Informatics in Medicine Unlocked最新文献_第4页

Prediction of the onset of the RSV epidemic with meteorological data using deep neural networks 利用深度神经网络预测RSV流行的气象数据

Q1 Medicine

Informatics in Medicine Unlocked

Pub Date : 2025-01-01 DOI: 10.1016/j.imu.2025.101659

Kazuo Yonekura , Miya Nishio , Momoko Kashiwado , Takuya Naruto , Masaaki Mori

Background

Respiratory syncytial virus (RSV) is a contagious virus that infects nearly all children by the age of two and is a leading cause of hospitalization and mortality among young children. Despite the recent approval of RSV vaccines for elderly and pregnant individuals, immune prophylaxis remains essential for pediatric cases. In Japan, the typical RSV season has shifted, making timely prediction crucial for effective clinical intervention.

Objective

This study aims to predict the onset of RSV epidemics in Japan using meteorological data, based on the hypothesis that meteorological data affect the spread of RSV.

Methods

We collected weekly RSV case counts from the Japanese National Institute of Infectious Diseases and daily meteorological data from the Japan Meteorological Agency for the period 2012–2023. Using aggregated weather features (mean, max, min), we constructed a binary classification task to identify the onset of RSV spread. Machine learning models including a support vector machine (SVM), XGBoost, and a deep neural network (DNN) were evaluated.

Results

The DNN outperformed other models, achieving the highest F1 score (0.71) and recall (0.83), particularly with a 3-week-ahead prediction horizon. The model demonstrated early detection capability across multiple prefectures, although performance varied geographically, with lower F1 scores in some northern regions.

Conclusion

Meteorological data can be effectively utilized to predict the onset of RSV epidemics in Japan. The proposed DNN-based model offers a promising tool for supporting timely prophylactic measures, although further refinement and integration of additional factors are needed to improve generalizability.

呼吸道合胞病毒（RSV）是一种传染性病毒，几乎所有2岁的儿童都会感染，是幼儿住院和死亡的主要原因。尽管最近批准了针对老年人和孕妇的呼吸道合胞病毒疫苗，但免疫预防对于儿科病例仍然至关重要。在日本，典型的RSV季节已经转移，因此及时预测对于有效的临床干预至关重要。目的基于气象资料影响RSV传播的假设，利用气象资料预测日本RSV流行的发生。方法收集2012-2023年日本国立传染病研究所每周RSV病例数和日本气象厅每日气象资料。利用汇总的天气特征（平均值、最大值、最小值），我们构建了一个二元分类任务来确定RSV传播的开始。评估了包括支持向量机（SVM）、XGBoost和深度神经网络（DNN）在内的机器学习模型。结果DNN优于其他模型，F1得分最高（0.71），召回率最高（0.83），特别是在3周的预测范围内。该模型显示了多个县的早期检测能力，尽管表现在地理上有所不同，一些北部地区的F1得分较低。结论气象资料可有效预测日本RSV疫情的发生。提出的基于dnn的模型为支持及时预防措施提供了一个有前途的工具，尽管需要进一步改进和整合其他因素以提高通用性。

{"title":"Prediction of the onset of the RSV epidemic with meteorological data using deep neural networks","authors":"Kazuo Yonekura , Miya Nishio , Momoko Kashiwado , Takuya Naruto , Masaaki Mori","doi":"10.1016/j.imu.2025.101659","DOIUrl":"10.1016/j.imu.2025.101659","url":null,"abstract":"<div><h3>Background</h3><div>Respiratory syncytial virus (RSV) is a contagious virus that infects nearly all children by the age of two and is a leading cause of hospitalization and mortality among young children. Despite the recent approval of RSV vaccines for elderly and pregnant individuals, immune prophylaxis remains essential for pediatric cases. In Japan, the typical RSV season has shifted, making timely prediction crucial for effective clinical intervention.</div></div><div><h3>Objective</h3><div>This study aims to predict the onset of RSV epidemics in Japan using meteorological data, based on the hypothesis that meteorological data affect the spread of RSV.</div></div><div><h3>Methods</h3><div>We collected weekly RSV case counts from the Japanese National Institute of Infectious Diseases and daily meteorological data from the Japan Meteorological Agency for the period 2012–2023. Using aggregated weather features (mean, max, min), we constructed a binary classification task to identify the onset of RSV spread. Machine learning models including a support vector machine (SVM), XGBoost, and a deep neural network (DNN) were evaluated.</div></div><div><h3>Results</h3><div>The DNN outperformed other models, achieving the highest F1 score (0.71) and recall (0.83), particularly with a 3-week-ahead prediction horizon. The model demonstrated early detection capability across multiple prefectures, although performance varied geographically, with lower F1 scores in some northern regions.</div></div><div><h3>Conclusion</h3><div>Meteorological data can be effectively utilized to predict the onset of RSV epidemics in Japan. The proposed DNN-based model offers a promising tool for supporting timely prophylactic measures, although further refinement and integration of additional factors are needed to improve generalizability.</div></div>","PeriodicalId":13953,"journal":{"name":"Informatics in Medicine Unlocked","volume":"57 ","pages":"Article 101659"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144240370","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}

引用次数: 0

Enhancing diabetes risk prediction: A comparative evaluation of bagging, boosting, and ensemble classifiers with SMOTE oversampling 增强糖尿病风险预测：用SMOTE过采样对bagging、boosting和ensemble分类器进行比较评价

Q1 Medicine

Informatics in Medicine Unlocked

Pub Date : 2025-01-01 DOI: 10.1016/j.imu.2025.101661

Rabia Asif , Darshana Upadhyay , Marzia Zaman , Srini Sampalli

Diabetes is a major global health concern, with millions of individuals at risk of developing this chronic condition. Early prediction and intervention are essential for effective diabetes management. This study explores advanced machine learning techniques, specifically bagging, boosting, and ensemble methods to improve diabetes risk prediction. Using three diverse datasets, namely, the Centers for Disease Control and Prevention (CDC) Diabetes Health Indicators dataset, the Early Stage Diabetes Risk Prediction System (ESDRP) dataset, and the PIMA Indian Diabetes dataset are utilized to evaluate the adaptability and robustness of the proposed models. Our approach addresses critical gaps in existing research, including the handling of highly imbalanced datasets through the Synthetic Minority Over-sampling Technique (SMOTE), the necessity of feature selection, and the underutilization of the CDC dataset in diabetes studies. We find that applying SMOTE to the CDC dataset significantly enhances model performance, with the CATBoost algorithm achieving an accuracy of 91 %. For the ESRPS dataset, ensemble methods demonstrate even stronger results, achieving 98 % accuracy using the top five features. This study not only contributes to the development of more accurate predictive models for diabetes risk but also provides insights into enhancing the robustness of machine learning methods in healthcare.

