Intelligence-based medicine最新文献_第4页

Explainable machine learning for early heart disease risk prediction: Insights from a clinical dataset in Bangladesh 可解释的机器学习用于早期心脏病风险预测：来自孟加拉国临床数据集的见解

Intelligence-based medicine

Pub Date : 2026-03-01 Epub Date: 2026-02-16 DOI: 10.1016/j.ibmed.2026.100352

Arpita Chakraborty , Utpol Kanti Das , Sadia Sazzad , Panna Das , Md. Mehedi Hassan Khan

Background:

Cardiovascular diseases remain one of the leading causes of mortality worldwide, particularly in low- and middle-income countries. Early and accurate prediction of heart disease is essential for timely intervention and improved patient outcomes. Machine learning techniques offer promising solutions; however, challenges such as class imbalance, lack of interpretability, and limited real-world validation persist.

Methods:

In this study, a machine learning-based heart disease prediction framework was developed using a real-world clinical dataset comprising 5000 patient records collected from healthcare facilities in Bangladesh. Data preprocessing included cleaning, feature encoding, train–test splitting, and class imbalance handling using the Synthetic Minority Oversampling Technique (SMOTE). Multiple machine learning models — Logistic Regression, Decision Tree, Support Vector Machine, and Random Forest — were evaluated using 10-fold stratified cross-validation. Model performance was assessed using accuracy, precision, recall, and F1-score. SHAP (SHapley Additive exPlanations) was employed to enhance model interpretability. The best-performing model was deployed as a web-based decision support system.

Results:

Among the evaluated models, the Random Forest classifier achieved the best performance, with an accuracy of 98%, recall of 96%, and F1-score of 96%. Ablation studies demonstrated the effectiveness of SMOTE, feature integration, and ensemble modeling. SHAP analysis identified clinically relevant features contributing to heart disease prediction, enhancing transparency and trust in model decisions.

Conclusions:

The proposed framework provides an accurate, interpretable, and practical solution for heart disease prediction using real-world clinical data. The integration of explainable machine learning and web-based deployment highlights its potential for clinical decision support. Future work will focus on multi-center prospective validation and adaptive model updating to further improve generalizability and real-world applicability.

背景：心血管疾病仍然是世界范围内死亡的主要原因之一，特别是在低收入和中等收入国家。早期准确预测心脏病对于及时干预和改善患者预后至关重要。机器学习技术提供了有前途的解决方案；然而，诸如类不平衡、缺乏可解释性和有限的实际验证等挑战仍然存在。方法：在本研究中，使用从孟加拉国医疗机构收集的5000例患者记录的真实临床数据集开发了基于机器学习的心脏病预测框架。数据预处理包括清洗、特征编码、训练-测试分割和使用合成少数派过采样技术（SMOTE）处理类不平衡。多个机器学习模型-逻辑回归，决策树，支持向量机和随机森林-使用10倍分层交叉验证进行评估。采用准确性、精密度、召回率和f1评分来评估模型的性能。采用SHapley加性解释（SHapley Additive explanatory）提高模型的可解释性。表现最好的模型被部署为基于web的决策支持系统。结果：在评估的模型中，随机森林分类器的准确率为98%，召回率为96%，f1得分为96%。消融研究证明了SMOTE、特征集成和集成建模的有效性。SHAP分析确定了有助于心脏病预测的临床相关特征，提高了模型决策的透明度和信任度。结论：提出的框架为利用真实世界的临床数据预测心脏病提供了一个准确、可解释和实用的解决方案。可解释的机器学习和基于网络的部署的集成突出了其在临床决策支持方面的潜力。未来的工作将集中在多中心前瞻性验证和自适应模型更新上，以进一步提高泛化性和现实适用性。

{"title":"Explainable machine learning for early heart disease risk prediction: Insights from a clinical dataset in Bangladesh","authors":"Arpita Chakraborty , Utpol Kanti Das , Sadia Sazzad , Panna Das , Md. Mehedi Hassan Khan","doi":"10.1016/j.ibmed.2026.100352","DOIUrl":"10.1016/j.ibmed.2026.100352","url":null,"abstract":"<div><h3>Background:</h3><div>Cardiovascular diseases remain one of the leading causes of mortality worldwide, particularly in low- and middle-income countries. Early and accurate prediction of heart disease is essential for timely intervention and improved patient outcomes. Machine learning techniques offer promising solutions; however, challenges such as class imbalance, lack of interpretability, and limited real-world validation persist.</div></div><div><h3>Methods:</h3><div>In this study, a machine learning-based heart disease prediction framework was developed using a real-world clinical dataset comprising 5000 patient records collected from healthcare facilities in Bangladesh. Data preprocessing included cleaning, feature encoding, train–test splitting, and class imbalance handling using the Synthetic Minority Oversampling Technique (SMOTE). Multiple machine learning models — Logistic Regression, Decision Tree, Support Vector Machine, and Random Forest — were evaluated using 10-fold stratified cross-validation. Model performance was assessed using accuracy, precision, recall, and F1-score. SHAP (SHapley Additive exPlanations) was employed to enhance model interpretability. The best-performing model was deployed as a web-based decision support system.</div></div><div><h3>Results:</h3><div>Among the evaluated models, the Random Forest classifier achieved the best performance, with an accuracy of 98%, recall of 96%, and F1-score of 96%. Ablation studies demonstrated the effectiveness of SMOTE, feature integration, and ensemble modeling. SHAP analysis identified clinically relevant features contributing to heart disease prediction, enhancing transparency and trust in model decisions.</div></div><div><h3>Conclusions:</h3><div>The proposed framework provides an accurate, interpretable, and practical solution for heart disease prediction using real-world clinical data. The integration of explainable machine learning and web-based deployment highlights its potential for clinical decision support. Future work will focus on multi-center prospective validation and adaptive model updating to further improve generalizability and real-world applicability.</div></div>","PeriodicalId":73399,"journal":{"name":"Intelligence-based medicine","volume":"13 ","pages":"Article 100352"},"PeriodicalIF":0.0,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147396429","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 : 2026-03-01 Epub 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":"2026-03-01","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

