Computer methods and programs in biomedicine最新文献_第9页

Development of advanced lung cancer inflammation index-based machine learning models for predicting stroke and mortality: A comparative and interpretable study 基于晚期肺癌炎症指数的预测中风和死亡率的机器学习模型的发展：一项比较和可解释的研究。

IF 4.8 2区医学 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computer methods and programs in biomedicine

Pub Date : 2026-02-01 Epub Date: 2025-12-03 DOI: 10.1016/j.cmpb.2025.109201

Jiaxin Fan , Xingzhi Guo , Shuai Cao , Rui Li , Shuqin Zhan

Background

The Advanced Lung Cancer Inflammation Index (ALI) is a novel composite index that enables a more holistic evaluation of inflammation and nutritional status than established single or commonly used indices. However, ALI has not been extensively studied in patients with stroke. In this study, we aimed to investigate: 1) the association between ALI and stroke risk and 2) the association between ALI and all-cause mortality among patients with stroke, as well as 3) develop and interpret machine learning (ML) models to predict stroke and prognosis.

Methods

Using data from the National Health and Nutrition Examination Survey (NHANES) 1999–2018, logistic regression and Cox regression assessed associations of ALI with stroke and mortality. Non-linear relationships were analysed using restricted cubic spline and subgroup stratification. Logistic regression (LR), extreme gradient boosting (XGBoost), random forest (RF), K-Nearest Neighbor (KNN), supported vector machine (SVM), and decision tree (DT) were developed for stroke and mortality prediction and evaluated using the area under the receiver operating characteristic curve (AUCROC), and metrics such as accuracy. Shapley additive explanations (SHAP) and Gini importance enhanced the dual-interpretability of the model.

Results

Among the 46,451 participants, higher ALI was associated with a lower stroke risk, whereas mortality decreased with increasing ALI before stabilising at an inflection point (ALI = 40.91, P _threshold < 0.001). Age stratification significantly modified the association between ALI and mortality. The RF model marginally outperformed the other models in terms of stroke identification (AUCROC: 0.9657, accuracy: 95.63 %) and mortality prediction (AUCROC: 0.7771, accuracy: 70.65 %). The SHAP and Gini importance analyses highlighted cardiovascular diseases as key factors for stroke prediction and age for mortality, with ALI being less influential.

Conclusions

Leveraging the nationally representative NHANES database, this exploratory analysis revealed that ALI presented a reverse dose-response association with the stroke risk and an “L-shaped” relationship with all-cause mortality among patients with stroke. Dual-interpretable RF models based on the ALI showed comparably promising potential among the six ML models for stroke identification and prognosis prediction.

背景：晚期肺癌炎症指数（ALI）是一种新型的复合指数，能够比现有的单一或常用指数更全面地评估炎症和营养状况。然而，在卒中患者中，ALI尚未得到广泛的研究。在这项研究中，我们旨在探讨：1)ALI与卒中风险之间的关系；2)ALI与卒中患者全因死亡率之间的关系；以及3)开发和解释机器学习（ML）模型来预测卒中和预后。方法：利用1999-2018年国家健康与营养调查（NHANES）的数据，采用logistic回归和Cox回归评估ALI与卒中和死亡率的关系。非线性关系分析采用限制三次样条和亚群分层。采用Logistic回归（LR）、极端梯度增强（XGBoost）、随机森林（RF）、k -最近邻（KNN）、支持向量机（SVM）和决策树（DT）进行脑卒中和死亡率预测，并利用受试者工作特征曲线下面积（AUCROC）和准确性等指标进行评估。Shapley加性解释（SHAP）和基尼重要性增强了模型的双重可解释性。结果：在46,451名参与者中，较高的ALI与较低的卒中风险相关，而死亡率随着ALI的增加而下降，然后在拐点稳定（ALI = 40.91， p阈值< 0.001）。年龄分层显著改变了ALI与死亡率之间的关系。RF模型在卒中识别（AUCROC: 0.9657，准确率：95.63%）和死亡率预测（AUCROC: 0.7771，准确率：70.65%）方面略优于其他模型。SHAP和基尼重要性分析强调心血管疾病是中风预测和年龄死亡率的关键因素，ALI的影响较小。结论：利用具有全国代表性的NHANES数据库，本探索性分析显示ALI与卒中风险呈反向剂量反应相关性，与卒中患者的全因死亡率呈“l型”关系。在6种ML模型中，基于ALI的双可解释RF模型在卒中识别和预后预测方面具有相当大的潜力。

