Intelligence-based medicine最新文献

STViTDA-Net: An explainable transformer-based framework with STGAN-ViT-MAE and deformable attention for multi-class skin cancer classification STViTDA-Net：一个可解释的基于变压器的框架，具有stgan - vita - mae和可变形的关注，用于多类别皮肤癌分类

Intelligence-based medicine

Pub Date : 2026-01-22 DOI: 10.1016/j.ibmed.2026.100350

Ravula Jyothsna , K. Prasanna , U. Moulali , V. Surya Narayana Reddy , D. Ramya Krishna , T. Praveen Kumar

Skin cancer continues to pose a major global health challenge, and its early identification is essential for improving patient outcomes. Traditional diagnostic practices rely heavily on clinician expertise and manual interpretation of dermoscopic images, making the process subjective, inconsistent, and time-consuming. To address these limitations, this work introduces STViTDA-Net, an explainable transformer-based framework designed for fast, objective, and scalable multi-class skin cancer classification. The model integrates three key components: STGAN for class-balanced dermoscopic image augmentation, ViT-MAE for robust hierarchical feature learning through masked patch reconstruction, and a Deformable Attention Transformer Encoder that adaptively focuses on irregular lesion boundaries and subtle spatial variations. Preprocessing with Error Level Analysis (ELA) enhances fine-grained diagnostic cues, while Grad-CAM provides interpretable heatmaps that highlight the regions influencing the model's predictions. Unlike manual dermoscopic evaluation, STViTDA-Net performs end-to-end inference within milliseconds and delivers consistent, expert-independent predictions supported by visual explanations. When evaluated on the ISIC2019 dataset comprising nine lesion categories, the model achieves 99.35 % accuracy, 99.0 % precision, 99.5 % recall, 99.2 % F1-score, and 99.2 % AUC-ROC, surpassing existing CNN and transformer baselines. By unifying class-balanced augmentation, adaptive feature encoding, deformable attention, and explainable outputs, STViTDA-Net establishes a powerful and efficient solution for automated dermatological diagnosis.

皮肤癌继续构成一项重大的全球健康挑战，其早期识别对于改善患者预后至关重要。传统的诊断实践在很大程度上依赖于临床医生的专业知识和对皮肤镜图像的人工解读，这使得该过程主观、不一致且耗时。为了解决这些限制，这项工作引入了STViTDA-Net，这是一个可解释的基于变压器的框架，旨在快速，客观和可扩展的多类别皮肤癌分类。该模型集成了三个关键组件：用于类平衡皮肤镜图像增强的STGAN，通过掩膜斑块重建进行鲁棒分层特征学习的viti - mae，以及自适应关注不规则病变边界和微妙空间变化的可变形注意力转换器编码器。误差水平分析（ELA）的预处理增强了细粒度的诊断线索，而Grad-CAM提供了可解释的热图，突出了影响模型预测的区域。与手动皮肤镜评估不同，STViTDA-Net在几毫秒内执行端到端推理，并提供一致的、独立于专家的预测，并得到视觉解释的支持。当在包含9个病变类别的ISIC2019数据集上进行评估时，该模型达到99.35%的准确率、99.0%的精度、99.5%的召回率、99.2%的f1得分和99.2%的AUC-ROC，超过了现有的CNN和transformer基线。通过统一类平衡增强、自适应特征编码、可变形的注意力和可解释的输出，STViTDA-Net建立了一个强大而有效的自动皮肤病诊断解决方案。

{"title":"STViTDA-Net: An explainable transformer-based framework with STGAN-ViT-MAE and deformable attention for multi-class skin cancer classification","authors":"Ravula Jyothsna , K. Prasanna , U. Moulali , V. Surya Narayana Reddy , D. Ramya Krishna , T. Praveen Kumar","doi":"10.1016/j.ibmed.2026.100350","DOIUrl":"10.1016/j.ibmed.2026.100350","url":null,"abstract":"<div><div>Skin cancer continues to pose a major global health challenge, and its early identification is essential for improving patient outcomes. Traditional diagnostic practices rely heavily on clinician expertise and manual interpretation of dermoscopic images, making the process subjective, inconsistent, and time-consuming. To address these limitations, this work introduces STViTDA-Net, an explainable transformer-based framework designed for fast, objective, and scalable multi-class skin cancer classification. The model integrates three key components: STGAN for class-balanced dermoscopic image augmentation, ViT-MAE for robust hierarchical feature learning through masked patch reconstruction, and a Deformable Attention Transformer Encoder that adaptively focuses on irregular lesion boundaries and subtle spatial variations. Preprocessing with Error Level Analysis (ELA) enhances fine-grained diagnostic cues, while Grad-CAM provides interpretable heatmaps that highlight the regions influencing the model's predictions. Unlike manual dermoscopic evaluation, STViTDA-Net performs end-to-end inference within milliseconds and delivers consistent, expert-independent predictions supported by visual explanations. When evaluated on the ISIC2019 dataset comprising nine lesion categories, the model achieves 99.35 % accuracy, 99.0 % precision, 99.5 % recall, 99.2 % F1-score, and 99.2 % AUC-ROC, surpassing existing CNN and transformer baselines. By unifying class-balanced augmentation, adaptive feature encoding, deformable attention, and explainable outputs, STViTDA-Net establishes a powerful and efficient solution for automated dermatological diagnosis.</div></div>","PeriodicalId":73399,"journal":{"name":"Intelligence-based medicine","volume":"13 ","pages":"Article 100350"},"PeriodicalIF":0.0,"publicationDate":"2026-01-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146077476","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

Emerging anomaly detection techniques for electronic health records: A survey 新兴的电子健康记录异常检测技术：综述

Intelligence-based medicine

Pub Date : 2026-01-21 DOI: 10.1016/j.ibmed.2026.100349

Soumendra N. Bhanja , Haoran Niu , Yang Chen , Olufemi A. Omitaomu , Angela Laurio , Amber Trickey , Vijayalakshmi Sampath , Jonathan R. Nebeker

Background

Anomaly detection in electronic health records (EHRs) is a cornerstone of biomedical informatics, with direct implications for patient safety, clinical decision-making, and the prevention of healthcare fraud. Once guided primarily by simple rule-based methods, the field has advanced rapidly, driven by increased computing power, richer and more detailed health data, and the rise of machine learning and deep learning techniques. The objective of this paper is to provide a comprehensive overview of modern approaches to detecting anomalies in EHRs, outlining their strengths, limitations, and relevance to key healthcare challenges. We review traditional statistical methods alongside newer ML- and DL-based strategies and hybrid models, with particular attention to how these techniques support transparency and build clinical trust.

