphology in mild coronary artery stenosis","authors":"Luyuan Chen , Haoyao Cao , Yiming Li , Mao Chen , Tinghui Zheng","doi":"10.1016/j.cmpb.2025.108602","DOIUrl":"10.1016/j.cmpb.2025.108602","url":null,"abstract":"<div><h3>Objectives</h3><div>As is well known, plaque morphology plays an important role in the hemodynamics of stenotic coronary arteries, thus their clinic outcomes. However, so far, there has been no research on how the cross-sectional shape of a stenotic lumen affects its hemodynamics. Therefore, this study aims to explore the impact of plaque cross-sectional shape on coronary hemodynamics under mild or moderate stenosis conditions (diameter stenosis degree ≤50 %).</div></div><div><h3>Methods</h3><div>A three-dimensional model of the coronary tree was established using CT images of a subject without coronary stenosis. Based on real CT images of patients, six types of plaque cross-sectional morphologies were created at the same location in one main left coronary artery model, controlling for 50 % and 25 % diameter stenosis, respectively. Computational fluid dynamics (CFD) simulations were performed on the six stenosed coronary models and one normal control model under the same boundary conditions. The differences in hemodynamic results among the models were compared.</div></div><div><h3>Results</h3><div>(1) Type III plaque caused the largest disturbance in the flow field. (2) In type IV plaque, the area with an oscillatory shear index (OSI) >0.1 accounted for 11.18 %. (3) Type V plaque exhibited the most prominent vortex flow lines. (4) Hemodynamic parameters within type VI plaques were most similar to those of normal coronary arteries. (5) Area stenosis better reflects the severity of coronary stenosis.</div></div><div><h3>Conclusion</h3><div>Different cross-sectional morphologies can lead to abnormalities in different hemodynamic parameters, leading to different clinical outcomes. Especially, type III plaques are most likely to cause vascular wall damage, while type V plaques warrant caution due to the risk of complications such as thrombosis. Considering plaque cross-sectional morphology can provide doctors with more information and theoretical support for diagnosis.</div></div>","PeriodicalId":10624,"journal":{"name":"Computer methods and programs in biomedicine","volume":"261 ","pages":"Article 108602"},"PeriodicalIF":4.9,"publicationDate":"2025-01-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143022594","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
: Myeloid-derived suppressor cells (MDSCs) are a crucial and diverse group of cells found in the tumor microenvironment (TME) that facilitate progression, invasion, and metastasis within solid tumors. CD84, a homophilic adhesion molecule expressed on MDSCs, plays a critical role in their accumulation and function within the TME. This study aims to investigate the protein-protein interactions of CD84 using molecular dynamics simulations and to explore potential therapeutic strategies targeting these interactions.
Methods
: Through computational techniques, we generated highly potent mutated CD84 mini-proteins and peptides as antagonists with significantly higher affinity for CD84 to mimic the key features of the IgV-like domain of the protein. Additionally, we engineered an antibody capable of blocking CD84. Binding affinities were assessed using dissociation constant (Kd) calculations.
Results
: Data analysis shows that the Kd values for the designed peptides ranged from 10 to 100 times stronger than those of the natural CD84 interactions, indicating efficient inhibition of CD84 interactions. Additionally, mutagenesis of the Ig-like V domain of CD84 resulted in variants with improved binding stability, with a Gibbs free energy change (ΔΔG) indicating enhanced interaction potential.
Conclusions
: This study provides insights into CD84 interactions and their implications for immunotherapy targeting MDSCs in solid tumors. However, experimental validation is necessary to confirm the findings of this study and evaluate peptide selectivity as potential molecular therapeutics.