DSCBAM-Net: A dual-attention deep learning framework for retinal vessel segmentation and feature-driven vascular analysis dscam - net：用于视网膜血管分割和特征驱动血管分析的双注意力深度学习框架

Intelligence-based medicine

Pub Date : 2026-03-01 Epub Date: 2026-03-02 DOI: 10.1016/j.ibmed.2026.100370

Rajatha , D.V. Ashoka

Retinal vessel segmentation is a crucial task in biomedical image analysis, enabling the understanding of vascular structures. Precise vessel segmentation reveals underlying vascular abnormalities in disease progression. Recent approaches have improved segmentation performance; however, maintaining computational efficiency while preserving accurate delineation of thin and low-contrast vessels remains an open engineering challenge. This study presents DSCBAM-Net, a lightweight dual-attention deep learning framework that integrates channel-wise and spatial attention mechanisms with multi-scale dilated convolution feature encoding to enhance vessel representation, particularly for thin and low-contrast vessel structures. The architecture follows a multi-stage design: (1) A robust encoder that extracts spatial and contextual features using standard and dilated convolutions enhanced with attention mechanisms. (2) A bottleneck module that fuses contextual features using parallel dilations. (3) A decoder with deep supervision for progressive vessel enhancement, enabling precise segmentation with fewer parameters and faster convergence. A composite hybrid loss function that integrates Dice, Focal-Tversky, and Top-k loss is introduced to address class imbalance and emphasize difficult vessel pixels. Following segmentation, various morphological features are extracted and used in a Retinal Vein Occlusion (RVO) detection and grading module, which estimates occlusion probability, stratifies the severity, and highlights dominant vascular drivers for explainability.

Extensive experiments were conducted on the merged dataset with an impressive dice score of 0.82. DSCBAM-Net also demonstrates superior cross-dataset performance, achieving a dice score of 0.879, 0.890, and 0.883 on DRIVE, STARE, and CHASE-DB1. In addition to segmentation, vessel-based structural features are extracted from the predicted masks to enable feature-driven vascular analysis and interpretability. Qualitative visualization, further highlights the effectiveness of the proposed architecture. Thus, DSCBAM-Net provides a robust and efficient solution for retinal vessel segmentation and downstream analytics tasks.

视网膜血管分割是生物医学图像分析中的一项重要任务，它使我们能够理解血管结构。精确的血管分割揭示了疾病进展中潜在的血管异常。最近的方法改进了分割性能；然而，保持计算效率，同时保持薄血管和低对比度血管的准确描绘仍然是一个开放的工程挑战。本研究提出了dscam - net，这是一个轻量级的双注意深度学习框架，它将通道和空间注意机制与多尺度扩展卷积特征编码集成在一起，以增强血管表征，特别是对于薄的和低对比度的血管结构。该架构遵循多阶段设计：(1)一个鲁棒编码器，使用标准卷积和扩展卷积增强了注意机制，提取空间和上下文特征。(2)瓶颈模块，使用并行扩展融合上下文特征。(3)具有深度监督的解码器，用于渐进式血管增强，以更少的参数和更快的收敛速度实现精确分割。引入了一个集成Dice、Focal-Tversky和Top-k损失的复合混合损失函数来解决职业不平衡问题，并强调困难的容器像素。在分割之后，提取各种形态特征并将其用于视网膜静脉闭塞（RVO）检测和分级模块，该模块可以估计闭塞概率，对严重程度进行分层，并突出显示主要的血管驱动因素以进行解释。在合并的数据集上进行了大量的实验，骰子得分达到了令人印象深刻的0.82。dscam - net还展示了卓越的跨数据集性能，在DRIVE、STARE和CHASE-DB1上的骰子得分分别为0.879、0.890和0.883。除了分割之外，还从预测的掩模中提取基于血管的结构特征，以实现特征驱动的血管分析和可解释性。定性可视化，进一步突出了所提出的体系结构的有效性。因此，dscam - net为视网膜血管分割和下游分析任务提供了一个强大而高效的解决方案。

{"title":"DSCBAM-Net: A dual-attention deep learning framework for retinal vessel segmentation and feature-driven vascular analysis","authors":"Rajatha , D.V. Ashoka","doi":"10.1016/j.ibmed.2026.100370","DOIUrl":"10.1016/j.ibmed.2026.100370","url":null,"abstract":"<div><div>Retinal vessel segmentation is a crucial task in biomedical image analysis, enabling the understanding of vascular structures. Precise vessel segmentation reveals underlying vascular abnormalities in disease progression. Recent approaches have improved segmentation performance; however, maintaining computational efficiency while preserving accurate delineation of thin and low-contrast vessels remains an open engineering challenge. This study presents DSCBAM-Net, a lightweight dual-attention deep learning framework that integrates channel-wise and spatial attention mechanisms with multi-scale dilated convolution feature encoding to enhance vessel representation, particularly for thin and low-contrast vessel structures. The architecture follows a multi-stage design: (1) A robust encoder that extracts spatial and contextual features using standard and dilated convolutions enhanced with attention mechanisms. (2) A bottleneck module that fuses contextual features using parallel dilations. (3) A decoder with deep supervision for progressive vessel enhancement, enabling precise segmentation with fewer parameters and faster convergence. A composite hybrid loss function that integrates Dice, Focal-Tversky, and Top-k loss is introduced to address class imbalance and emphasize difficult vessel pixels. Following segmentation, various morphological features are extracted and used in a Retinal Vein Occlusion (RVO) detection and grading module, which estimates occlusion probability, stratifies the severity, and highlights dominant vascular drivers for explainability.</div><div>Extensive experiments were conducted on the merged dataset with an impressive dice score of 0.82. DSCBAM-Net also demonstrates superior cross-dataset performance, achieving a dice score of 0.879, 0.890, and 0.883 on DRIVE, STARE, and CHASE-DB1. In addition to segmentation, vessel-based structural features are extracted from the predicted masks to enable feature-driven vascular analysis and interpretability. Qualitative visualization, further highlights the effectiveness of the proposed architecture. Thus, DSCBAM-Net provides a robust and efficient solution for retinal vessel segmentation and downstream analytics tasks.</div></div>","PeriodicalId":73399,"journal":{"name":"Intelligence-based medicine","volume":"13 ","pages":"Article 100370"},"PeriodicalIF":0.0,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147396424","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 the role of synthetic data in the future of AI in healthcare: A scoping review of frameworks, challenges, and implications 探索人工智能在医疗保健领域的未来中合成数据的作用：框架、挑战和影响的范围审查