糖尿病是一个主要的全球健康问题，数百万人有患这种慢性疾病的风险。早期预测和干预对于有效的糖尿病管理至关重要。本研究探索先进的机器学习技术，特别是bagging、boosting和ensemble方法来改善糖尿病风险预测。使用三个不同的数据集，即疾病控制和预防中心（CDC）糖尿病健康指标数据集，早期糖尿病风险预测系统（ESDRP）数据集和PIMA印度糖尿病数据集来评估所提出模型的适应性和鲁棒性。我们的方法解决了现有研究中的关键空白，包括通过合成少数派过采样技术（SMOTE）处理高度不平衡的数据集，特征选择的必要性，以及糖尿病研究中CDC数据集的未充分利用。我们发现，将SMOTE应用于CDC数据集可以显著提高模型性能，CATBoost算法的准确率达到91%。对于ESRPS数据集，集成方法显示出更强的结果，使用前五个特征达到98%的准确率。这项研究不仅有助于开发更准确的糖尿病风险预测模型，而且还为增强医疗保健领域机器学习方法的鲁棒性提供了见解。

{"title":"Enhancing diabetes risk prediction: A comparative evaluation of bagging, boosting, and ensemble classifiers with SMOTE oversampling","authors":"Rabia Asif , Darshana Upadhyay , Marzia Zaman , Srini Sampalli","doi":"10.1016/j.imu.2025.101661","DOIUrl":"10.1016/j.imu.2025.101661","url":null,"abstract":"<div><div>Diabetes is a major global health concern, with millions of individuals at risk of developing this chronic condition. Early prediction and intervention are essential for effective diabetes management. This study explores advanced machine learning techniques, specifically bagging, boosting, and ensemble methods to improve diabetes risk prediction. Using three diverse datasets, namely, the Centers for Disease Control and Prevention (CDC) Diabetes Health Indicators dataset, the Early Stage Diabetes Risk Prediction System (ESDRP) dataset, and the PIMA Indian Diabetes dataset are utilized to evaluate the adaptability and robustness of the proposed models. Our approach addresses critical gaps in existing research, including the handling of highly imbalanced datasets through the Synthetic Minority Over-sampling Technique (SMOTE), the necessity of feature selection, and the underutilization of the CDC dataset in diabetes studies. We find that applying SMOTE to the CDC dataset significantly enhances model performance, with the CATBoost algorithm achieving an accuracy of 91 %. For the ESRPS dataset, ensemble methods demonstrate even stronger results, achieving 98 % accuracy using the top five features. This study not only contributes to the development of more accurate predictive models for diabetes risk but also provides insights into enhancing the robustness of machine learning methods in healthcare.</div></div>","PeriodicalId":13953,"journal":{"name":"Informatics in Medicine Unlocked","volume":"57 ","pages":"Article 101661"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144535756","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}

引用次数: 0

Challenges in AI-driven multi-omics data analysis for Oncology: Addressing dimensionality, sparsity, transparency and ethical considerations 人工智能驱动的肿瘤学多组学数据分析的挑战：解决维度、稀疏性、透明度和伦理考虑

Q1 Medicine

Informatics in Medicine Unlocked

Pub Date : 2025-01-01 DOI: 10.1016/j.imu.2025.101679

Maryem Ouhmouk , Shakuntala Baichoo , Mounia Abik

Artificial intelligence, particularly deep learning, is becoming increasingly prominent in multi-omics research, especially since traditional statistical models struggle to handle the complexity and high dimensionality of such data. By effectively combining different types of omics data, AI techniques can unveil hidden connections, detect biomarkers, and improve disease prediction through the integration of multi-omics layers and modalities, which can lead to significant advancements in precision medicine. In this review, we gathered published methods of deep learning-based multi-omics integration specialized in oncology since 2020. We concentrated exclusively on studies utilizing cancer omics data mainly sourced from The Cancer Genome Atlas (TCGA) database. As a result, we identified 32 articles that generally fulfilled the criteria. We studied their techniques and their ability to handle challenges in analyzing multi-omics data, particularly regarding missing data, dimensionality, and processing workflows. We also discuss how well these methods consider explainability, interpretability, and ethical aspects in developing solutions that treat private medical and sensitive information.

From the 32 studies, we can divide deep learning-based multi-omics integration methods into two types: non-generative and generative models. Non-generative approaches, such as feedforward neural networks (FFNs), graph convolutional networks (GCNs), and autoencoders, are designed to extract features and perform classification directly. On the other hand, generative methods such as variational autoencoders (VAEs), generative adversarial networks (GANs), and generative pretrained transformers (GPTs) focus on creating adaptable representations that can be shared across multiple modalities. These methods have advanced the handling of missing data and dimensionality, outperforming traditional approaches. However, most reviewed models remain at the proof-of-concept stage, with limited clinical validation or real-world deployment.