Methods

This paper presents a thorough and critical survey through systematic review (PRISMA-based) of the latest anomaly detection strategies in time-sequence data domains within electronic health record systems.

Results

We explore a broad spectrum of methodologies, including statistical models, supervised and unsupervised learning approaches, hybrid frameworks, and state-of-the-art ML-based techniques that collectively advance the precision and scalability of detecting anomalies in complex clinical datasets. In addition to mapping current capabilities, we address the enduring challenges that hinder widespread implementation and provide a forward-looking perspective on the future of anomaly detection in the data-rich landscape of modern healthcare.

Summary

The advancement in AI-based approaches is reported along with the basic principles of the individual approaches and their applicability. The increased availability of high-quality data, advancements in DL approaches, and enhanced computation power are leading to more frequent adaptation of DL-based approaches. Emerging DL-based approaches that have been adapted in other domains or recently applied in the EHR domain are also discussed in detail. Although DL-based approaches can improve model predictions by incorporating comorbidities, their application is limited in low-frequency data domains (e.g., when the total available data remains in the single digits). Therefore, the user must carefully consider the application based on data availability.

电子健康记录（EHRs）中的异常检测是生物医学信息学的基石，对患者安全、临床决策和预防医疗欺诈具有直接影响。曾经主要由简单的基于规则的方法指导，该领域在计算能力增强、更丰富和更详细的健康数据以及机器学习和深度学习技术的兴起的推动下迅速发展。本文的目的是全面概述检测电子病历异常的现代方法，概述其优势、局限性以及与关键医疗保健挑战的相关性。我们回顾了传统的统计方法以及新的基于ML和dl的策略和混合模型，特别关注这些技术如何支持透明度和建立临床信任。方法通过系统回顾（基于prisma），对电子健康记录系统中时间序列数据域的最新异常检测策略进行了全面而批判性的调查。我们探索了广泛的方法，包括统计模型、监督和无监督学习方法、混合框架和最先进的基于ml的技术，这些技术共同提高了在复杂临床数据集中检测异常的精度和可扩展性。除了映射当前功能之外，我们还解决了阻碍广泛实施的持久挑战，并提供了在数据丰富的现代医疗保健环境中异常检测的前瞻性视角。综述了基于人工智能方法的研究进展，介绍了各种方法的基本原理及其适用性。高质量数据可用性的增加、深度学习方法的进步以及计算能力的增强导致了基于深度学习的方法的更频繁的适应。还详细讨论了在其他领域中采用或最近在EHR领域中应用的新兴基于dl的方法。尽管基于dl的方法可以通过纳入合并症来改进模型预测，但它们的应用仅限于低频数据域（例如，当总可用数据保持在个位数时）。因此，用户必须根据数据可用性仔细考虑应用程序。

{"title":"Emerging anomaly detection techniques for electronic health records: A survey","authors":"Soumendra N. Bhanja , Haoran Niu , Yang Chen , Olufemi A. Omitaomu , Angela Laurio , Amber Trickey , Vijayalakshmi Sampath , Jonathan R. Nebeker","doi":"10.1016/j.ibmed.2026.100349","DOIUrl":"10.1016/j.ibmed.2026.100349","url":null,"abstract":"<div><h3>Background</h3><div>Anomaly detection in electronic health records (EHRs) is a cornerstone of biomedical informatics, with direct implications for patient safety, clinical decision-making, and the prevention of healthcare fraud. Once guided primarily by simple rule-based methods, the field has advanced rapidly, driven by increased computing power, richer and more detailed health data, and the rise of machine learning and deep learning techniques. The objective of this paper is to provide a comprehensive overview of modern approaches to detecting anomalies in EHRs, outlining their strengths, limitations, and relevance to key healthcare challenges. We review traditional statistical methods alongside newer ML- and DL-based strategies and hybrid models, with particular attention to how these techniques support transparency and build clinical trust.</div></div><div><h3>Methods</h3><div>This paper presents a thorough and critical survey through systematic review (PRISMA-based) of the latest anomaly detection strategies in time-sequence data domains within electronic health record systems.</div></div><div><h3>Results</h3><div>We explore a broad spectrum of methodologies, including statistical models, supervised and unsupervised learning approaches, hybrid frameworks, and state-of-the-art ML-based techniques that collectively advance the precision and scalability of detecting anomalies in complex clinical datasets. In addition to mapping current capabilities, we address the enduring challenges that hinder widespread implementation and provide a forward-looking perspective on the future of anomaly detection in the data-rich landscape of modern healthcare.</div></div><div><h3>Summary</h3><div>The advancement in AI-based approaches is reported along with the basic principles of the individual approaches and their applicability. The increased availability of high-quality data, advancements in DL approaches, and enhanced computation power are leading to more frequent adaptation of DL-based approaches. Emerging DL-based approaches that have been adapted in other domains or recently applied in the EHR domain are also discussed in detail. Although DL-based approaches can improve model predictions by incorporating comorbidities, their application is limited in low-frequency data domains (e.g., when the total available data remains in the single digits). Therefore, the user must carefully consider the application based on data availability.</div></div>","PeriodicalId":73399,"journal":{"name":"Intelligence-based medicine","volume":"13 ","pages":"Article 100349"},"PeriodicalIF":0.0,"publicationDate":"2026-01-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146077477","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

Improving glioma grade classification with a hybrid CNN-transformer model 用CNN-transformer混合模型改进胶质瘤分级

Intelligence-based medicine

Pub Date : 2026-01-19 DOI: 10.1016/j.ibmed.2025.100334

Sreedevi Gutta, Shyam Sundhar Yathirajam

Purpose

Accurate glioma grading is critical for treatment planning and prognosis. While convolutional neural networks (CNNs) capture fine-grained local features and transformers model long-range dependencies, each alone has limitations. This work investigates whether a hybrid CNN–Transformer architecture can improve classification performance on multi-sequence MRI.

Approach

We evaluated ten machine learning/deep learning models, including a radiomics approach, simple convolutional neural network (CNN), long-short term memory (LSTM), ensemble model, ResNet-based transfer learning, Vision Transformers (ViT, DeiT-base, DeiT-small, and DeiT-tiny), and a hybrid model that combines CNN features with DeiT Tiny. Performance was assessed using accuracy, precision, recall, and F1-score. Model interpretability was explored with Grad-CAM, attention maps, and t-SNE feature visualizations.