{"title":"Targeting CD84 protein on myeloid-derived suppressor cells as a novel immunotherapy in solid tumors","authors":"Saeed Mobini , Milad Chizari , Elham Rismani , Ladan Mafakher , Mohammad Javad Sadrzadeh , Massoud Vosough","doi":"10.1016/j.cmpb.2025.108607","DOIUrl":"10.1016/j.cmpb.2025.108607","url":null,"abstract":"<div><h3>Background and Objective</h3><div>: Myeloid-derived suppressor cells (MDSCs) are a crucial and diverse group of cells found in the tumor microenvironment (TME) that facilitate progression, invasion, and metastasis within solid tumors. CD84, a homophilic adhesion molecule expressed on MDSCs, plays a critical role in their accumulation and function within the TME. This study aims to investigate the protein-protein interactions of CD84 using molecular dynamics simulations and to explore potential therapeutic strategies targeting these interactions.</div></div><div><h3>Methods</h3><div>: Through computational techniques, we generated highly potent mutated CD84 mini-proteins and peptides as antagonists with significantly higher affinity for CD84 to mimic the key features of the IgV-like domain of the protein. Additionally, we engineered an antibody capable of blocking CD84. Binding affinities were assessed using dissociation constant (Kd) calculations.</div></div><div><h3>Results</h3><div>: Data analysis shows that the Kd values for the designed peptides ranged from 10 to 100 times stronger than those of the natural CD84 interactions, indicating efficient inhibition of CD84 interactions. Additionally, mutagenesis of the Ig-like V domain of CD84 resulted in variants with improved binding stability, with a Gibbs free energy change (ΔΔG) indicating enhanced interaction potential.</div></div><div><h3>Conclusions</h3><div>: This study provides insights into CD84 interactions and their implications for immunotherapy targeting MDSCs in solid tumors. However, experimental validation is necessary to confirm the findings of this study and evaluate peptide selectivity as potential molecular therapeutics.</div></div>","PeriodicalId":10624,"journal":{"name":"Computer methods and programs in biomedicine","volume":"261 ","pages":"Article 108607"},"PeriodicalIF":4.9,"publicationDate":"2025-01-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143028094","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Alzheimer's disease (AD), the most prevalent form of dementia, remains enigmatic in its origins despite the widely accepted "amyloid hypothesis," which implicates amyloid-beta peptide aggregates in its pathogenesis and progression. Despite advancements in technology and healthcare, the incidence of AD continues to rise. The traditional drug development process remains time-consuming, often taking years to bring an AD treatment to market. Drug repurposing has emerged as a promising strategy for developing cost-effective and efficient therapeutic options by identifying new uses for existing approved drugs, thus accelerating drug development.
Objectives
This study aimed to examine two key drug repurposing methodologies in general diseases and specifically in AD, which are artificial intelligent (AI) approach and molecular docking approach. In addition, the hybrid approach that integrates AI with molecular docking techniques will be explored too.
Methodology
This study systematically compiled a comprehensive collection of relevant academic articles, scientific papers, and research studies which were published up until November 2024 (as of the writing of this review paper). The final selection of papers was filtered to include studies related to Alzheimer's disease and general diseases, and then categorized into three groups: AI articles, molecular docking articles, and hybrid articles.
Results
As a result, 331 papers were identified that employed AI for drug repurposing in general diseases, and 58 papers focused specifically in AD. For molecular docking in drug repurposing, 588 papers addressed general diseases, while 46 papers were dedicated to AD. The hybrid approach combining AI and molecular docking in drug repurposing has 52 papers for general diseases and 9 for AD. A comparative review was done across the methods, results, strengths, and limitations in those studies. Challenges of drug repurposing in AD are explored and future prospects are proposed.
Discussion and conclusion
Drug repurposing emerges as a compelling and effective strategy within AD research. Both AI and molecular docking methods exhibit significant potential in this domain. AI algorithms yield more precise predictions, thus facilitating the exploration of new therapeutic avenues for existing drugs. Similarly, molecular docking techniques revolutionize drug-target interaction modelling, employing refined algorithms to screen extensive drug databases against specific target proteins. This review offers valuable insights for guiding the utilization of AI, molecular docking, or their hybrid in AD drug repurposing endeavors. The hope is to speed up the timeline of drug discovery which could improve the therapeutic approach to AD.