Intelligence-based medicine

Pub Date : 2026-03-01 Epub Date: 2025-12-29 DOI: 10.1016/j.ibmed.2025.100342

Mohammad Ishtiaque Rahman , Md Razuan Hossain , S.M. Sayem , Forhan Bin Emdad

Synthetic data has emerged as a transformative tool in healthcare, particularly in areas such as medical imaging, electronic health records (EHRs), and clinical trial simulation, where data privacy, diversity, and accessibility are critical. This scoping review examines current approaches to synthetic data generation in healthcare, with a focus on AI model training, privacy preservation, and bias mitigation. A comprehensive search of PubMed, IEEE Xplore, and ACM Digital Library yielded 2906 studies, of which 42 met the inclusion criteria. Key data generation techniques included generative adversarial networks (GANs), variational autoencoders (VAEs), diffusion models, Bayesian networks, federated learning, recurrent neural networks (RNNs), large language models (LLMs), agent-based models, graph-based generators, and SMOTE-based oversampling. Applications ranged from diagnostic model development to privacy-preserving data sharing and educational simulation. However, the field faces persistent challenges, including inconsistent validation practices, the absence of standard benchmarks, high computational demands, and ethical concerns related to consent and bias. This review underscores the need for standardized evaluation protocols, clearer regulatory guidance, and multidisciplinary collaboration to ensure the safe, equitable, and effective use of synthetic data in healthcare AI. In addition to technical advances, the review highlights the socio-technical implications of synthetic data adoption, including its impact on health equity, patient trust, and clinical decision-making.

合成数据已成为医疗保健领域的变革性工具，特别是在医疗成像、电子健康记录（EHRs）和临床试验模拟等领域，这些领域的数据隐私、多样性和可访问性至关重要。本范围审查审查了医疗保健中合成数据生成的当前方法，重点是人工智能模型训练、隐私保护和偏见缓解。综合检索PubMed、IEEE explore和ACM数字图书馆得到2906篇研究，其中42篇符合纳入标准。关键的数据生成技术包括生成对抗网络（GANs）、变分自编码器（VAEs）、扩散模型、贝叶斯网络、联邦学习、循环神经网络（rnn）、大型语言模型（LLMs）、基于代理的模型、基于图的生成器和基于smote的过采样。应用范围从诊断模型开发到保护隐私的数据共享和教育模拟。然而，该领域面临着持续的挑战，包括不一致的验证实践、缺乏标准基准、高计算需求以及与同意和偏见相关的伦理问题。本综述强调需要标准化的评估方案、更清晰的监管指导和多学科合作，以确保在医疗保健人工智能中安全、公平和有效地使用合成数据。除了技术进步外，该审查还强调了采用综合数据的社会技术影响，包括其对卫生公平、患者信任和临床决策的影响。

{"title":"Exploring the role of synthetic data in the future of AI in healthcare: A scoping review of frameworks, challenges, and implications","authors":"Mohammad Ishtiaque Rahman , Md Razuan Hossain , S.M. Sayem , Forhan Bin Emdad","doi":"10.1016/j.ibmed.2025.100342","DOIUrl":"10.1016/j.ibmed.2025.100342","url":null,"abstract":"<div><div>Synthetic data has emerged as a transformative tool in healthcare, particularly in areas such as medical imaging, electronic health records (EHRs), and clinical trial simulation, where data privacy, diversity, and accessibility are critical. This scoping review examines current approaches to synthetic data generation in healthcare, with a focus on AI model training, privacy preservation, and bias mitigation. A comprehensive search of PubMed, IEEE Xplore, and ACM Digital Library yielded 2906 studies, of which 42 met the inclusion criteria. Key data generation techniques included generative adversarial networks (GANs), variational autoencoders (VAEs), diffusion models, Bayesian networks, federated learning, recurrent neural networks (RNNs), large language models (LLMs), agent-based models, graph-based generators, and SMOTE-based oversampling. Applications ranged from diagnostic model development to privacy-preserving data sharing and educational simulation. However, the field faces persistent challenges, including inconsistent validation practices, the absence of standard benchmarks, high computational demands, and ethical concerns related to consent and bias. This review underscores the need for standardized evaluation protocols, clearer regulatory guidance, and multidisciplinary collaboration to ensure the safe, equitable, and effective use of synthetic data in healthcare AI. In addition to technical advances, the review highlights the socio-technical implications of synthetic data adoption, including its impact on health equity, patient trust, and clinical decision-making.</div></div>","PeriodicalId":73399,"journal":{"name":"Intelligence-based medicine","volume":"13 ","pages":"Article 100342"},"PeriodicalIF":0.0,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145926925","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

Securing patient-specific ECG data in telemedicine through adaptive wavelet-based watermarking 通过自适应小波水印保护远程医疗中患者特定的心电数据

Intelligence-based medicine

Pub Date : 2026-03-01 Epub Date: 2026-02-02 DOI: 10.1016/j.ibmed.2026.100357

Rania Hamami , Narima Zermi , Larbi Boubchir , Amine Khaldi , Med Redouane Kafi , Aditya Kumar Sahu , Narimene Mimoune