Results

Transformer-based models consistently outperformed CNNs and recurrent architectures, with DeiT Tiny achieving the best standalone performance (F1 = 0.90). The hybrid CNN+DeiT Tiny achieved the highest overall performance (F1 = 0.96, precision = 0.95, recall = 0.97), while maintaining practical efficiency (13 ms/slice inference, 89 MB model size). Interpretability analyses showed that the hybrid model effectively integrates local and global features, and t-SNE confirmed strong feature separability between glioma grades.

Conclusions

Combining CNNs and transformers yields superior accuracy, generalization, and interpretability for glioma grading, while remaining computationally feasible for real-time use. The hybrid model's accuracy and efficiency can help make glioma grading more reliable and useful in real clinical practice.

目的准确的胶质瘤分级对治疗计划和预后至关重要。虽然卷积神经网络（cnn）捕获细粒度的局部特征和变压器模型的远程依赖关系，但每一个单独都有局限性。这项工作研究了混合CNN-Transformer架构是否可以提高多序列MRI的分类性能。我们评估了十种机器学习/深度学习模型，包括放射组学方法、简单卷积神经网络（CNN）、长短期记忆（LSTM）、集成模型、基于resnet的迁移学习、视觉变形器（ViT、DeiT-base、DeiT-small和DeiT- Tiny），以及将CNN特征与DeiT- Tiny相结合的混合模型。使用准确性、精密度、召回率和f1评分来评估性能。通过Grad-CAM、注意图和t-SNE特征可视化来探索模型的可解释性。结果基于transformer的模型始终优于cnn和循环架构，其中DeiT Tiny获得了最佳的独立性能（F1 = 0.90）。混合CNN+DeiT Tiny实现了最高的整体性能（F1 = 0.96，精度= 0.95，召回率= 0.97），同时保持了实际效率（13 ms/slice inference， 89 MB模型大小）。可解释性分析表明，混合模型有效地整合了局部和全局特征，t-SNE证实了胶质瘤分级之间具有很强的特征可分离性。结论：结合cnn和transformer对胶质瘤分级具有更高的准确性、通用性和可解释性，同时在实时使用中仍然具有计算可行性。混合模型的准确性和效率有助于使胶质瘤分级更可靠，在实际临床实践中更有用。

{"title":"Improving glioma grade classification with a hybrid CNN-transformer model","authors":"Sreedevi Gutta, Shyam Sundhar Yathirajam","doi":"10.1016/j.ibmed.2025.100334","DOIUrl":"10.1016/j.ibmed.2025.100334","url":null,"abstract":"<div><h3>Purpose</h3><div>Accurate glioma grading is critical for treatment planning and prognosis. While convolutional neural networks (CNNs) capture fine-grained local features and transformers model long-range dependencies, each alone has limitations. This work investigates whether a hybrid CNN–Transformer architecture can improve classification performance on multi-sequence MRI.</div></div><div><h3>Approach</h3><div>We evaluated ten machine learning/deep learning models, including a radiomics approach, simple convolutional neural network (CNN), long-short term memory (LSTM), ensemble model, ResNet-based transfer learning, Vision Transformers (ViT, DeiT-base, DeiT-small, and DeiT-tiny), and a hybrid model that combines CNN features with DeiT Tiny. Performance was assessed using accuracy, precision, recall, and F1-score. Model interpretability was explored with Grad-CAM, attention maps, and t-SNE feature visualizations.</div></div><div><h3>Results</h3><div>Transformer-based models consistently outperformed CNNs and recurrent architectures, with DeiT Tiny achieving the best standalone performance (F1 = 0.90). The hybrid CNN+DeiT Tiny achieved the highest overall performance (F1 = 0.96, precision = 0.95, recall = 0.97), while maintaining practical efficiency (13 ms/slice inference, 89 MB model size). Interpretability analyses showed that the hybrid model effectively integrates local and global features, and t-SNE confirmed strong feature separability between glioma grades.</div></div><div><h3>Conclusions</h3><div>Combining CNNs and transformers yields superior accuracy, generalization, and interpretability for glioma grading, while remaining computationally feasible for real-time use. The hybrid model's accuracy and efficiency can help make glioma grading more reliable and useful in real clinical practice.</div></div>","PeriodicalId":73399,"journal":{"name":"Intelligence-based medicine","volume":"13 ","pages":"Article 100334"},"PeriodicalIF":0.0,"publicationDate":"2026-01-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146037646","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

Advanced AI framework for accurate detection and classification of brain tumours from MRI images 用于从MRI图像中准确检测和分类脑肿瘤的先进人工智能框架

Intelligence-based medicine

Pub Date : 2026-01-19 DOI: 10.1016/j.ibmed.2026.100348

M. Rajesh , K. Swaminathan , K. Vengatesan , Usha Moorthy , Sathishkumar Veerappampalayam Easwaramoorthy

Brain tumours adversely affect patient outcomes owing to their intricacy and the difficulties associated with diagnosis. The accuracy and timeliness of diagnosis are hindered by the subjectivity and unpredictability inherent in manual magnetic resonance imaging (MRI) interpretation. We present novel research on artificial intelligence systems capable of detecting, segmenting, and categorising brain cancers utilising MRI data, which may assist in addressing these issues. The system utilises advanced convolutional neural network (CNN) designs and unique explainability methods; it is designed for application in therapeutic and evidential contexts. This approach addresses deficiencies in cancer categorisation, differentiation, and AI interpretability, hence enhancing the accuracy and reliability of diagnosis. The method's efficacy and practical utility were evidenced through validation on extensive MRI datasets encompassing gliomas, meningiomas, pituitary tumours, and healthy controls. An AI-driven diagnostic tool can increase clinical decision-making, decrease diagnostic error rates, expedite therapy initiation, and improve patient outcomes.