{"title":"Development of advanced lung cancer inflammation index-based machine learning models for predicting stroke and mortality: A comparative and interpretable study","authors":"Jiaxin Fan , Xingzhi Guo , Shuai Cao , Rui Li , Shuqin Zhan","doi":"10.1016/j.cmpb.2025.109201","DOIUrl":"10.1016/j.cmpb.2025.109201","url":null,"abstract":"<div><h3>Background</h3><div>The Advanced Lung Cancer Inflammation Index (ALI) is a novel composite index that enables a more holistic evaluation of inflammation and nutritional status than established single or commonly used indices. However, ALI has not been extensively studied in patients with stroke. In this study, we aimed to investigate: 1) the association between ALI and stroke risk and 2) the association between ALI and all-cause mortality among patients with stroke, as well as 3) develop and interpret machine learning (ML) models to predict stroke and prognosis.</div></div><div><h3>Methods</h3><div>Using data from the National Health and Nutrition Examination Survey (NHANES) 1999–2018, logistic regression and Cox regression assessed associations of ALI with stroke and mortality. Non-linear relationships were analysed using restricted cubic spline and subgroup stratification. Logistic regression (LR), extreme gradient boosting (XGBoost), random forest (RF), K-Nearest Neighbor (KNN), supported vector machine (SVM), and decision tree (DT) were developed for stroke and mortality prediction and evaluated using the area under the receiver operating characteristic curve (AUCROC), and metrics such as accuracy. Shapley additive explanations (SHAP) and Gini importance enhanced the dual-interpretability of the model.</div></div><div><h3>Results</h3><div>Among the 46,451 participants, higher ALI was associated with a lower stroke risk, whereas mortality decreased with increasing ALI before stabilising at an inflection point (ALI = 40.91, <em>P</em> <sub>threshold</sub> < 0.001). Age stratification significantly modified the association between ALI and mortality. The RF model marginally outperformed the other models in terms of stroke identification (AUCROC: 0.9657, accuracy: 95.63 %) and mortality prediction (AUCROC: 0.7771, accuracy: 70.65 %). The SHAP and Gini importance analyses highlighted cardiovascular diseases as key factors for stroke prediction and age for mortality, with ALI being less influential.</div></div><div><h3>Conclusions</h3><div>Leveraging the nationally representative NHANES database, this exploratory analysis revealed that ALI presented a reverse dose-response association with the stroke risk and an “L-shaped” relationship with all-cause mortality among patients with stroke. Dual-interpretable RF models based on the ALI showed comparably promising potential among the six ML models for stroke identification and prognosis prediction.</div></div>","PeriodicalId":10624,"journal":{"name":"Computer methods and programs in biomedicine","volume":"275 ","pages":"Article 109201"},"PeriodicalIF":4.8,"publicationDate":"2026-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145721423","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Adapt or specialize? A comprehensive evaluation of adapted SAM versus task-specific CNNs for fetal abdominal segmentation 适应还是专业化？对胎儿腹部分割的适应性SAM与任务特异性cnn的综合评估

IF 4.8 2区医学 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computer methods and programs in biomedicine

Pub Date : 2026-02-01 Epub Date: 2025-11-26 DOI: 10.1016/j.cmpb.2025.109178

Maria Chiara Fiorentino , Lorenzo Federici , Alessandro Pietro La Camera , Enrico Gianluca Caiani

Background:

The fetal abdomen is crucial in prenatal screening, offering key insights into fetal growth and congenital anomalies. However, segmenting internal abdominal structures in ultrasound (US) remains challenging due to anatomical variability, overlapping organs, and low contrast. While CNN-based models have shown strong performance in fetal head analysis, most existing methods focus on biometric measurements (e.g., head or abdominal circumference), leaving internal abdominal organ segmentation largely underexplored. Recently, foundation models like the Segment Anything Model (SAM) have emerged as flexible alternatives, enabling zero- or few-shot segmentation. Yet, their performance on fetal US remains poorly understood, and the need for adaptation is still an open question.