人工智能，特别是深度学习，在多组学研究中变得越来越突出，特别是因为传统的统计模型难以处理此类数据的复杂性和高维性。通过有效结合不同类型的组学数据，人工智能技术可以通过多组学层和模式的整合，揭示隐藏的联系，检测生物标志物，改善疾病预测，这可能会导致精准医疗的重大进步。在这篇综述中，我们收集了自2020年以来发表的基于深度学习的肿瘤学多组学整合方法。我们专注于利用主要来自癌症基因组图谱（TCGA）数据库的癌症组学数据的研究。结果，我们确定了32篇基本符合标准的文章。我们研究了他们的技术和他们处理多组学数据分析挑战的能力，特别是在缺失数据、维度和处理工作流方面。我们还讨论了这些方法在开发处理私人医疗和敏感信息的解决方案时如何很好地考虑可解释性、可解释性和伦理方面。从这32项研究中，我们可以将基于深度学习的多组学集成方法分为非生成模型和生成模型两类。非生成方法，如前馈神经网络（ffn）、图卷积网络（GCNs）和自动编码器，被设计用于提取特征并直接执行分类。另一方面，生成方法，如变分自编码器（VAEs）、生成对抗网络（GANs）和生成预训练变压器（GPTs）专注于创建可跨多种模式共享的自适应表示。这些方法提高了对缺失数据和维度的处理，优于传统方法。然而，大多数被审查的模型仍处于概念验证阶段，缺乏临床验证或实际应用。

{"title":"Challenges in AI-driven multi-omics data analysis for Oncology: Addressing dimensionality, sparsity, transparency and ethical considerations","authors":"Maryem Ouhmouk , Shakuntala Baichoo , Mounia Abik","doi":"10.1016/j.imu.2025.101679","DOIUrl":"10.1016/j.imu.2025.101679","url":null,"abstract":"<div><div>Artificial intelligence, particularly deep learning, is becoming increasingly prominent in multi-omics research, especially since traditional statistical models struggle to handle the complexity and high dimensionality of such data. By effectively combining different types of omics data, AI techniques can unveil hidden connections, detect biomarkers, and improve disease prediction through the integration of multi-omics layers and modalities, which can lead to significant advancements in precision medicine. In this review, we gathered published methods of deep learning-based multi-omics integration specialized in oncology since 2020. We concentrated exclusively on studies utilizing cancer omics data mainly sourced from The Cancer Genome Atlas (TCGA) database. As a result, we identified 32 articles that generally fulfilled the criteria. We studied their techniques and their ability to handle challenges in analyzing multi-omics data, particularly regarding missing data, dimensionality, and processing workflows. We also discuss how well these methods consider explainability, interpretability, and ethical aspects in developing solutions that treat private medical and sensitive information.</div><div>From the 32 studies, we can divide deep learning-based multi-omics integration methods into two types: non-generative and generative models. Non-generative approaches, such as feedforward neural networks (FFNs), graph convolutional networks (GCNs), and autoencoders, are designed to extract features and perform classification directly. On the other hand, generative methods such as variational autoencoders (VAEs), generative adversarial networks (GANs), and generative pretrained transformers (GPTs) focus on creating adaptable representations that can be shared across multiple modalities. These methods have advanced the handling of missing data and dimensionality, outperforming traditional approaches. However, most reviewed models remain at the proof-of-concept stage, with limited clinical validation or real-world deployment.</div></div>","PeriodicalId":13953,"journal":{"name":"Informatics in Medicine Unlocked","volume":"57 ","pages":"Article 101679"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144766898","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}

引用次数: 0

A longitudinal analysis of morphological shape variation of spleen in patients with fontan surgery fontan手术患者脾脏形态变化的纵向分析

Q1 Medicine

Informatics in Medicine Unlocked

Pub Date : 2025-01-01 DOI: 10.1016/j.imu.2025.101665

Varatharajan Nainamalai , Håvard Bjørke Jenssen , Mostafa Rezaeitaleshmahalleh , Djinaud Prophete , Jordan Gosnell , Sarah Khan , Marcus Haw , Jingfeng Jiang , Joseph Vettukattil

Background

Splenic size serves as a surrogate biomarker for predicting portal vein hyper-tension and liver abnormalities in subjects with Fontan Associated Liver Disease (FALD). We analyze the long-term shape variation of the spleen in FALD subjects using morphological shape features of radiomic features.

Methods

We used 154 (84 from computed tomography and 70 from magnetic resonance) image volumes obtained from 36 individuals who underwent stage 3 Fontan procedure and 145 computed tomography images from controls to assess splenomegaly. To understand the splenomegaly, thirteen shape features of the spleen over three 10-year intervals, and variations between controls and FALD subjects were analyzed.

Results

The spleen enlargement was observed in all intervals of the post-surgery period. Also, a significant difference (level of significance α = 0.05, p < α) was observed between the morphological shape features of controls and the Fontan Associated Liver Disease subjects.

Conclusion

Morphological shape features clearly distinguish between controls and subjects after Fontan stage 3 correction.

背景：脾大小可作为预测Fontan相关性肝病（FALD）患者门静脉高压和肝脏异常的替代生物标志物。我们利用放射学特征的形态学特征分析了FALD受试者脾脏的长期形状变化。方法：我们使用36例行3期Fontan手术的患者的154张图像（84张来自计算机断层扫描，70张来自磁共振成像）和对照组的145张计算机断层扫描图像来评估脾肿大。为了了解脾肿大，我们分析了三个10年间隔的13个脾脏形状特征，以及对照组和FALD受试者之间的差异。结果术后各时间段均可见脾脏肿大。同时，显著性差异(显著水平α = 0.05, p <；对照组与Fontan相关性肝病患者的形态特征之间存在α)。结论Fontan 3期矫正后的形态学特征与对照组有明显区别。