脑肿瘤由于其复杂性和与诊断相关的困难而对患者的预后产生不利影响。诊断的准确性和及时性受到人工磁共振成像（MRI）解释固有的主观性和不可预测性的阻碍。我们提出了利用MRI数据检测、分割和分类脑癌的人工智能系统的新研究，这可能有助于解决这些问题。该系统采用先进的卷积神经网络（CNN）设计和独特的可解释性方法；它被设计用于治疗和证据背景下的应用。这种方法解决了癌症分类、鉴别和人工智能可解释性方面的缺陷，从而提高了诊断的准确性和可靠性。通过对包括胶质瘤、脑膜瘤、垂体瘤和健康对照在内的广泛MRI数据集的验证，证明了该方法的有效性和实用性。人工智能驱动的诊断工具可以增加临床决策，降低诊断错误率，加快治疗启动，改善患者预后。

{"title":"Advanced AI framework for accurate detection and classification of brain tumours from MRI images","authors":"M. Rajesh , K. Swaminathan , K. Vengatesan , Usha Moorthy , Sathishkumar Veerappampalayam Easwaramoorthy","doi":"10.1016/j.ibmed.2026.100348","DOIUrl":"10.1016/j.ibmed.2026.100348","url":null,"abstract":"<div><div>Brain tumours adversely affect patient outcomes owing to their intricacy and the difficulties associated with diagnosis. The accuracy and timeliness of diagnosis are hindered by the subjectivity and unpredictability inherent in manual magnetic resonance imaging (MRI) interpretation. We present novel research on artificial intelligence systems capable of detecting, segmenting, and categorising brain cancers utilising MRI data, which may assist in addressing these issues. The system utilises advanced convolutional neural network (CNN) designs and unique explainability methods; it is designed for application in therapeutic and evidential contexts. This approach addresses deficiencies in cancer categorisation, differentiation, and AI interpretability, hence enhancing the accuracy and reliability of diagnosis. The method's efficacy and practical utility were evidenced through validation on extensive MRI datasets encompassing gliomas, meningiomas, pituitary tumours, and healthy controls. An AI-driven diagnostic tool can increase clinical decision-making, decrease diagnostic error rates, expedite therapy initiation, and improve patient outcomes.</div></div>","PeriodicalId":73399,"journal":{"name":"Intelligence-based medicine","volume":"13 ","pages":"Article 100348"},"PeriodicalIF":0.0,"publicationDate":"2026-01-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146037647","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

TransFusion-BCNet: A transformer-driven multi-modal fusion and explainable deep learning framework for breast cancer diagnosis 输血- bcnet：用于乳腺癌诊断的转换器驱动的多模式融合和可解释的深度学习框架

Intelligence-based medicine

Pub Date : 2026-01-12 DOI: 10.1016/j.ibmed.2026.100346

Ahmed Kateb Jumaah Al-Nussairi , Saleem Malik , Yasser Taha Alzubaidi , S Gopal Krishna Patro , Kasim Sakran Abass , Iman Basheti , Mohammad Khishe

Most AI breast cancer detection systems use single-modality imaging algorithms, limiting clinical reliability. Early and accurate detection improves therapy and mortality. These challenges are addressed by Transformer-driven multi-modal fusion and explainable deep learning system, TransFusion-BCNet for breast cancer diagnosis. The framework consists of three parts. The TriFusion-Transformer (TriFT) performs three-tier fusion: intra-modality fusion across multiple mammogram views and imaging sources, inter-modality fusion combining mammogram, ultrasound, MRI, and clinical features, and decision-level fusion for robust outcome prediction. TriFT detects complicated connections across heterogeneous modalities, unlike classical fusion. Second, we present the FusionAttribution Map (FAMap), a dual-level interpretability mechanism that generates imaging data region-level saliency maps and modality-level contribution scores to evaluate input source influence. This openness helps clinicians understand where and which modality drives predictions. Third, the MetaFusion Optimizer (MFO) adjusts fusion weights, network depth, and learning parameters via evolutionary search and gradient-based fine-tuning. Traditional optimizers lack model generalization and training stability. This staged technique improved both. TransFusion-BCNet outperforms CNN–Transformer hybrids with 99.4 % accuracy, 99.0 % precision, 99.2 % recall, and 99.1 % F1-score in extensive CBIS-DDSM,BUSI, TCGA-BRCA and RIDER Breast MRI datasets. With TriFT, FAMap, and MFO, TransFusion-BCNet provides a robust, transparent, and clinically interpretable diagnostic framework, improving AI in breast cancer screening and decision assistance.

大多数人工智能乳腺癌检测系统使用单模态成像算法，限制了临床可靠性。早期和准确的检测可以改善治疗和死亡率。这些挑战由transformer驱动的多模态融合和可解释的深度学习系统——用于乳腺癌诊断的输血- bcnet来解决。该框架由三部分组成。TriFT进行三层融合：跨多个乳房x光片视图和成像源的模态内融合，结合乳房x光片、超声、MRI和临床特征的模态间融合，以及用于稳健结果预测的决策级融合。与传统的融合不同，TriFT可以检测跨异构模式的复杂连接。其次，我们提出了FusionAttribution Map (FAMap)，这是一种双层可解释性机制，可生成成像数据区域级显著性图和模式级贡献分数，以评估输入源的影响。这种开放性有助于临床医生了解在哪里以及哪种模式驱动预测。第三，MetaFusion Optimizer （MFO）通过进化搜索和基于梯度的微调来调整融合权重、网络深度和学习参数。传统的优化算法缺乏模型泛化和训练稳定性。这种分阶段的技术改善了这两方面。在广泛的CBIS-DDSM、BUSI、TCGA-BRCA和RIDER乳腺MRI数据集中，输血- bcnet的准确率为99.4%，精密度为99.0%，召回率为99.2%，f1评分为99.1%，优于CNN-Transformer混合方案。通过TriFT、FAMap和MFO，输血- bcnet提供了一个强大、透明和临床可解释的诊断框架，改善了人工智能在乳腺癌筛查和决策辅助方面的应用。

{"title":"TransFusion-BCNet: A transformer-driven multi-modal fusion and explainable deep learning framework for breast cancer diagnosis","authors":"Ahmed Kateb Jumaah Al-Nussairi , Saleem Malik , Yasser Taha Alzubaidi , S Gopal Krishna Patro , Kasim Sakran Abass , Iman Basheti , Mohammad Khishe","doi":"10.1016/j.ibmed.2026.100346","DOIUrl":"10.1016/j.ibmed.2026.100346","url":null,"abstract":"<div><div>Most AI breast cancer detection systems use single-modality imaging algorithms, limiting clinical reliability. Early and accurate detection improves therapy and mortality. These challenges are addressed by Transformer-driven multi-modal fusion and explainable deep learning system, TransFusion-BCNet for breast cancer diagnosis. The framework consists of three parts. The TriFusion-Transformer (TriFT) performs three-tier fusion: intra-modality fusion across multiple mammogram views and imaging sources, inter-modality fusion combining mammogram, ultrasound, MRI, and clinical features, and decision-level fusion for robust outcome prediction. TriFT detects complicated connections across heterogeneous modalities, unlike classical fusion. Second, we present the FusionAttribution Map (FAMap), a dual-level interpretability mechanism that generates imaging data region-level saliency maps and modality-level contribution scores to evaluate input source influence. This openness helps clinicians understand where and which modality drives predictions. Third, the MetaFusion Optimizer (MFO) adjusts fusion weights, network depth, and learning parameters via evolutionary search and gradient-based fine-tuning. Traditional optimizers lack model generalization and training stability. This staged technique improved both. TransFusion-BCNet outperforms CNN–Transformer hybrids with 99.4 % accuracy, 99.0 % precision, 99.2 % recall, and 99.1 % F1-score in extensive CBIS-DDSM,BUSI, TCGA-BRCA and RIDER Breast MRI datasets. With TriFT, FAMap, and MFO, TransFusion-BCNet provides a robust, transparent, and clinically interpretable diagnostic framework, improving AI in breast cancer screening and decision assistance.</div></div>","PeriodicalId":73399,"journal":{"name":"Intelligence-based medicine","volume":"13 ","pages":"Article 100346"},"PeriodicalIF":0.0,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146077478","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 risk prediction for diabetes, hypertension, and heart disease using SMOTE-ENN balancing with PCA and gradient boosting in healthcare AI 在医疗保健AI中使用SMOTE-ENN平衡与PCA和梯度增强来增强糖尿病、高血压和心脏病的风险预测