Watermarking proves to be an effective technique for safeguarding crucial medical information. In this research, we propose a robust and imperceptible watermarking method designed to enhance the security of telemedicine-transmitted medical electrocardiogram (ECG) data. Embedding a mark in medical ECGs enables precise patient identification, reduces the risk of confusion during scans, and helps prevent diagnostic errors that could have adverse consequences. To ensure the security of ECG signals exchanged in telemedicine, our approach involves a frequency-domain watermarking method that conceals electronic patient records within the corresponding ECG signals. In this methodology, the signal undergoes a conversion into a 2D image, followed by a three-layer transform to extract the frequency content of the medical image. The low-frequency subbands undergo Schur decomposition, and the watermark bits are subsequently incorporated into the values of the upper triangular matrix. According to experimental results, these proposed techniques maintain a significant level of watermarked ECG quality while demonstrating high resistance to standard attacks. Experimental results show that the proposed SWT–Schur-based watermarking scheme achieves an average PSNR of 44.56 dB and an NCC higher than 0.95 under most common signal processing attacks. The average embedding capacity is 0.27 bits per pixel (BPP), while preserving the diagnostic quality of the ECG signals.

事实证明，水印是保护重要医疗信息的一种有效技术。在本研究中，我们提出了一种鲁棒且不易察觉的水印方法，旨在提高远程医疗传输的医疗心电图数据的安全性。在医用心电图中嵌入标记可以精确地识别患者，降低扫描期间混淆的风险，并有助于防止可能产生不良后果的诊断错误。为了确保远程医疗中交换的心电信号的安全性，我们的方法涉及到一种频域水印方法，该方法将电子病历隐藏在相应的心电信号中。在这种方法中，将信号转换成二维图像，然后进行三层变换以提取医学图像的频率内容。低频子带进行舒尔分解，水印比特随后被纳入上三角矩阵的值中。实验结果表明，这些技术在保持水印心电图质量的同时，对标准攻击具有很高的抵抗力。实验结果表明，在大多数常见的信号处理攻击下，基于swt - schur的水印方案的平均PSNR为44.56 dB， NCC大于0.95。在保证心电信号诊断质量的前提下，平均嵌入容量为0.27比特/像素（BPP）。

{"title":"Securing patient-specific ECG data in telemedicine through adaptive wavelet-based watermarking","authors":"Rania Hamami , Narima Zermi , Larbi Boubchir , Amine Khaldi , Med Redouane Kafi , Aditya Kumar Sahu , Narimene Mimoune","doi":"10.1016/j.ibmed.2026.100357","DOIUrl":"10.1016/j.ibmed.2026.100357","url":null,"abstract":"<div><div>Watermarking proves to be an effective technique for safeguarding crucial medical information. In this research, we propose a robust and imperceptible watermarking method designed to enhance the security of telemedicine-transmitted medical electrocardiogram (ECG) data. Embedding a mark in medical ECGs enables precise patient identification, reduces the risk of confusion during scans, and helps prevent diagnostic errors that could have adverse consequences. To ensure the security of ECG signals exchanged in telemedicine, our approach involves a frequency-domain watermarking method that conceals electronic patient records within the corresponding ECG signals. In this methodology, the signal undergoes a conversion into a 2D image, followed by a three-layer transform to extract the frequency content of the medical image. The low-frequency subbands undergo Schur decomposition, and the watermark bits are subsequently incorporated into the values of the upper triangular matrix. According to experimental results, these proposed techniques maintain a significant level of watermarked ECG quality while demonstrating high resistance to standard attacks. Experimental results show that the proposed SWT–Schur-based watermarking scheme achieves an average PSNR of 44.56 dB and an NCC higher than 0.95 under most common signal processing attacks. The average embedding capacity is 0.27 bits per pixel (BPP), while preserving the diagnostic quality of the ECG signals.</div></div>","PeriodicalId":73399,"journal":{"name":"Intelligence-based medicine","volume":"13 ","pages":"Article 100357"},"PeriodicalIF":0.0,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146188083","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 smart system for accurate detection and classification of cardio vascular diseases using advanced analysis 一个智能系统，用于准确检测和分类心血管疾病使用先进的分析

Intelligence-based medicine

Pub Date : 2026-03-01 Epub Date: 2026-01-22 DOI: 10.1016/j.ibmed.2026.100351

Sapana Bhushan Raghuwanshi, Nilesh Ashok Suryawanshi

Cardiovascular diseases (CVDs) are the leading cause of global mortality, highlighting the need for accurate and early diagnosis to enhance patient outcomes. However, effectively detecting and classifying CVDs using ECG signals remains a critical challenge. Hence, a novel Adaptive Temporal Wavelet-Kalman Fusion Transformer Network is proposed for enhanced CVD diagnosis. Initially, existing models struggle to capture nonlinear ventricular repolarization dynamics and subtle T-wave heterogeneity, leading to incomplete feature extraction. So, a novel Temporal Sparse Wavelet-Fourier Attention Transformer integrates Sparse Wavelet-Fourier Transform Decomposition to extract transient ECG variations, enhancing feature extraction. In contrast, Temporal Convolutional Transformers (TCTs) captures temporal dependencies for improved detection. Besides, existing ECG-based models fail to distinguish autonomic dysfunction (AD)-induced variations from pathological abnormalities, leading to misclassifications, and delayed diagnoses in CVD detection. Thus, an Adaptive Temporal Kalman Deep Fusion Forest Network integrates Adaptive Hybrid Empirical Kalman Decomposition to filter autonomic noise. At the same time, the Temporal Residual Gated Fusion Forest Network extracts spatial features, enhancing classification robustness. Further, the framework is validated using Stratified K-Fold Cross-Validation to ensure fairness and minimize bias, with experimental results demonstrating high accuracy, precision, and low MSE, leading to improved detection and classification of CVD.