{"title":"A longitudinal analysis of morphological shape variation of spleen in patients with fontan surgery","authors":"Varatharajan Nainamalai , Håvard Bjørke Jenssen , Mostafa Rezaeitaleshmahalleh , Djinaud Prophete , Jordan Gosnell , Sarah Khan , Marcus Haw , Jingfeng Jiang , Joseph Vettukattil","doi":"10.1016/j.imu.2025.101665","DOIUrl":"10.1016/j.imu.2025.101665","url":null,"abstract":"<div><h3>Background</h3><div>Splenic size serves as a surrogate biomarker for predicting portal vein hyper-tension and liver abnormalities in subjects with Fontan Associated Liver Disease (FALD). We analyze the long-term shape variation of the spleen in FALD subjects using morphological shape features of radiomic features.</div></div><div><h3>Methods</h3><div>We used 154 (84 from computed tomography and 70 from magnetic resonance) image volumes obtained from 36 individuals who underwent stage 3 Fontan procedure and 145 computed tomography images from controls to assess splenomegaly. To understand the splenomegaly, thirteen shape features of the spleen over three 10-year intervals, and variations between controls and FALD subjects were analyzed.</div></div><div><h3>Results</h3><div>The spleen enlargement was observed in all intervals of the post-surgery period. Also, a significant difference (level of significance <em>α</em> = 0.05, <em>p < α</em>) was observed between the morphological shape features of controls and the Fontan Associated Liver Disease subjects.</div></div><div><h3>Conclusion</h3><div>Morphological shape features clearly distinguish between controls and subjects after Fontan stage 3 correction.</div></div>","PeriodicalId":13953,"journal":{"name":"Informatics in Medicine Unlocked","volume":"57 ","pages":"Article 101665"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144679522","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}

引用次数: 0

TransUNetB: An advanced Transformer–UNet framework for efficient and explainable brain tumor segmentation TransUNetB：一个先进的Transformer-UNet框架，用于高效和可解释的脑肿瘤分割

Q1 Medicine

Informatics in Medicine Unlocked

Pub Date : 2025-01-01 DOI: 10.1016/j.imu.2025.101706

Katura Gania Khushubu , Abdullah Al Masum , Md Habibur Rahman , Shakh Md Shakib Hasan , Md Imranul Hoque Bhuiyan , Mohammad Rasel Mahmud , S.M. Masfequier Rahman Swapno , Abhishek Appaji

This study presents TransUNetB, a hybrid architecture that combines Transformer and UNet for multi-class brain tumor segmentation. This model integrates global context modeling with precise spatial localization. A lightweight Transformer encoder at the bottleneck captures long-range dependencies, while the U-Net's skip pathways preserve fine anatomical details. Additionally, a multi-scale decoder fusion module consolidates features at various resolutions, enhancing the clarity of tumor boundaries in heterogeneous, low-contrast conditions. Our contributions are threefold: (1) a simple, efficient design that integrates bottleneck self-attention with multi-scale fusion for robust ED/TC/ET segmentation; (2) a comprehensive ablation of design choices—attention type, positional encoding, fusion strategy, loss formulation, and patch size—quantifying their impact on accuracy and efficiency; and (3) an explainability analysis using Grad-CAM with quantitative focus/entropy measures to verify that salient regions align with clinical tumor substructures. Evaluated on the BraTS 2020 and BraTS 2021 datasets, TransUNetB achieves a Dice score of 98.90 % and an Intersection over Union (IoU) score of 96.10 %. It outperforms strong CNN and vision-transformer baselines while maintaining a competitive runtime of approximately 63 ms per image. These results suggest that combining global attention with spatially faithful decoding provides a favorable trade-off between accuracy and efficiency for clinical deployment. We also discuss the generalization of our model beyond MRI cohorts, practical constraints in resource-limited settings, and future research avenues, including attention-guided fusion and broader multi-center validation.

本研究提出了一种结合Transformer和UNet的混合架构TransUNetB，用于多类别脑肿瘤分割。该模型将全局上下文建模与精确的空间定位相结合。瓶颈处的轻量级Transformer编码器捕获远程依赖关系，而U-Net的跳过路径保留了精细的解剖细节。此外，一个多尺度解码器融合模块在不同分辨率下整合特征，增强了异质性、低对比度条件下肿瘤边界的清晰度。我们的贡献有三个：(1)一个简单、高效的设计，将瓶颈自关注与多尺度融合相结合，用于稳健的ED/TC/ET分割；(2)综合分析设计选择——注意力类型、位置编码、融合策略、损失公式和补丁大小——量化它们对准确性和效率的影响；(3)使用带有定量焦点/熵测量的Grad-CAM进行可解释性分析，以验证突出区域与临床肿瘤亚结构一致。在BraTS 2020和BraTS 2021数据集上进行评估，TransUNetB的Dice得分为98.90%，IoU得分为96.10%。它优于强大的CNN和视觉转换基线，同时保持每张图像约63毫秒的竞争性运行时间。这些结果表明，将全局注意力与空间忠实解码相结合，在临床部署的准确性和效率之间提供了有利的权衡。我们还讨论了我们的模型在MRI队列之外的推广，资源有限环境下的实际限制，以及未来的研究途径，包括注意力引导融合和更广泛的多中心验证。

{"title":"TransUNetB: An advanced Transformer–UNet framework for efficient and explainable brain tumor segmentation","authors":"Katura Gania Khushubu , Abdullah Al Masum , Md Habibur Rahman , Shakh Md Shakib Hasan , Md Imranul Hoque Bhuiyan , Mohammad Rasel Mahmud , S.M. Masfequier Rahman Swapno , Abhishek Appaji","doi":"10.1016/j.imu.2025.101706","DOIUrl":"10.1016/j.imu.2025.101706","url":null,"abstract":"<div><div>This study presents TransUNetB, a hybrid architecture that combines Transformer and UNet for multi-class brain tumor segmentation. This model integrates global context modeling with precise spatial localization. A lightweight Transformer encoder at the bottleneck captures long-range dependencies, while the U-Net's skip pathways preserve fine anatomical details. Additionally, a multi-scale decoder fusion module consolidates features at various resolutions, enhancing the clarity of tumor boundaries in heterogeneous, low-contrast conditions. Our contributions are threefold: (1) a simple, efficient design that integrates bottleneck self-attention with multi-scale fusion for robust ED/TC/ET segmentation; (2) a comprehensive ablation of design choices—attention type, positional encoding, fusion strategy, loss formulation, and patch size—quantifying their impact on accuracy and efficiency; and (3) an explainability analysis using Grad-CAM with quantitative focus/entropy measures to verify that salient regions align with clinical tumor substructures. Evaluated on the BraTS 2020 and BraTS 2021 datasets, TransUNetB achieves a Dice score of 98.90 % and an Intersection over Union (IoU) score of 96.10 %. It outperforms strong CNN and vision-transformer baselines while maintaining a competitive runtime of approximately 63 ms per image. These results suggest that combining global attention with spatially faithful decoding provides a favorable trade-off between accuracy and efficiency for clinical deployment. We also discuss the generalization of our model beyond MRI cohorts, practical constraints in resource-limited settings, and future research avenues, including attention-guided fusion and broader multi-center validation.</div></div>","PeriodicalId":13953,"journal":{"name":"Informatics in Medicine Unlocked","volume":"59 ","pages":"Article 101706"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145463705","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}