{"title":"Drug repurposing using artificial intelligence, molecular docking, and hybrid approaches: A comprehensive review in general diseases vs Alzheimer's disease","authors":"Natasha Azeelen Zulhafiz , Teow-Chong Teoh , Ai-Vyrn Chin , Siow-Wee Chang","doi":"10.1016/j.cmpb.2025.108604","DOIUrl":"10.1016/j.cmpb.2025.108604","url":null,"abstract":"<div><h3>Background</h3><div>Alzheimer's disease (AD), the most prevalent form of dementia, remains enigmatic in its origins despite the widely accepted \"amyloid hypothesis,\" which implicates amyloid-beta peptide aggregates in its pathogenesis and progression. Despite advancements in technology and healthcare, the incidence of AD continues to rise. The traditional drug development process remains time-consuming, often taking years to bring an AD treatment to market. Drug repurposing has emerged as a promising strategy for developing cost-effective and efficient therapeutic options by identifying new uses for existing approved drugs, thus accelerating drug development.</div></div><div><h3>Objectives</h3><div>This study aimed to examine two key drug repurposing methodologies in general diseases and specifically in AD, which are artificial intelligent (AI) approach and molecular docking approach. In addition, the hybrid approach that integrates AI with molecular docking techniques will be explored too.</div></div><div><h3>Methodology</h3><div>This study systematically compiled a comprehensive collection of relevant academic articles, scientific papers, and research studies which were published up until November 2024 (as of the writing of this review paper). The final selection of papers was filtered to include studies related to Alzheimer's disease and general diseases, and then categorized into three groups: AI articles, molecular docking articles, and hybrid articles.</div></div><div><h3>Results</h3><div>As a result, 331 papers were identified that employed AI for drug repurposing in general diseases, and 58 papers focused specifically in AD. For molecular docking in drug repurposing, 588 papers addressed general diseases, while 46 papers were dedicated to AD. The hybrid approach combining AI and molecular docking in drug repurposing has 52 papers for general diseases and 9 for AD. A comparative review was done across the methods, results, strengths, and limitations in those studies. Challenges of drug repurposing in AD are explored and future prospects are proposed.</div></div><div><h3>Discussion and conclusion</h3><div>Drug repurposing emerges as a compelling and effective strategy within AD research. Both AI and molecular docking methods exhibit significant potential in this domain. AI algorithms yield more precise predictions, thus facilitating the exploration of new therapeutic avenues for existing drugs. Similarly, molecular docking techniques revolutionize drug-target interaction modelling, employing refined algorithms to screen extensive drug databases against specific target proteins. This review offers valuable insights for guiding the utilization of AI, molecular docking, or their hybrid in AD drug repurposing endeavors. The hope is to speed up the timeline of drug discovery which could improve the therapeutic approach to AD.</div></div>","PeriodicalId":10624,"journal":{"name":"Computer methods and programs in biomedicine","volume":"261 ","pages":"Article 108604"},"PeriodicalIF":4.9,"publicationDate":"2025-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143001439","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-01-13DOI: 10.1016/j.cmpb.2025.108603
Shuai Zhang , Jiali Lv , Jinglan Zhang , Zhe Fan , Bingbing Gu , Bingbing Fan , Chunxia Li , Cheng Wang , Tao Zhang
Background and Objective
Colorectal cancer (CRC) represents a heterogeneous malignancy that has concerned global burden of incidence and mortality. The traditional tumor-node-metastasis staging system has exhibited certain limitations. With the advancement of omics technologies, researchers are directing their focus on developing a more precise multi-omics molecular classification. Therefore, the utilization of unsupervised multi-omics integrative clustering methods in CRC, advocating for the establishment of a comprehensive benchmark with practical guidelines.
Methods
In this study, we obtained CRC multi-omics data, encompassing DNA methylation, gene expression, and protein expression from the cancer genome atlas (TCGA)database. We then generated interrelated CRC multi-omics data with various structures based on realistic multi-omics correlations, and performed a comprehensive evaluation of eight representative methods categorized as early integration, intermediate integration, and late integration using complementary benchmarks for subtype classification accuracy. Lastly, we employed these methods to integrate real-world CRC multi-omics data, survival and differential analysis were used to highlight differences among newly identified multi-omics subtypes.
Results
Through in-depth comparisons, we observed that similarity network fusion (SNF) exhibited exceptional performance in integrating multi-omics data derived from simulations. Additionally, SNF effectively distinguished CRC patients into five subgroups with the highest classification accuracy. Moreover, we found significant survival differences and molecular distinctions among SNF subtypes.