Intelligence-based medicine

Pub Date : 2026-01-08 DOI: 10.1016/j.ibmed.2025.100339

Tapon Paul , Md Assaduzzaman , Nafiz Fahad , Md Jakir Hossen

Predicting chronic disease remains a crucial problem, particularly in low-resource environments where accurate and timely predictions are of utmost importance. Machine learning methods do not adequately generalize, encounter data imbalance and computation issues when used. In this study, an improved model that makes predictions using Synthetic minority over-sampling technique and edited nearest neighbor balancing techniques and Principal Component Analysis and Gradient Boosting to predict chronic diseases based on readily available clinical profiles is proposed. The dataset used in this study was retrieved from provide precise source: Kaggle dataset. It contains over 1500 anonymized patient records and includes 15 features such as demographic, lifestyle, and clinical measures. Standardized encoding labels, and adjusting classes using SMOTE-ENN have been completed before the PCA was conducted to improve computation speed and reduce overfitting. Decision Tree, Random Forest, LightGBM, XGBoost have been used for comparison to suggest the best performing model seen to be Gradient Boosting. PCA is performed, and the Gradient Boosting approach produces better results. Precision measures how often the classification system is correct when making a positive test result, while recall is determined using a contingency table, and the F1 score, the possibility of modeling outcomes out of 100 trials. The model proposed in the experiment provides the following outputs: Accuracy (CV: 99.33 %, CI: 98.90 %–99.50 %), Precision (CV: 99 %, CI: 98 %–99.5 %), Recall (CV: 99 %, CI: 98 %–99.5 %), F1-Score (CV: 99 %, CI: 98 %–99.5 %). The model's performance was evaluated using cross-validation, yielding an accuracy of 98.90 %. The classifying system performance is specified by the ROC-AUC ranking. It outperforms the model making indefinite projections; its ROC-AUC value is greater than 0.99. The suggested model is a robust, interpretable, and high-precision approach for the early detection of chronic conditions. Therefore, the suggested machine learning system can deliver a considerable promise with respect to creating patient-oriented outcomes.

预测慢性病仍然是一个关键问题，特别是在资源匮乏的环境中，准确和及时的预测至关重要。机器学习方法泛化不充分，使用时会遇到数据不平衡和计算问题。在本研究中，提出了一种改进的模型，该模型使用合成少数过采样技术和编辑最近邻平衡技术以及主成分分析和梯度增强来预测基于现有临床资料的慢性疾病。本研究使用的数据集检索自提供精确来源：Kaggle数据集。它包含1500多名匿名患者记录，包括15个特征，如人口统计、生活方式和临床措施。为了提高计算速度和减少过拟合，在PCA之前已经完成了标准化的编码标签和使用SMOTE-ENN进行类调整。决策树，随机森林，LightGBM， XGBoost已经被用于比较，以表明表现最好的模型是梯度增强。进行了主成分分析，结果表明梯度增强方法效果较好。精确度衡量的是分类系统在产生阳性测试结果时的正确频率，而召回率是使用列联表和F1分数来确定的，F1分数是100次试验中建模结果的可能性。实验中提出的模型提供了以下输出：准确率（CV: 99.33%, CI: 98.90% - 99.50%），精度（CV: 99%, CI: 98% - 99.5%），召回率（CV: 99%, CI: 98% - 99.5%）， F1-Score （CV: 99%, CI: 98% - 99.5%）。使用交叉验证对模型的性能进行了评估，准确度为98.90%。分类系统的性能由ROC-AUC排序指定。它优于模型进行不确定的预测；ROC-AUC值大于0.99。所建议的模型是一种稳健的、可解释的、高精度的慢性疾病早期检测方法。因此，建议的机器学习系统在创造以患者为导向的结果方面可以提供相当大的承诺。

{"title":"Enhancing risk prediction for diabetes, hypertension, and heart disease using SMOTE-ENN balancing with PCA and gradient boosting in healthcare AI","authors":"Tapon Paul , Md Assaduzzaman , Nafiz Fahad , Md Jakir Hossen","doi":"10.1016/j.ibmed.2025.100339","DOIUrl":"10.1016/j.ibmed.2025.100339","url":null,"abstract":"<div><div>Predicting chronic disease remains a crucial problem, particularly in low-resource environments where accurate and timely predictions are of utmost importance. Machine learning methods do not adequately generalize, encounter data imbalance and computation issues when used. In this study, an improved model that makes predictions using Synthetic minority over-sampling technique and edited nearest neighbor balancing techniques and Principal Component Analysis and Gradient Boosting to predict chronic diseases based on readily available clinical profiles is proposed. The dataset used in this study was retrieved from provide precise source: Kaggle dataset. It contains over 1500 anonymized patient records and includes 15 features such as demographic, lifestyle, and clinical measures. Standardized encoding labels, and adjusting classes using SMOTE-ENN have been completed before the PCA was conducted to improve computation speed and reduce overfitting. Decision Tree, Random Forest, LightGBM, XGBoost have been used for comparison to suggest the best performing model seen to be Gradient Boosting. PCA is performed, and the Gradient Boosting approach produces better results. Precision measures how often the classification system is correct when making a positive test result, while recall is determined using a contingency table, and the F1 score, the possibility of modeling outcomes out of 100 trials. The model proposed in the experiment provides the following outputs: Accuracy (CV: 99.33 %, CI: 98.90 %–99.50 %), Precision (CV: 99 %, CI: 98 %–99.5 %), Recall (CV: 99 %, CI: 98 %–99.5 %), F1-Score (CV: 99 %, CI: 98 %–99.5 %). The model's performance was evaluated using cross-validation, yielding an accuracy of 98.90 %. The classifying system performance is specified by the ROC-AUC ranking. It outperforms the model making indefinite projections; its ROC-AUC value is greater than 0.99. The suggested model is a robust, interpretable, and high-precision approach for the early detection of chronic conditions. Therefore, the suggested machine learning system can deliver a considerable promise with respect to creating patient-oriented outcomes.</div></div>","PeriodicalId":73399,"journal":{"name":"Intelligence-based medicine","volume":"13 ","pages":"Article 100339"},"PeriodicalIF":0.0,"publicationDate":"2026-01-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146037648","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 AI-based oral decision support systems: Hybrid image processing for detecting impacted maxillary canines in orthopantomograms 增强基于人工智能的口腔决策支持系统：混合图像处理在矫形断层摄影中检测上颌埋伏齿