心血管疾病是全球死亡的主要原因，因此需要进行准确和早期诊断，以提高患者的预后。然而，利用心电信号有效地检测和分类心血管疾病仍然是一个关键的挑战。为此，提出了一种新的自适应时域小波-卡尔曼融合变压器网络，用于增强CVD诊断。最初，现有模型难以捕捉非线性心室复极化动力学和微妙的t波异质性，导致特征提取不完整。为此，提出了一种新颖的时间稀疏小波-傅立叶注意转换器，结合稀疏小波-傅立叶变换分解提取瞬态心电变化，增强了特征提取能力。相比之下，时间卷积变压器（tct）捕获时间依赖性以改进检测。此外，现有的基于ecg的模型无法区分自主神经功能障碍（AD）引起的变异和病理异常，导致CVD检测中的错误分类和延迟诊断。因此，自适应时间卡尔曼深度融合森林网络集成了自适应混合经验卡尔曼分解来过滤自主噪声。同时，时间残差门控融合森林网络提取空间特征，增强分类鲁棒性。此外，使用分层K-Fold交叉验证对该框架进行了验证，以确保公平性和最小化偏差，实验结果表明，该框架具有较高的准确性、精密度和较低的MSE，从而提高了CVD的检测和分类。

{"title":"A smart system for accurate detection and classification of cardio vascular diseases using advanced analysis","authors":"Sapana Bhushan Raghuwanshi, Nilesh Ashok Suryawanshi","doi":"10.1016/j.ibmed.2026.100351","DOIUrl":"10.1016/j.ibmed.2026.100351","url":null,"abstract":"<div><div>Cardiovascular diseases (CVDs) are the leading cause of global mortality, highlighting the need for accurate and early diagnosis to enhance patient outcomes. However, effectively detecting and classifying CVDs using ECG signals remains a critical challenge. Hence, a novel <strong>Adaptive Temporal Wavelet-Kalman Fusion Transformer Network</strong> is proposed for enhanced CVD diagnosis. Initially, existing models struggle to capture nonlinear ventricular repolarization dynamics and subtle T-wave heterogeneity, leading to incomplete feature extraction. So, a novel Temporal Sparse Wavelet-Fourier Attention Transformer integrates Sparse Wavelet-Fourier Transform Decomposition to extract transient ECG variations, enhancing feature extraction. In contrast, Temporal Convolutional Transformers (TCTs) captures temporal dependencies for improved detection. Besides, existing ECG-based models fail to distinguish autonomic dysfunction (AD)-induced variations from pathological abnormalities, leading to misclassifications, and delayed diagnoses in CVD detection. Thus, an Adaptive Temporal Kalman Deep Fusion Forest Network integrates Adaptive Hybrid Empirical Kalman Decomposition to filter autonomic noise. At the same time, the Temporal Residual Gated Fusion Forest Network extracts spatial features, enhancing classification robustness. Further, the framework is validated using Stratified K-Fold Cross-Validation to ensure fairness and minimize bias, with experimental results demonstrating high accuracy, precision, and low MSE, leading to improved detection and classification of CVD.</div></div>","PeriodicalId":73399,"journal":{"name":"Intelligence-based medicine","volume":"13 ","pages":"Article 100351"},"PeriodicalIF":0.0,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146188916","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

LIO-VisionAR: Intelligence-enabled augmented reality guidance for laser indirect ophthalmoscope-based retinal laser therapy LIO-VisionAR：用于激光间接检眼镜视网膜激光治疗的智能增强现实指导

Intelligence-based medicine

Pub Date : 2026-03-01 Epub Date: 2026-01-28 DOI: 10.1016/j.ibmed.2026.100353

Sangjun Eom , Tiffany Ma , Miroslav Pajic , Maria Gorlatova , Majda Hadziahmetovic

Objective

Laser indirect ophthalmoscope (LIO) retinal therapy is a complex procedure that demands precision. We present LIO-VisionAR, an intelligence-enabled augmented reality (AR) guidance system designed to support safer and more effective training for LIO-based retinal laser therapy.

Methods

A custom retina model with retinopathy areas was developed and integrated into a human phantom model. A virtual retina model and simulator were developed using the color fundus photo to compute the magnification and laser targeting guidance based on the user's AR head-mounted device movement. Randomized user trials compared conventional and AR-guided retinal laser tasks, while multimodal behavioral telemetry were recorded for quantitative performance analysis and proof-of-concept skill inference.

Results

A total of 11 experts and 12 non-experts were included in the study. With AR guidance, laser targeting accuracy increased from 70.8 % to 82.6 % for experts and from 65.7 % to 81.7 % for non-experts. AR guidance increased laser instrumentation time, reflecting a deliberate speed–accuracy trade-off. Analysis of AR-captured behavioral telemetry showed that gaze exploration and temporal control features were associated with performance, and unsupervised clustering revealed distinct behavioral strategies linked to progressively higher accuracy. A composite performance-based skill score exhibited a moderate positive association with laser accuracy (Spearman ρ = 0.45, p = 0.032). Over 80 % of experts agreed that our system is appropriate for teaching and could improve retinal laser therapy training and safety.

Conclusions

LIO-VisionAR improves procedural accuracy under simulated conditions and demonstrates a concrete pathway toward adaptive, intelligence-based AR guidance for ophthalmic microsurgical training.