Conclusions
The findings consistently demonstrate that SNF outperforms other methods in CRC multi-omics integrative clustering. The significant survival differences and molecular distinctions among SNF subtypes provide novel insights into the multi-omics perspective on CRC heterogeneity with potential clinical treatment.
{"title":"Benchmarking multi-omics integrative clustering methods for subtype identification in colorectal cancer","authors":"Shuai Zhang , Jiali Lv , Jinglan Zhang , Zhe Fan , Bingbing Gu , Bingbing Fan , Chunxia Li , Cheng Wang , Tao Zhang","doi":"10.1016/j.cmpb.2025.108603","DOIUrl":"10.1016/j.cmpb.2025.108603","url":null,"abstract":"<div><h3>Background and Objective</h3><div>Colorectal cancer (CRC) represents a heterogeneous malignancy that has concerned global burden of incidence and mortality. The traditional tumor-node-metastasis staging system has exhibited certain limitations. With the advancement of omics technologies, researchers are directing their focus on developing a more precise multi-omics molecular classification. Therefore, the utilization of unsupervised multi-omics integrative clustering methods in CRC, advocating for the establishment of a comprehensive benchmark with practical guidelines.</div></div><div><h3>Methods</h3><div>In this study, we obtained CRC multi-omics data, encompassing DNA methylation, gene expression, and protein expression from the cancer genome atlas (TCGA)database. We then generated interrelated CRC multi-omics data with various structures based on realistic multi-omics correlations, and performed a comprehensive evaluation of eight representative methods categorized as early integration, intermediate integration, and late integration using complementary benchmarks for subtype classification accuracy. Lastly, we employed these methods to integrate real-world CRC multi-omics data, survival and differential analysis were used to highlight differences among newly identified multi-omics subtypes.</div></div><div><h3>Results</h3><div>Through in-depth comparisons, we observed that similarity network fusion (SNF) exhibited exceptional performance in integrating multi-omics data derived from simulations. Additionally, SNF effectively distinguished CRC patients into five subgroups with the highest classification accuracy. Moreover, we found significant survival differences and molecular distinctions among SNF subtypes.</div></div><div><h3>Conclusions</h3><div>The findings consistently demonstrate that SNF outperforms other methods in CRC multi-omics integrative clustering. The significant survival differences and molecular distinctions among SNF subtypes provide novel insights into the multi-omics perspective on CRC heterogeneity with potential clinical treatment.</div></div>","PeriodicalId":10624,"journal":{"name":"Computer methods and programs in biomedicine","volume":"261 ","pages":"Article 108603"},"PeriodicalIF":4.9,"publicationDate":"2025-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143000917","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-01-13DOI: 10.1016/j.cmpb.2025.108600
Wenbing Lv , Junyi Peng , Jiaping Hu , Yijun Lu , Zidong Zhou , Hui Xu , Kongzai Xing , Xiaodong Zhang , Lijun Lu
Background and objectives
Accurate prediction of progression in knee osteoarthritis (KOA) is significant for early personalized intervention. Previous methods commonly focused on quantifying features from a specific sub-structure imaged at baseline and resulted in limited performance. We proposed a longitudinal MRI sub-structural texture-guided graph convolution network (LMSST-GCN) for improved KOA progression prediction.
Methods
600 KOA participants from the Osteoarthritis Initiative underwent 3 longitudinal MRI scans at baseline, 12 and 24 months. 3D nnU-net was adopted to segment 32 sub-structures of each knee on both IW and DESS sequences at each time point. 105 radiomics features were extracted from each sub-structure, mRMR was used for feature selection, and only the most representative feature was retained to characterize its texture. Each patient was encoded into a 1D vector with 192 features by concatenating all features from 32 sub-structures on the 2 sequences at the 3 time points. Then a population graph was constructed with each vertex representing each patient and edges determining their connection/similarity. The graph was further fed into EdgeGCN to generate the probability of progression. A clinical model and three kinds of machine-learning models including Support Vector Machine (SVM), Random Forest, and Extreme Gradient Boosting (XGBoost) were also constructed for comparison. Interpretability analysis by using GNNExplainer was conducted to explain the association between specific knee sub-structure and KOA progression.