Intelligence-based medicine

Pub Date : 2026-01-07 DOI: 10.1016/j.ibmed.2026.100345

Roopitha C.H , Veena Mayya , V. Sivakumar , Vathsala Patil , Divya Pai , Pavithra Varchas

Timely identification of impacted canines is crucial for preventing complications such as root resorption, misalignment, and damage to neighboring teeth. This research evaluates image processing techniques for AI-based oral decision support systems using orthopantomogram (OPG) images to determine effective preprocessing methods to improve deep learning (DL) model performance. The approach involved preprocessing OPG images followed by object detection using YOLOv5. Preprocessing techniques included median filtering, Gaussian blur, CLAHE, image sharpening, histogram stretching, and CLAHE combined with image sharpening. Performance was evaluated using precision, recall, F1 score, mAP, and IoU metrics. The CLAHE-sharpening combination achieved superior performance with precision of 0.934, recall of 0.932, F1 score of 0.931, mAP of 0.948, and IoU of 0.741, significantly outperforming unprocessed images. Grad-CAM visualizations confirmed that preprocessing enabled the model to identify relevant regions effectively. This study emphasizes the importance of preprocessing methods in improving the diagnostic accuracy in dental radiography for improved treatment planning.

及时识别患牙对于预防并发症如牙根吸收、牙齿错位和对邻近牙齿的损害至关重要。本研究评估了基于人工智能的口头决策支持系统的图像处理技术，使用正体层图（OPG）图像来确定有效的预处理方法，以提高深度学习（DL）模型的性能。该方法包括预处理OPG图像，然后使用YOLOv5进行目标检测。预处理技术包括中值滤波、高斯模糊、CLAHE、图像锐化、直方图拉伸、CLAHE与图像锐化相结合。使用精度、召回率、F1分数、mAP和IoU指标评估性能。clahe -锐化组合的精度为0.934，召回率为0.932，F1评分为0.931，mAP为0.948，IoU为0.741，明显优于未经处理的图像。Grad-CAM可视化证实，预处理使模型能够有效地识别相关区域。本研究强调了预处理方法在提高牙科x线摄影诊断准确性方面的重要性，以改善治疗计划。

{"title":"Enhancing AI-based oral decision support systems: Hybrid image processing for detecting impacted maxillary canines in orthopantomograms","authors":"Roopitha C.H , Veena Mayya , V. Sivakumar , Vathsala Patil , Divya Pai , Pavithra Varchas","doi":"10.1016/j.ibmed.2026.100345","DOIUrl":"10.1016/j.ibmed.2026.100345","url":null,"abstract":"<div><div>Timely identification of impacted canines is crucial for preventing complications such as root resorption, misalignment, and damage to neighboring teeth. This research evaluates image processing techniques for AI-based oral decision support systems using orthopantomogram (OPG) images to determine effective preprocessing methods to improve deep learning (DL) model performance. The approach involved preprocessing OPG images followed by object detection using YOLOv5. Preprocessing techniques included median filtering, Gaussian blur, CLAHE, image sharpening, histogram stretching, and CLAHE combined with image sharpening. Performance was evaluated using precision, recall, F1 score, mAP, and IoU metrics. The CLAHE-sharpening combination achieved superior performance with precision of 0.934, recall of 0.932, F1 score of 0.931, mAP of 0.948, and IoU of 0.741, significantly outperforming unprocessed images. Grad-CAM visualizations confirmed that preprocessing enabled the model to identify relevant regions effectively. This study emphasizes the importance of preprocessing methods in improving the diagnostic accuracy in dental radiography for improved treatment planning.</div></div>","PeriodicalId":73399,"journal":{"name":"Intelligence-based medicine","volume":"13 ","pages":"Article 100345"},"PeriodicalIF":0.0,"publicationDate":"2026-01-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145926930","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

The second-generation 3D-Printed localization grid for MRI-guided interventional procedures 用于mri引导介入手术的第二代3d打印定位网格

Intelligence-based medicine

Pub Date : 2026-01-06 DOI: 10.1016/j.ibmed.2026.100343

Queenie T.K. Shea , Wing Ki Wong , Louis Lee

Patient-specific three-dimensional (3D)-printed magnetic resonance imaging localization grids (MR-Grids) have demonstrated feasibility as an assistive tool for image-guided interventional procedures. However, a limitation of conventional designs arises when the target lesion lies in an imaging plane where the discrete grid markers are not visible. To address this challenge, we proposed a patient-specific MR-Grid incorporating a contrast-filled flexible tubing system arranged in a criss-cross pattern, ensuring visibility across all imaging slices.

The proposed MR-Grid comprises two primary components: (1) a 3D-printed patient-specific scaffold designed to conform to individual anatomical contours, and (2) a contrast-filled flexible tubing system inserted into the grooves of the scaffold.

The MR-Grid was tested in an interventional procedure using a biopsy phantom containing MRI-visible lesions to validate its utility. The grid facilitated precise needle insertion by identifying the optimal entry point under MR guidance, demonstrating its potential to improve accuracy and efficiency in image-guided interventions.