目的激光间接检眼镜（LIO）视网膜治疗是一项复杂的手术，要求精度高。我们提出了一种智能增强现实（AR）引导系统，旨在支持更安全，更有效的基于lio的视网膜激光治疗培训。方法建立具有视网膜病变区域的定制视网膜模型，并将其整合到人体幻影模型中。利用彩色眼底照片建立了虚拟视网膜模型和模拟器，计算了基于用户AR头戴式设备运动的放大倍率和激光瞄准制导。随机用户试验比较了传统和ar制导视网膜激光任务，同时记录了多模态行为遥测，用于定量性能分析和概念验证技能推断。结果共纳入专家11人，非专家12人。使用AR制导，专家激光瞄准精度从70.8%增加到82.6%，非专家激光瞄准精度从65.7%增加到81.7%。AR制导增加了激光仪器时间，反映了有意的速度精度权衡。ar捕捉的行为遥测分析表明，凝视探索和时间控制特征与性能相关，无监督聚类揭示了不同的行为策略，这些策略与逐渐提高的准确性相关。基于性能的综合技能得分与激光精度呈正相关（Spearman ρ = 0.45, p = 0.032）。超过80%的专家认为我们的系统适合教学，可以提高视网膜激光治疗的培训和安全性。结论slio - visionar提高了模拟条件下的操作精度，为自适应智能AR指导眼科显微外科训练提供了具体途径。

{"title":"LIO-VisionAR: Intelligence-enabled augmented reality guidance for laser indirect ophthalmoscope-based retinal laser therapy","authors":"Sangjun Eom , Tiffany Ma , Miroslav Pajic , Maria Gorlatova , Majda Hadziahmetovic","doi":"10.1016/j.ibmed.2026.100353","DOIUrl":"10.1016/j.ibmed.2026.100353","url":null,"abstract":"<div><h3>Objective</h3><div>Laser indirect ophthalmoscope (LIO) retinal therapy is a complex procedure that demands precision. We present LIO-VisionAR, an intelligence-enabled augmented reality (AR) guidance system designed to support safer and more effective training for LIO-based retinal laser therapy.</div></div><div><h3>Methods</h3><div>A custom retina model with retinopathy areas was developed and integrated into a human phantom model. A virtual retina model and simulator were developed using the color fundus photo to compute the magnification and laser targeting guidance based on the user's AR head-mounted device movement. Randomized user trials compared conventional and AR-guided retinal laser tasks, while multimodal behavioral telemetry were recorded for quantitative performance analysis and proof-of-concept skill inference.</div></div><div><h3>Results</h3><div>A total of 11 experts and 12 non-experts were included in the study. With AR guidance, laser targeting accuracy increased from 70.8 % to 82.6 % for experts and from 65.7 % to 81.7 % for non-experts. AR guidance increased laser instrumentation time, reflecting a deliberate speed–accuracy trade-off. Analysis of AR-captured behavioral telemetry showed that gaze exploration and temporal control features were associated with performance, and unsupervised clustering revealed distinct behavioral strategies linked to progressively higher accuracy. A composite performance-based skill score exhibited a moderate positive association with laser accuracy (Spearman ρ = 0.45, p = 0.032). Over 80 % of experts agreed that our system is appropriate for teaching and could improve retinal laser therapy training and safety.</div></div><div><h3>Conclusions</h3><div>LIO-VisionAR improves procedural accuracy under simulated conditions and demonstrates a concrete pathway toward adaptive, intelligence-based AR guidance for ophthalmic microsurgical training.</div></div>","PeriodicalId":73399,"journal":{"name":"Intelligence-based medicine","volume":"13 ","pages":"Article 100353"},"PeriodicalIF":0.0,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146188802","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

Optimization framework for overcoming tyrosine kinase inhibitor resistance: Multi-objective selection, scheduling, and adaptive therapy 克服酪氨酸激酶抑制剂耐药性的优化框架：多目标选择、调度和适应性治疗

Intelligence-based medicine

Pub Date : 2026-03-01 Epub Date: 2025-12-18 DOI: 10.1016/j.ibmed.2025.100332

Mohanad A. Deif , Mohamed A. Hafez , Mohammad Khishe

Tyrosine kinase inhibitors are key drugs in targeted cancer therapy but often fail when resistance emerges. Many predictive methods focus on accuracy alone, while calibration and kinase selectivity, which matter for clinical use, receive less attention. We present a single framework that treats resistance prediction and dosing decisions as a three-objective problem: minimize misclassification, reduce calibration error, and increase selectivity. Using calibrated probabilities and tuned thresholds, baseline models improved in ROC–AUC and expected calibration error across stratified, scaffold, and mutation-cold splits. Pareto analysis with hypervolume and coverage showed that including selectivity changes the relative ranking of inhibitors and exposes trade-offs that accuracy alone cannot capture. On the treatment side, we compared continuous dosing, hysteresis switching, and adaptive model predictive control in a two-compartment tumor model. Adaptive control lowered total dose by about 18% and extended simulated survival by more than 25 weeks. These results provide a clear proof of concept that combining machine learning, multi-objective optimization, and adaptive therapy can improve prediction quality and guide personalized dosing to better manage resistance.

酪氨酸激酶抑制剂是靶向癌症治疗的关键药物，但往往在出现耐药性时失效。许多预测方法只关注准确性，而校准和激酶选择性对临床使用很重要，却很少受到关注。我们提出了一个单一的框架，将耐药性预测和给药决策视为一个三目标问题：最大限度地减少错误分类，减少校准误差，增加选择性。使用校准概率和调整阈值，基线模型改进了ROC-AUC和分层、支架和突变冷分裂的预期校准误差。使用超容量和覆盖的帕累托分析表明，包括选择性改变了抑制剂的相对排名，并暴露了仅靠准确性无法捕获的权衡。在治疗方面，我们比较了两室肿瘤模型中的连续给药、迟滞开关和自适应模型预测控制。适应性控制将总剂量降低了18%，并将模拟生存期延长了25周以上。这些结果提供了一个清晰的概念证明，结合机器学习、多目标优化和自适应治疗可以提高预测质量，并指导个性化给药，以更好地管理耐药性。