引用次数: 0

Dynamic system of tuberculosis elimination modelling in Indonesia: A study in Kuningan, West Java, Indonesia 印度尼西亚消除结核病模型的动态系统：在印度尼西亚西爪哇库宁安的一项研究

Q1 Medicine

Informatics in Medicine Unlocked

Pub Date : 2025-01-01 DOI: 10.1016/j.imu.2025.101714

Dyah Aryani Perwitasari , Imaniar Noor Faridah , Casnan Casnan , Nourma Nurjanah , Ina Listiana , Hendy Ristiono , Juwita Ramadhani

Tuberculosis (TB) remains a major public health burden in Indonesia, which currently ranks second worldwide in TB incidence. This study aimed to develop and validate a dynamic systems model to simulate TB elimination strategies at the district level in Kuningan, West Java. We conducted a cross-sectional study involving 30 TB patients, complemented with retrospective data from Ciniru Health Center, the national TB surveillance system (SITB), and Minimal Standard of Services (2019–2023). Patient-level data were collected using structured questionnaires on adherence, beliefs, and knowledge, as well as household environmental assessments. A system dynamics model was constructed using Powersim Studio 10 Academic to simulate the transmission and control of TB. Model validation against historical data showed strong predictive accuracy (absolute mean error <5 %). Simulation of two scenarios—Moderate and extreme—revealed substantial differences in outcomes by 2040. Under the Moderate scenario, improvements in lifestyle, nutrition, and adherence reduced the number of TB cases from 82 to 55, and suspects treated decreased from 702 to 614. Under the extreme scenario, further reductions were observed, with TB cases declining to 10 and suspects treated to 612. Key determinants influencing TB burden included nutrition, lifestyle, household environment, and management of drug side effects. This dynamic modeling framework provides valuable insights for localized TB control strategies. It emphasizes that elimination efforts should go beyond biomedical interventions by addressing behavioral, environmental, and psychosocial determinants. The model offers a practical tool to guide district-level planning and supports the national TB elimination agenda.

结核病仍然是印度尼西亚的一个主要公共卫生负担，目前印度尼西亚的结核病发病率在世界上排名第二。本研究旨在开发和验证一个动态系统模型，以模拟西爪哇库宁安地区一级的结核病消除战略。我们进行了一项涉及30名结核病患者的横断面研究，并辅以Ciniru卫生中心、国家结核病监测系统（SITB）和最低服务标准（2019-2023）的回顾性数据。采用结构化问卷收集患者层面的数据，包括依从性、信念和知识，以及家庭环境评估。采用Powersim Studio 10 Academic软件建立系统动力学模型，模拟结核的传播与控制。对历史数据的模型验证显示出很强的预测准确性（绝对平均误差<； 5%）。对中度和极端两种情景的模拟显示，到2040年，结果将存在实质性差异。在中等情景下，生活方式、营养和依从性的改善将结核病病例从82例减少到55例，接受治疗的疑似病例从702例减少到614例。在极端情况下，观察到进一步减少，结核病病例减少到10例，治疗的疑似病例减少到612例。影响结核病负担的关键决定因素包括营养、生活方式、家庭环境和药物副作用的管理。这一动态建模框架为本地化结核病控制策略提供了有价值的见解。它强调消除努力应超越生物医学干预，解决行为、环境和社会心理决定因素。该模式为指导地区一级的规划和支持国家消除结核病议程提供了实用工具。

{"title":"Dynamic system of tuberculosis elimination modelling in Indonesia: A study in Kuningan, West Java, Indonesia","authors":"Dyah Aryani Perwitasari , Imaniar Noor Faridah , Casnan Casnan , Nourma Nurjanah , Ina Listiana , Hendy Ristiono , Juwita Ramadhani","doi":"10.1016/j.imu.2025.101714","DOIUrl":"10.1016/j.imu.2025.101714","url":null,"abstract":"<div><div>Tuberculosis (TB) remains a major public health burden in Indonesia, which currently ranks second worldwide in TB incidence. This study aimed to develop and validate a dynamic systems model to simulate TB elimination strategies at the district level in Kuningan, West Java. We conducted a cross-sectional study involving 30 TB patients, complemented with retrospective data from Ciniru Health Center, the national TB surveillance system (SITB), and Minimal Standard of Services (2019–2023). Patient-level data were collected using structured questionnaires on adherence, beliefs, and knowledge, as well as household environmental assessments. A system dynamics model was constructed using Powersim Studio 10 Academic to simulate the transmission and control of TB. Model validation against historical data showed strong predictive accuracy (absolute mean error <5 %). Simulation of two scenarios—Moderate and extreme—revealed substantial differences in outcomes by 2040. Under the Moderate scenario, improvements in lifestyle, nutrition, and adherence reduced the number of TB cases from 82 to 55, and suspects treated decreased from 702 to 614. Under the extreme scenario, further reductions were observed, with TB cases declining to 10 and suspects treated to 612. Key determinants influencing TB burden included nutrition, lifestyle, household environment, and management of drug side effects. This dynamic modeling framework provides valuable insights for localized TB control strategies. It emphasizes that elimination efforts should go beyond biomedical interventions by addressing behavioral, environmental, and psychosocial determinants. The model offers a practical tool to guide district-level planning and supports the national TB elimination agenda.</div></div>","PeriodicalId":13953,"journal":{"name":"Informatics in Medicine Unlocked","volume":"59 ","pages":"Article 101714"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145619996","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}