患者特异性三维（3D）打印的磁共振成像定位网格（MR-Grids）已被证明可作为图像引导介入手术的辅助工具。然而，当目标病变位于不可见的离散网格标记的成像平面时，传统设计的局限性就出现了。为了应对这一挑战，我们提出了一种针对患者的MR-Grid，其中包括以纵横交错模式排列的造影剂填充柔性管系统，确保所有成像切片的可见性。提出的MR-Grid包括两个主要组成部分：(1)3d打印的患者专用支架，设计符合个人解剖轮廓，以及(2)插入支架凹槽的对比填充柔性管道系统。MR-Grid在介入手术中进行了测试，使用包含mri可见病变的活检假体来验证其实用性。网格通过识别MR引导下的最佳切入点，促进了针的精确插入，展示了其在图像引导干预中提高准确性和效率的潜力。

{"title":"The second-generation 3D-Printed localization grid for MRI-guided interventional procedures","authors":"Queenie T.K. Shea , Wing Ki Wong , Louis Lee","doi":"10.1016/j.ibmed.2026.100343","DOIUrl":"10.1016/j.ibmed.2026.100343","url":null,"abstract":"<div><div>Patient-specific three-dimensional (3D)-printed magnetic resonance imaging localization grids (MR-Grids) have demonstrated feasibility as an assistive tool for image-guided interventional procedures. However, a limitation of conventional designs arises when the target lesion lies in an imaging plane where the discrete grid markers are not visible. To address this challenge, we proposed a patient-specific MR-Grid incorporating a contrast-filled flexible tubing system arranged in a criss-cross pattern, ensuring visibility across all imaging slices.</div><div>The proposed MR-Grid comprises two primary components: (1) a 3D-printed patient-specific scaffold designed to conform to individual anatomical contours, and (2) a contrast-filled flexible tubing system inserted into the grooves of the scaffold.</div><div>The MR-Grid was tested in an interventional procedure using a biopsy phantom containing MRI-visible lesions to validate its utility. The grid facilitated precise needle insertion by identifying the optimal entry point under MR guidance, demonstrating its potential to improve accuracy and efficiency in image-guided interventions.</div></div>","PeriodicalId":73399,"journal":{"name":"Intelligence-based medicine","volume":"13 ","pages":"Article 100343"},"PeriodicalIF":0.0,"publicationDate":"2026-01-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145926931","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

Automated right ventricular assessment in pediatric echocardiography via deep learning improves measurement reliability and reduces variability 通过深度学习的儿童超声心动图自动右心室评估提高了测量可靠性并减少了变异性

Intelligence-based medicine

Pub Date : 2026-01-06 DOI: 10.1016/j.ibmed.2026.100344

Ping He , Faith Zhu , Mariella Vargas-Gutierrez , Rakhika Kumar , Wei Hui , Yalin Lin , Mark K. Friedberg , Luc Mertens , Lauren Erdman

Background

Right ventricular (RV) function is important for pediatric cardiac evaluation but accurate and reproducible quantification of RV function is challenging. This study aimed to develop a deep learning (DL) model for RV functional assessment from echocardiography (ECHO) which out-performs current, manual methods.

Methods

We trained multiple DL segmentation models, using a dataset of 664 pediatric ECHOs, and proceeded with the best performing model for evaluation. DL model performance was assessed using the dice similarity coefficient (DSC) for segmentation, mean absolute error (MAE) for RVFAC. Blinded expert evaluation was conducted between ground truth and model generated segmentation outputs. A detailed analysis of inter-observer variability identified the main sources of RVFAC variability among four experts and the DL model, as well as opportunities for the model to improve RV assessment in practice.

Findings

The FCBFormer architecture yielded the best segmentation quality with DSC of 0.926 and MAE of 5.913 % for RVFAC prediction. Blinded expert review revealed that model generated segmentation was favored over human in 57.3 % of evaluated cases. All sources of variation were overcome by the RVFAC model: RV contour delineation, RV cardiac cycle selection, and RV end-diastolic/end-systolic frame identification.

Interpretation

This study demonstrates the feasibility of DL-based automated RV functional assessment for pediatric patients, offering a promising approach for more consistent and systematic longitudinal tracking of RV function than manual ECHO assessment.

背景右心室（RV）功能对儿童心脏评估很重要，但准确和可重复的量化右心室功能具有挑战性。本研究旨在开发一种深度学习（DL）模型，用于超声心动图（ECHO）的RV功能评估，该模型优于当前的手动方法。方法采用664例儿童超声数据集，对多个深度学习分割模型进行训练，选取表现最好的模型进行评价。使用骰子相似系数（DSC）进行分割，使用RVFAC的平均绝对误差（MAE）评估DL模型的性能。在地面真实值和模型生成的分割输出之间进行盲法专家评价。通过对观察者间变异性的详细分析，确定了四名专家和DL模型之间RVFAC变异性的主要来源，以及该模型在实践中改进RV评估的机会。结果FCBFormer结构对RVFAC预测的分割质量最好，DSC为0.926，MAE为5.913%。盲法专家评审显示，在57.3%的评估病例中，模型生成的分割优于人类。RVFAC模型克服了所有变异的来源：右心室轮廓描绘、右心室心动周期选择和右心室舒张末期/收缩末期框架识别。本研究证明了基于dl的儿童右心室功能自动评估的可行性，提供了一种比手动ECHO评估更一致和系统的右心室功能纵向跟踪的有希望的方法。

{"title":"Automated right ventricular assessment in pediatric echocardiography via deep learning improves measurement reliability and reduces variability","authors":"Ping He , Faith Zhu , Mariella Vargas-Gutierrez , Rakhika Kumar , Wei Hui , Yalin Lin , Mark K. Friedberg , Luc Mertens , Lauren Erdman","doi":"10.1016/j.ibmed.2026.100344","DOIUrl":"10.1016/j.ibmed.2026.100344","url":null,"abstract":"<div><h3>Background</h3><div>Right ventricular (RV) function is important for pediatric cardiac evaluation but accurate and reproducible quantification of RV function is challenging. This study aimed to develop a deep learning (DL) model for RV functional assessment from echocardiography (ECHO) which out-performs current, manual methods.</div></div><div><h3>Methods</h3><div>We trained multiple DL segmentation models, using a dataset of 664 pediatric ECHOs, and proceeded with the best performing model for evaluation. DL model performance was assessed using the dice similarity coefficient (DSC) for segmentation, mean absolute error (MAE) for RVFAC. Blinded expert evaluation was conducted between ground truth and model generated segmentation outputs. A detailed analysis of inter-observer variability identified the main sources of RVFAC variability among four experts and the DL model, as well as opportunities for the model to improve RV assessment in practice.</div></div><div><h3>Findings</h3><div>The FCBFormer architecture yielded the best segmentation quality with DSC of 0.926 and MAE of 5.913 % for RVFAC prediction. Blinded expert review revealed that model generated segmentation was favored over human in 57.3 % of evaluated cases. All sources of variation were overcome by the RVFAC model: RV contour delineation, RV cardiac cycle selection, and RV end-diastolic/end-systolic frame identification.</div></div><div><h3>Interpretation</h3><div>This study demonstrates the feasibility of DL-based automated RV functional assessment for pediatric patients, offering a promising approach for more consistent and systematic longitudinal tracking of RV function than manual ECHO assessment.</div></div>","PeriodicalId":73399,"journal":{"name":"Intelligence-based medicine","volume":"13 ","pages":"Article 100344"},"PeriodicalIF":0.0,"publicationDate":"2026-01-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145978173","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 AI for breast cancer detection: Biglycan biomarker classification with transfer learning 用于乳腺癌检测的可解释人工智能：Biglycan生物标志物分类与迁移学习