{"title":"Optimization framework for overcoming tyrosine kinase inhibitor resistance: Multi-objective selection, scheduling, and adaptive therapy","authors":"Mohanad A. Deif , Mohamed A. Hafez , Mohammad Khishe","doi":"10.1016/j.ibmed.2025.100332","DOIUrl":"10.1016/j.ibmed.2025.100332","url":null,"abstract":"<div><div>Tyrosine kinase inhibitors are key drugs in targeted cancer therapy but often fail when resistance emerges. Many predictive methods focus on accuracy alone, while calibration and kinase selectivity, which matter for clinical use, receive less attention. We present a single framework that treats resistance prediction and dosing decisions as a three-objective problem: minimize misclassification, reduce calibration error, and increase selectivity. Using calibrated probabilities and tuned thresholds, baseline models improved in ROC–AUC and expected calibration error across stratified, scaffold, and mutation-cold splits. Pareto analysis with hypervolume and coverage showed that including selectivity changes the relative ranking of inhibitors and exposes trade-offs that accuracy alone cannot capture. On the treatment side, we compared continuous dosing, hysteresis switching, and adaptive model predictive control in a two-compartment tumor model. Adaptive control lowered total dose by about 18% and extended simulated survival by more than 25 weeks. These results provide a clear proof of concept that combining machine learning, multi-objective optimization, and adaptive therapy can improve prediction quality and guide personalized dosing to better manage resistance.</div></div>","PeriodicalId":73399,"journal":{"name":"Intelligence-based medicine","volume":"13 ","pages":"Article 100332"},"PeriodicalIF":0.0,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145926988","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 deep learning vision–language model for diagnosing pediatric dental diseases 儿童牙病诊断的深度学习视觉语言模型

Intelligence-based medicine

Pub Date : 2026-03-01 Epub Date: 2026-02-23 DOI: 10.1016/j.ibmed.2026.100364

Tuan D. Pham

Background:

Automated diagnosis of pediatric dental diseases from panoramic radiographs remains challenging due to anatomical variability and limited availability of specialist expertise. Vision–language models offer a potential approach by integrating visual and textual information to improve diagnostic performance and interpretability.

Objective:

To develop and evaluate a deep learning vision–language model for differentiating between caries and periapical infections in pediatric panoramic radiographs.

Methods:

A multimodal framework was proposed that combines visual features extracted from panoramic radiographs using non-linear dynamics and textural encoding with textual descriptions generated by a large language model. The fused multimodal representations were used to train a one-dimensional convolutional neural network classifier. Model performance was evaluated using accuracy, sensitivity, precision, F1 score, and area under the receiver operating characteristic curve (AUC).

Results:

Experiments conducted on a small, single-center dataset demonstrated that the proposed model outperformed conventional image-only convolutional neural networks and standalone language-based approaches, achieving an accuracy of 90%, sensitivity of 92%, specificity of 83%, precision of 92%, F1 score of 0.90, and an AUC of 0.96 within this dataset. However, the limited sample size and absence of external or prospective clinical validation restrict the generalizability and immediate clinical applicability of these findings.

Conclusions:

The results suggest that integrating visual and textual representations can enhance diagnostic performance for pediatric dental disease classification. Nevertheless, the findings should be regarded as preliminary and hypothesis-generating. Future work will involve larger, multi-center studies, external validation, and prospective clinical evaluation to establish robustness, generalizability, and real-world clinical impact of vision–language models in pediatric dental diagnostics.

背景：由于解剖结构的变异性和专家技术的有限性，从全景x线片自动诊断儿童牙科疾病仍然具有挑战性。视觉语言模型通过集成视觉和文本信息来提高诊断性能和可解释性，提供了一种潜在的方法。目的：建立并评价一种用于儿童全景x线片龋齿和根尖周感染鉴别的深度学习视觉语言模型。方法：提出了一种多模态框架，将利用非线性动力学和纹理编码提取的全景x线照片视觉特征与大型语言模型生成的文本描述相结合。利用融合的多模态表示训练一维卷积神经网络分类器。通过准确度、灵敏度、精密度、F1评分和受试者工作特征曲线下面积来评价模型的性能。结果：在一个小型单中心数据集上进行的实验表明，所提出的模型优于传统的仅图像卷积神经网络和基于独立语言的方法，在该数据集内实现了90%的准确率，92%的灵敏度，83%的特异性，92%的精度，F1得分为0.90，AUC为0.96。然而，有限的样本量和缺乏外部或前瞻性临床验证限制了这些发现的普遍性和直接临床适用性。结论：视觉与文字相结合的方法可以提高儿童牙病分类的诊断效能。然而，这些发现应该被认为是初步的和产生假设的。未来的工作将包括更大的、多中心的研究、外部验证和前瞻性临床评估，以建立视觉语言模型在儿童牙科诊断中的稳健性、通用性和现实世界的临床影响。

{"title":"A deep learning vision–language model for diagnosing pediatric dental diseases","authors":"Tuan D. Pham","doi":"10.1016/j.ibmed.2026.100364","DOIUrl":"10.1016/j.ibmed.2026.100364","url":null,"abstract":"<div><h3>Background:</h3><div>Automated diagnosis of pediatric dental diseases from panoramic radiographs remains challenging due to anatomical variability and limited availability of specialist expertise. Vision–language models offer a potential approach by integrating visual and textual information to improve diagnostic performance and interpretability.</div></div><div><h3>Objective:</h3><div>To develop and evaluate a deep learning vision–language model for differentiating between caries and periapical infections in pediatric panoramic radiographs.</div></div><div><h3>Methods:</h3><div>A multimodal framework was proposed that combines visual features extracted from panoramic radiographs using non-linear dynamics and textural encoding with textual descriptions generated by a large language model. The fused multimodal representations were used to train a one-dimensional convolutional neural network classifier. Model performance was evaluated using accuracy, sensitivity, precision, F1 score, and area under the receiver operating characteristic curve (AUC).</div></div><div><h3>Results:</h3><div>Experiments conducted on a small, single-center dataset demonstrated that the proposed model outperformed conventional image-only convolutional neural networks and standalone language-based approaches, achieving an accuracy of 90%, sensitivity of 92%, specificity of 83%, precision of 92%, F1 score of 0.90, and an AUC of 0.96 within this dataset. However, the limited sample size and absence of external or prospective clinical validation restrict the generalizability and immediate clinical applicability of these findings.</div></div><div><h3>Conclusions:</h3><div>The results suggest that integrating visual and textual representations can enhance diagnostic performance for pediatric dental disease classification. Nevertheless, the findings should be regarded as preliminary and hypothesis-generating. Future work will involve larger, multi-center studies, external validation, and prospective clinical evaluation to establish robustness, generalizability, and real-world clinical impact of vision–language models in pediatric dental diagnostics.</div></div>","PeriodicalId":73399,"journal":{"name":"Intelligence-based medicine","volume":"13 ","pages":"Article 100364"},"PeriodicalIF":0.0,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147396307","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