引用次数: 0

Explainable deep learning models for HER2 IHC scoring in breast cancer diagnosis 乳腺癌诊断中HER2 IHC评分的可解释深度学习模型

Q1 Medicine

Informatics in Medicine Unlocked

Pub Date : 2025-01-01 DOI: 10.1016/j.imu.2025.101700

Md Serajun Nabi , Mohammad Faizal Ahmad Fauzi , Hezerul Bin Abdul Karim , Phaik-Leng Cheah , Seow-Fan Chiew , Lai-Meng Looi

Accurate and interpretable HER2 IHC scoring is crucial for guiding breast cancer treatment. However, manual evaluation remains inconsistent and subjective. This study proposes a deep learning framework that integrates both a custom Convolutional Neural Network (CNN) and a fine-tuned DenseNet121 model for automated HER2 scoring using the HER-IHC-40x dataset. Preprocessing involves HSV-based patch filtering and expert validation to ensure data relevance. To improve transparency and address the black-box nature of AI models, we employed explainable AI (XAI) techniques. Gradient-weighted Class Activation Mapping (Grad-CAM) and SHapley Additive exPlanations (SHAP) provide visual explanations at the pixel and region levels. These techniques enhance interpretability, ensuring clinical confidence by clearly visualizing and attributing model decisions, particularly in borderline HER2 cases (Class 1+ and 2+), where manual misinterpretations are common. The experimental results show that both CNN and DenseNet121 achieved 93% accuracy with excellent class-wise consistency. CNN, in particular, demonstrated higher prediction confidence and lower training loss, indicating superior calibration. The integration of explainability modules ensures improved clinical transparency and improves trust in AI-driven decision-making. Comparison with the existing literature confirms the strength of the proposed method in predictive capacity and interpretability, contributing to a robust AI-assisted breast cancer diagnosis.

准确和可解释的HER2 IHC评分对于指导乳腺癌治疗至关重要。然而，人工评估仍然是不一致的和主观的。本研究提出了一个深度学习框架，该框架集成了自定义卷积神经网络（CNN）和微调的DenseNet121模型，用于使用HER-IHC-40x数据集自动进行HER2评分。预处理包括基于hsv的补丁过滤和专家验证，以确保数据的相关性。为了提高透明度和解决人工智能模型的黑箱性质，我们采用了可解释的人工智能（XAI）技术。梯度加权类激活映射（Grad-CAM）和SHapley加性解释（SHAP）在像素和区域级别提供视觉解释。这些技术提高了可解释性，通过清晰地可视化和归因模型决定来确保临床信心，特别是在HER2的边缘病例（1+和2+级），在这些病例中，人工误解是常见的。实验结果表明，CNN和DenseNet121都达到了93%的准确率，并且具有优异的分类一致性。尤其是CNN，它的预测置信度更高，训练损失更小，这表明它的校准效果更好。可解释性模块的整合确保了临床透明度的提高，并提高了对人工智能驱动决策的信任。与现有文献的比较证实了所提出方法在预测能力和可解释性方面的优势，有助于实现强大的人工智能辅助乳腺癌诊断。

{"title":"Explainable deep learning models for HER2 IHC scoring in breast cancer diagnosis","authors":"Md Serajun Nabi , Mohammad Faizal Ahmad Fauzi , Hezerul Bin Abdul Karim , Phaik-Leng Cheah , Seow-Fan Chiew , Lai-Meng Looi","doi":"10.1016/j.imu.2025.101700","DOIUrl":"10.1016/j.imu.2025.101700","url":null,"abstract":"<div><div>Accurate and interpretable HER2 IHC scoring is crucial for guiding breast cancer treatment. However, manual evaluation remains inconsistent and subjective. This study proposes a deep learning framework that integrates both a custom Convolutional Neural Network (CNN) and a fine-tuned DenseNet121 model for automated HER2 scoring using the HER-IHC-40x dataset. Preprocessing involves HSV-based patch filtering and expert validation to ensure data relevance. To improve transparency and address the black-box nature of AI models, we employed explainable AI (XAI) techniques. Gradient-weighted Class Activation Mapping (Grad-CAM) and SHapley Additive exPlanations (SHAP) provide visual explanations at the pixel and region levels. These techniques enhance interpretability, ensuring clinical confidence by clearly visualizing and attributing model decisions, particularly in borderline HER2 cases (Class 1+ and 2+), where manual misinterpretations are common. The experimental results show that both CNN and DenseNet121 achieved 93% accuracy with excellent class-wise consistency. CNN, in particular, demonstrated higher prediction confidence and lower training loss, indicating superior calibration. The integration of explainability modules ensures improved clinical transparency and improves trust in AI-driven decision-making. Comparison with the existing literature confirms the strength of the proposed method in predictive capacity and interpretability, contributing to a robust AI-assisted breast cancer diagnosis.</div></div>","PeriodicalId":13953,"journal":{"name":"Informatics in Medicine Unlocked","volume":"58 ","pages":"Article 101700"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145324046","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}

引用次数: 0

Use of natural language processing to predict the diagnoses of patients with gait impairments 使用自然语言处理来预测步态障碍患者的诊断

Q1 Medicine

Informatics in Medicine Unlocked

Pub Date : 2025-01-01 DOI: 10.1016/j.imu.2025.101712

Bastian Widmer , Giusi Moffa , H. Elke Viehweger , Morgan Sangeux

Background

In clinical practice, unstructured medical data stored in electronic health records presents challenges for information extraction. This study aims to evaluate the feasibility of using natural language processing (NLP) techniques on German clinical text, as well as its English version, translated with Google Translate, for automating the diagnosis classification of patients with gait impairments.

Materials and methods

We trained four different types of models on a dataset containing 892 diagnoses to classify patients into four subgroups: unilateral cerebral palsy (CP), bilateral CP, undefined CP, and no CP. The approaches included rule-based methods, term frequency, embeddings, and Bidirectional Encoder Representations from Transformers. We assessed model performance using metrics such as accuracy, precision, recall, and F₁-Measure. To compare the classification results between the original German text and the translated English text, we applied McNemar's test.