Results
The proposed LMSST-GCN model and its variants (AUC ≥ 0.82) significantly outperformed the clinical model (AUC ≤ 0.72) and machine learning models (AUC ≤ 0.77, p ≤ 0.05 for all). Model performance benefits from the involvement of more sequences and more time points, the highest AUC of 0.85 was achieved by LMSST-GCN model constructed by using all available information. The interpretability analysis demonstrated that the loss of cartilage and sclerosis of subchondral bone at the tibial medial central region, the injury of lateral meniscus, and abnormal changes of the infrapatellar fat pad are more responsible for progression.
Conclusions
The proposed LMSST-GCN model characterized the texture of all knee sub-structures on longitudinal multi-sequence MRI and identified patients prone to progression in the scenario of vertex classification in a population graph, providing a novel strategy for improved prediction of KOA progression. The code was made publicly available at https://github.com/JunyiPeng-SMU/EdgeGCN.
{"title":"LMSST-GCN: Longitudinal MRI sub-structural texture guided graph convolution network for improved progression prediction of knee osteoarthritis","authors":"Wenbing Lv , Junyi Peng , Jiaping Hu , Yijun Lu , Zidong Zhou , Hui Xu , Kongzai Xing , Xiaodong Zhang , Lijun Lu","doi":"10.1016/j.cmpb.2025.108600","DOIUrl":"10.1016/j.cmpb.2025.108600","url":null,"abstract":"<div><h3>Background and objectives</h3><div>Accurate prediction of progression in knee osteoarthritis (KOA) is significant for early personalized intervention. Previous methods commonly focused on quantifying features from a specific sub-structure imaged at baseline and resulted in limited performance. We proposed a longitudinal MRI sub-structural texture-guided graph convolution network (LMSST-GCN) for improved KOA progression prediction.</div></div><div><h3>Methods</h3><div>600 KOA participants from the Osteoarthritis Initiative underwent 3 longitudinal MRI scans at baseline, 12 and 24 months. 3D nnU-net was adopted to segment 32 sub-structures of each knee on both IW and DESS sequences at each time point. 105 radiomics features were extracted from each sub-structure, mRMR was used for feature selection, and only the most representative feature was retained to characterize its texture. Each patient was encoded into a 1D vector with 192 features by concatenating all features from 32 sub-structures on the 2 sequences at the 3 time points. Then a population graph was constructed with each vertex representing each patient and edges determining their connection/similarity. The graph was further fed into EdgeGCN to generate the probability of progression. A clinical model and three kinds of machine-learning models including Support Vector Machine (SVM), Random Forest, and Extreme Gradient Boosting (XGBoost) were also constructed for comparison. Interpretability analysis by using GNNExplainer was conducted to explain the association between specific knee sub-structure and KOA progression.</div></div><div><h3>Results</h3><div>The proposed LMSST-GCN model and its variants (AUC ≥ 0.82) significantly outperformed the clinical model (AUC ≤ 0.72) and machine learning models (AUC ≤ 0.77, <em>p</em> ≤ 0.05 for all). Model performance benefits from the involvement of more sequences and more time points, the highest AUC of 0.85 was achieved by LMSST-GCN model constructed by using all available information. The interpretability analysis demonstrated that the loss of cartilage and sclerosis of subchondral bone at the tibial medial central region, the injury of lateral meniscus, and abnormal changes of the infrapatellar fat pad are more responsible for progression.</div></div><div><h3>Conclusions</h3><div>The proposed LMSST-GCN model characterized the texture of all knee sub-structures on longitudinal multi-sequence MRI and identified patients prone to progression in the scenario of vertex classification in a population graph, providing a novel strategy for improved prediction of KOA progression. The code was made publicly available at <span><span>https://github.com/JunyiPeng-SMU/EdgeGCN</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":10624,"journal":{"name":"Computer methods and programs in biomedicine","volume":"261 ","pages":"Article 108600"},"PeriodicalIF":4.9,"publicationDate":"2025-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143001403","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号