CCSA-RF enhanced lightweight multi-scale CNN framework for robust lung cancer classification CCSA-RF增强轻量级多尺度CNN框架的鲁棒肺癌分类

Intelligence-based medicine

Pub Date : 2026-03-01 Epub Date: 2026-02-21 DOI: 10.1016/j.ibmed.2026.100363

R. Gayathri, R. Thilagavathy

The early and accurate diagnosis of lung cancer is an urgent issue owing to the noise of imaging, nonhomogeneous nodule structures, and high cost of computing the current deep learning architectures. Most modern methods find it difficult to strike a balance between diagnostic accuracy, strength, and efficiency, especially with multimodal medical images. To overcome these shortcomings, this study proposes a single lightweight-based learning framework that combines preprocessing using wavelets, feature extraction under attention, chaos-based feature selection, and multiscale convolutional learning. First, a Ricker wavelet invariant center-weighted mean (RWICWM) filter was used to reduce noise without unimportant structures in the diagnostics. An Attention-Based DenseNet (ATT-DenseNet) is then used to extract discriminative features by highlighting the parts of the image that contain cancer-related features with channel-wise attention. The chaotic convolutional spectral analysis random forest (CCSA-RF) mechanism was proposed to select informative features and eliminate redundancy. Finally, a lightweight multi-scale CNN (LMS-CNN) is employed to achieve effective and precise classification of lung cancer. On the LC25000 and LIDC-IDRI datasets, the proposed framework attained an accuracy of 96.54, recall of 96.79, precision of 96.90, and F1-score of 96.32, with both datasets reporting similar results. The findings indicate that the proposed method enhances the classification accuracy and strength and still has a low computational complexity, which renders it applicable for real-world clinical use. The proposed research offers a computationally and theoretically sound solution that sets the state-of-the-art in lung cancer diagnostics.

由于目前的深度学习架构存在成像噪声、结节结构不均匀以及计算成本高等问题，早期准确诊断肺癌是一个迫切需要解决的问题。大多数现代方法发现很难在诊断准确性、强度和效率之间取得平衡，特别是对于多模态医学图像。为了克服这些缺点，本研究提出了一种基于轻量级的学习框架，该框架结合了小波预处理、关注下的特征提取、基于混沌的特征选择和多尺度卷积学习。首先，采用Ricker小波不变中心加权均值（RWICWM）滤波，在诊断过程中剔除不重要结构的噪声；然后使用基于注意力的DenseNet （at -DenseNet）通过突出显示图像中包含与通道相关的癌症相关特征的部分来提取判别特征。提出了混沌卷积谱分析随机森林（CCSA-RF）机制来选择信息特征并消除冗余。最后，采用轻量级多尺度CNN （LMS-CNN）对肺癌进行有效、精确的分类。在LC25000和LIDC-IDRI数据集上，该框架的准确率为96.54，召回率为96.79，精密度为96.90，f1得分为96.32，两个数据集报告的结果相似。结果表明，该方法在提高分类精度和强度的同时，仍然具有较低的计算复杂度，适用于实际临床应用。拟议的研究提供了一个计算和理论上合理的解决方案，设置最先进的肺癌诊断。

{"title":"CCSA-RF enhanced lightweight multi-scale CNN framework for robust lung cancer classification","authors":"R. Gayathri, R. Thilagavathy","doi":"10.1016/j.ibmed.2026.100363","DOIUrl":"10.1016/j.ibmed.2026.100363","url":null,"abstract":"<div><div>The early and accurate diagnosis of lung cancer is an urgent issue owing to the noise of imaging, nonhomogeneous nodule structures, and high cost of computing the current deep learning architectures. Most modern methods find it difficult to strike a balance between diagnostic accuracy, strength, and efficiency, especially with multimodal medical images. To overcome these shortcomings, this study proposes a single lightweight-based learning framework that combines preprocessing using wavelets, feature extraction under attention, chaos-based feature selection, and multiscale convolutional learning. First, a Ricker wavelet invariant center-weighted mean (RWICWM) filter was used to reduce noise without unimportant structures in the diagnostics. An Attention-Based DenseNet (ATT-DenseNet) is then used to extract discriminative features by highlighting the parts of the image that contain cancer-related features with channel-wise attention. The chaotic convolutional spectral analysis random forest (CCSA-RF) mechanism was proposed to select informative features and eliminate redundancy. Finally, a lightweight multi-scale CNN (LMS-CNN) is employed to achieve effective and precise classification of lung cancer. On the LC25000 and LIDC-IDRI datasets, the proposed framework attained an accuracy of 96.54, recall of 96.79, precision of 96.90, and F1-score of 96.32, with both datasets reporting similar results. The findings indicate that the proposed method enhances the classification accuracy and strength and still has a low computational complexity, which renders it applicable for real-world clinical use. The proposed research offers a computationally and theoretically sound solution that sets the state-of-the-art in lung cancer diagnostics.</div></div>","PeriodicalId":73399,"journal":{"name":"Intelligence-based medicine","volume":"13 ","pages":"Article 100363"},"PeriodicalIF":0.0,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147396308","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