Results

All models exhibited strong and comparable performance across the evaluated metrics, except for embedding-based models. There was no statistically significant difference in the misclassification distributions when the models were applied to the German text compared to its English translation.

Conclusions

These findings underscore the potential of NLP to transform medical informatics by accurately and efficiently classifying orthopaedic texts in German. Furthermore, translating diagnoses from German into English is a viable approach, facilitating the use of models trained on English text for multilingual diagnosis classification.

在临床实践中，存储在电子病历中的非结构化医疗数据对信息提取提出了挑战。本研究旨在评估使用自然语言处理（NLP）技术对德语临床文本及其英文版本进行自动化诊断分类的可行性，并使用谷歌Translate进行翻译。材料和方法我们在包含892个诊断的数据集上训练了四种不同类型的模型，将患者分为四个亚组：单侧脑瘫（CP）、双侧脑瘫、未定义脑瘫和无脑瘫。方法包括基于规则的方法、词频、嵌入和来自变压器的双向编码器表示。我们使用准确性、精密度、召回率和F1-Measure等指标评估模型性能。为了比较德语原文和翻译后的英语文本的分类结果，我们使用McNemar检验。结果除了基于嵌入的模型外，所有模型在评估指标中都表现出很强的可比性。当模型应用于德语文本时，与英语文本相比，错误分类分布没有统计学上的显着差异。结论这些发现强调了NLP通过准确有效地分类德语骨科文本来改变医学信息学的潜力。此外，将诊断从德语翻译成英语是一种可行的方法，有助于使用英语文本训练的模型进行多语言诊断分类。

{"title":"Use of natural language processing to predict the diagnoses of patients with gait impairments","authors":"Bastian Widmer , Giusi Moffa , H. Elke Viehweger , Morgan Sangeux","doi":"10.1016/j.imu.2025.101712","DOIUrl":"10.1016/j.imu.2025.101712","url":null,"abstract":"<div><h3>Background</h3><div>In clinical practice, unstructured medical data stored in electronic health records presents challenges for information extraction. This study aims to evaluate the feasibility of using natural language processing (NLP) techniques on German clinical text, as well as its English version, translated with Google Translate, for automating the diagnosis classification of patients with gait impairments.</div></div><div><h3>Materials and methods</h3><div>We trained four different types of models on a dataset containing 892 diagnoses to classify patients into four subgroups: unilateral cerebral palsy (CP), bilateral CP, undefined CP, and no CP. The approaches included rule-based methods, term frequency, embeddings, and Bidirectional Encoder Representations from Transformers. We assessed model performance using metrics such as accuracy, precision, recall, and F<sub>1</sub>-Measure. To compare the classification results between the original German text and the translated English text, we applied McNemar's test.</div></div><div><h3>Results</h3><div>All models exhibited strong and comparable performance across the evaluated metrics, except for embedding-based models. There was no statistically significant difference in the misclassification distributions when the models were applied to the German text compared to its English translation.</div></div><div><h3>Conclusions</h3><div>These findings underscore the potential of NLP to transform medical informatics by accurately and efficiently classifying orthopaedic texts in German. Furthermore, translating diagnoses from German into English is a viable approach, facilitating the use of models trained on English text for multilingual diagnosis classification.</div></div>","PeriodicalId":13953,"journal":{"name":"Informatics in Medicine Unlocked","volume":"59 ","pages":"Article 101712"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145526150","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}

引用次数: 0

Exploring health informaticians' perception of the Saudi health informatics competency framework (SHICF) 探索卫生信息学家对沙特卫生信息学能力框架（SHICF）的看法

Q1 Medicine

Informatics in Medicine Unlocked

Pub Date : 2025-01-01 DOI: 10.1016/j.imu.2025.101709

Khulud Alkadi , Lujain Aljarallah , Hadeel Albedewi

Background

Saudi Arabia has recently made significant progress in health informatics, investing in eHealth infrastructure and training health informaticians. The Saudi Health Informatics Competency Framework (SHICF), established in 2021, ensures health informaticians possess the skills and knowledge to address twenty-first-century healthcare challenges and support Saudi Vision 2030.

Objectives

This study explores practicing health informaticians' perceptions of the current SHICF in Saudi Arabia, aiming to suggest improvements to the framework.

Methods

This qualitative descriptive cross-sectional study used semi-structured interviews, with health informaticians in Riyadh, Saudi Arabia, from academia, public, and private sectors. Interviews included demographics and 20 questions exploring perceptions of SHICF. Inductive thematic analysis identified themes and sub-themes reflecting participants’ views on SHICF.

Results

A total of 10 interviewees were included in the study. Most viewed the current SHICF as comprehensive but requiring clarification. They also suggested tailoring competencies to Saudi Arabian laws, technical standards, and various educational levels.

Conclusion

Our findings highlight the need to refine the Saudi Health Informatics Competency Framework (SHICF) to a more tailored, current version that reflects recent developments in Saudi health informatics.

沙特阿拉伯最近在卫生信息学方面取得了重大进展，投资于电子卫生基础设施和培训卫生信息学家。沙特卫生信息能力框架（SHICF）于2021年建立，确保卫生信息学家拥有应对21世纪卫生保健挑战的技能和知识，并支持沙特2030年愿景。本研究探讨了沙特阿拉伯执业卫生信息学家对当前SHICF的看法，旨在提出改进框架的建议。方法：本定性描述性横断面研究采用半结构化访谈，访谈对象为沙特阿拉伯利雅得的学术界、公共部门和私营部门的卫生信息学家。访谈包括人口统计数据和20个问题，探讨了对shif的看法。归纳性专题分析确定了反映与会者对可持续发展论坛看法的主题和分主题。结果共纳入10名受访者。大多数人认为目前的shf是全面的，但需要澄清。他们还建议根据沙特阿拉伯的法律、技术标准和不同的教育水平来调整能力。结论：我们的研究结果强调需要将沙特卫生信息学能力框架（SHICF）完善为一个更有针对性的最新版本，以反映沙特卫生信息学的最新发展。