Intelligence-based medicine

Pub Date : 2026-01-01 DOI: 10.1016/j.ibmed.2025.100340

Md. Mominul Islam , Naime Akter , Md. Assaduzzaman , Md. Monir Hossain Shimul , Rahmatul Kabir Rasel Sarker

Background

Breast cancer is a leading global cause of cancer-related mortality, where early diagnosis is essential for improved survival outcomes. Although deep learning has shown strong performance in pathology image classification, many models remain difficult to interpret, which limits clinical trust and practical adoption. This study aimed to develop an explainable deep learning framework for classifying breast tissue based on Biglycan (BGN) biomarker expression.

Methods

Immunohistochemical photomicrographs from the publicly available Biglycan Breast Cancer Dataset were processed and classified into cancerous and healthy categories. Three transfer learning models, EfficientNet-B0, DenseNet-161, and ResNet-50, were fine-tuned using ImageNet pre-trained weights under a unified training setting with the Adam optimizer (learning rate = 0.001, batch size = 32, epochs = 50). Model performance was evaluated using accuracy, precision, recall, and F1-score. To enhance interpretability, Gradient-weighted Class Activation Mapping (Grad-CAM) was applied to visualize tissue regions that contributed most to the model's predictions.

Results

EfficientNet-B0 achieved the best overall performance with 99 % accuracy, precision of 0.97, recall of 1.00 for the cancerous class, and an F1-score of 0.99. Grad-CAM heatmaps indicated that the model focused on diagnostically relevant tissue regions associated with strong BGN-related staining patterns. The best-performing model was integrated into a lightweight web-based application to enable real-time image upload and prediction.

Conclusion

This study presents an explainable transfer learning approach for BGN-driven breast tissue classification with strong performance on the Biglycan dataset. The integration of Grad-CAM provides region-level visual explanations that improve transparency, while the web deployment demonstrates a practical pathway for accessible decision support in digital pathology.

乳腺癌是全球癌症相关死亡的主要原因，早期诊断对于改善生存结果至关重要。尽管深度学习在病理图像分类方面表现出色，但许多模型仍然难以解释，这限制了临床信任和实际应用。本研究旨在开发一个可解释的深度学习框架，用于基于Biglycan （BGN）生物标志物表达对乳腺组织进行分类。方法对公开的Biglycan乳腺癌数据集的免疫组化显微照片进行处理，并将其分为癌变和健康两类。使用ImageNet预训练权值，在Adam优化器的统一训练设置下（学习率= 0.001，批大小= 32,epoch = 50），对effentnet - b0、DenseNet-161和ResNet-50三个迁移学习模型进行了微调。使用准确性、精密度、召回率和f1评分来评估模型的性能。为了提高可解释性，应用梯度加权类激活映射（Grad-CAM）来可视化对模型预测贡献最大的组织区域。结果fficientnet - b0的准确率为99%，精密度为0.97，查全率为1.00，f1评分为0.99。Grad-CAM热图显示，该模型专注于与bgn相关的强染色模式相关的诊断相关组织区域。性能最好的模型被集成到一个轻量级的基于web的应用程序中，以实现实时图像上传和预测。本研究提出了一种可解释的迁移学习方法，用于bgnn驱动的乳腺组织分类，在Biglycan数据集上具有较强的性能。Grad-CAM的集成提供了区域级的可视化解释，提高了透明度，而网络部署展示了数字病理学中可访问的决策支持的实用途径。

{"title":"Explainable AI for breast cancer detection: Biglycan biomarker classification with transfer learning","authors":"Md. Mominul Islam , Naime Akter , Md. Assaduzzaman , Md. Monir Hossain Shimul , Rahmatul Kabir Rasel Sarker","doi":"10.1016/j.ibmed.2025.100340","DOIUrl":"10.1016/j.ibmed.2025.100340","url":null,"abstract":"<div><h3>Background</h3><div>Breast cancer is a leading global cause of cancer-related mortality, where early diagnosis is essential for improved survival outcomes. Although deep learning has shown strong performance in pathology image classification, many models remain difficult to interpret, which limits clinical trust and practical adoption. This study aimed to develop an explainable deep learning framework for classifying breast tissue based on Biglycan (BGN) biomarker expression.</div></div><div><h3>Methods</h3><div>Immunohistochemical photomicrographs from the publicly available Biglycan Breast Cancer Dataset were processed and classified into cancerous and healthy categories. Three transfer learning models, EfficientNet-B0, DenseNet-161, and ResNet-50, were fine-tuned using ImageNet pre-trained weights under a unified training setting with the Adam optimizer (learning rate = 0.001, batch size = 32, epochs = 50). Model performance was evaluated using accuracy, precision, recall, and F1-score. To enhance interpretability, Gradient-weighted Class Activation Mapping (Grad-CAM) was applied to visualize tissue regions that contributed most to the model's predictions.</div></div><div><h3>Results</h3><div>EfficientNet-B0 achieved the best overall performance with 99 % accuracy, precision of 0.97, recall of 1.00 for the cancerous class, and an F1-score of 0.99. Grad-CAM heatmaps indicated that the model focused on diagnostically relevant tissue regions associated with strong BGN-related staining patterns. The best-performing model was integrated into a lightweight web-based application to enable real-time image upload and prediction.</div></div><div><h3>Conclusion</h3><div>This study presents an explainable transfer learning approach for BGN-driven breast tissue classification with strong performance on the Biglycan dataset. The integration of Grad-CAM provides region-level visual explanations that improve transparency, while the web deployment demonstrates a practical pathway for accessible decision support in digital pathology.</div></div>","PeriodicalId":73399,"journal":{"name":"Intelligence-based medicine","volume":"13 ","pages":"Article 100340"},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145926928","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