{"title":"Exploring health informaticians' perception of the Saudi health informatics competency framework (SHICF)","authors":"Khulud Alkadi , Lujain Aljarallah , Hadeel Albedewi","doi":"10.1016/j.imu.2025.101709","DOIUrl":"10.1016/j.imu.2025.101709","url":null,"abstract":"<div><h3>Background</h3><div>Saudi Arabia has recently made significant progress in health informatics, investing in eHealth infrastructure and training health informaticians. The Saudi Health Informatics Competency Framework (SHICF), established in 2021, ensures health informaticians possess the skills and knowledge to address twenty-first-century healthcare challenges and support Saudi Vision 2030.</div></div><div><h3>Objectives</h3><div>This study explores practicing health informaticians' perceptions of the current SHICF in Saudi Arabia, aiming to suggest improvements to the framework.</div></div><div><h3>Methods</h3><div>This qualitative descriptive cross-sectional study used semi-structured interviews, with health informaticians in Riyadh, Saudi Arabia, from academia, public, and private sectors. Interviews included demographics and 20 questions exploring perceptions of SHICF. Inductive thematic analysis identified themes and sub-themes reflecting participants’ views on SHICF.</div></div><div><h3>Results</h3><div>A total of 10 interviewees were included in the study. Most viewed the current SHICF as comprehensive but requiring clarification. They also suggested tailoring competencies to Saudi Arabian laws, technical standards, and various educational levels.</div></div><div><h3>Conclusion</h3><div>Our findings highlight the need to refine the Saudi Health Informatics Competency Framework (SHICF) to a more tailored, current version that reflects recent developments in Saudi health informatics.</div></div>","PeriodicalId":13953,"journal":{"name":"Informatics in Medicine Unlocked","volume":"59 ","pages":"Article 101709"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145526151","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}

引用次数: 0

Domain specific transfer learning and classifier chains in Alzheimer's disease detection using 3D convolutional neural networks 基于三维卷积神经网络的阿尔茨海默病检测领域特定迁移学习和分类器链

Q1 Medicine

Informatics in Medicine Unlocked

Pub Date : 2025-01-01 DOI: 10.1016/j.imu.2025.101711

Lisa-Marie Bente , Luca Himstedt , Tim Kacprowski , Alzheimer's Disease Neuroimaging Initiative

This study examines different configurations of deep convolutional neural networks (CNNs) and the effect of using domain-specific transfer learning for distinguishing Alzheimer's Disease and Mild Cognitive Impairment from normal controls. The data used to train our models was provided by ADNI and included 1118 3D FDG-PET scans in total. We train a binary and a multiclass classifier, as well as chains of binary classifiers, for consecutive multiclass classification. Two chains were trained with different orders: chain A classified cognitively normal (CN) vs. non-CN, followed by Alzheimer's disease (AD) vs. mild cognitive impairment (MCI). Classifier chain B classified AD vs. non-AD first, followed by MCI vs. CN. All classifiers were trained with and without the use of domain-specific transfer learning, using weights from Med3D. All models achieve comparable performance to the state-of-the-art. Classifier chain A even achieved superior performance with an accuracy of 96 %, F1 score of 95 % and AUROC of 99 %. Using domain-specific transfer learning resulted in worse performance among the majority of the models, producing decreases in accuracy of up to 55 %. These results show the potential of binary classifier chains and open some questions about the use of domain-specific transfer learning.

本研究考察了深度卷积神经网络（cnn）的不同配置，以及使用特定领域迁移学习区分阿尔茨海默病和轻度认知障碍与正常对照的效果。用于训练我们的模型的数据由ADNI提供，总共包括1118个3D FDG-PET扫描。我们训练了一个二元分类器和一个多类分类器，以及二元分类器链，用于连续的多类分类。以不同顺序训练两条链：A链分类为认知正常（CN）与非CN，其次是阿尔茨海默病（AD）与轻度认知障碍（MCI）。分类器链B首先对AD和非AD进行分类，其次是MCI和CN。使用Med3D的权值，使用或不使用特定领域迁移学习对所有分类器进行训练。所有型号都达到了最先进的性能。分类器链A甚至以96%的准确率、95%的F1分数和99%的AUROC取得了优异的表现。使用特定领域的迁移学习导致大多数模型的性能更差，导致准确率下降高达55%。这些结果显示了二元分类器链的潜力，并对特定领域迁移学习的使用提出了一些问题。

{"title":"Domain specific transfer learning and classifier chains in Alzheimer's disease detection using 3D convolutional neural networks","authors":"Lisa-Marie Bente , Luca Himstedt , Tim Kacprowski , Alzheimer's Disease Neuroimaging Initiative","doi":"10.1016/j.imu.2025.101711","DOIUrl":"10.1016/j.imu.2025.101711","url":null,"abstract":"<div><div>This study examines different configurations of deep convolutional neural networks (CNNs) and the effect of using domain-specific transfer learning for distinguishing Alzheimer's Disease and Mild Cognitive Impairment from normal controls. The data used to train our models was provided by ADNI and included 1118 3D FDG-PET scans in total. We train a binary and a multiclass classifier, as well as chains of binary classifiers, for consecutive multiclass classification. Two chains were trained with different orders: chain A classified cognitively normal (CN) vs. non-CN, followed by Alzheimer's disease (AD) vs. mild cognitive impairment (MCI). Classifier chain B classified AD vs. non-AD first, followed by MCI vs. CN. All classifiers were trained with and without the use of domain-specific transfer learning, using weights from Med3D. All models achieve comparable performance to the state-of-the-art. Classifier chain A even achieved superior performance with an accuracy of 96 %, F1 score of 95 % and AUROC of 99 %. Using domain-specific transfer learning resulted in worse performance among the majority of the models, producing decreases in accuracy of up to 55 %. These results show the potential of binary classifier chains and open some questions about the use of domain-specific transfer learning.</div></div>","PeriodicalId":13953,"journal":{"name":"Informatics in Medicine Unlocked","volume":"59 ","pages":"Article 101711"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145526152","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}

引用次数: 0