Methods:

We compare two segmentation strategies: (1) task-specific CNNs, including UNet, Attention UNet, nnUNet, DeepLabv3+, and their focal-loss variants; and (2) adapted SAM-based models. The latter includes zero-shot variants (SAMPoint, SAMBBox), pre-trained models (SAM Med2DBBox, MedSAMBBox), and adapted configurations, including lightweight fine-tuned models (SAM-LoRA, MedSAM-FrozenEncoder) and SAM Med2D variants with adapted layers. Experiments are conducted on a curated dataset of fetal abdominal US images with manual segmentations of the liver, stomach, artery, and umbilical vein. Performance is evaluated using Dice Similarity Coefficient, Intersection over Union, and precision. Statistical significance is assessed via pairwise Friedman chi-square tests.

Results & Conclusions:

Zero-shot SAM variants performed poorly, particularly on small or low-contrast structures. In contrast, adapted SAM models consistently outperformed CNNs, reaching DSC scores up to 0.90 (liver) and 0.80 (artery). Prompt-based interaction enables semi-automated, human-in-the-loop workflows, supporting clinical applicability.

背景：胎儿腹部在产前筛查中是至关重要的，为胎儿生长和先天性异常提供了关键的见解。然而，由于解剖变异、器官重叠和低对比度，在超声（US）中分割腹部内部结构仍然具有挑战性。虽然基于cnn的模型在胎儿头部分析中表现出色，但大多数现有方法都侧重于生物特征测量（例如头部或腹部围），而对腹部内部器官分割的探索很大程度上不足。最近，像分段任意模型（SAM）这样的基础模型已经作为灵活的替代方案出现，可以实现零次或几次分段。然而，他们对胎儿US的表现仍然知之甚少，适应的必要性仍然是一个悬而未决的问题。方法：我们比较了两种分割策略：(1)针对特定任务的cnn，包括UNet、Attention UNet、nnUNet、DeepLabv3+及其焦点丢失变体；(2)适应的基于sam的模型。后者包括零射击变体（SAMPoint, SAMBBox），预训练模型（SAM Med2DBBox, MedSAMBBox）和适应配置，包括轻量级微调模型（SAM- lora, MedSAM-FrozenEncoder）和具有适应层的SAM Med2D变体。实验是在一个精心策划的胎儿腹部US图像数据集上进行的，该数据集具有人工分割的肝脏、胃、动脉和脐静脉。性能评估使用骰子相似系数，交集超过联合，和精度。通过两两Friedman卡方检验评估统计显著性。结果&结论：零射击SAM变体表现不佳，特别是在小或低对比度结构上。相比之下，适应性SAM模型的表现一直优于cnn， DSC得分高达0.90（肝脏）和0.80（动脉）。基于提示的交互实现了半自动化、人在循环的工作流程，支持临床应用。

{"title":"Adapt or specialize? A comprehensive evaluation of adapted SAM versus task-specific CNNs for fetal abdominal segmentation","authors":"Maria Chiara Fiorentino , Lorenzo Federici , Alessandro Pietro La Camera , Enrico Gianluca Caiani","doi":"10.1016/j.cmpb.2025.109178","DOIUrl":"10.1016/j.cmpb.2025.109178","url":null,"abstract":"<div><h3>Background:</h3><div>The fetal abdomen is crucial in prenatal screening, offering key insights into fetal growth and congenital anomalies. However, segmenting internal abdominal structures in ultrasound (US) remains challenging due to anatomical variability, overlapping organs, and low contrast. While CNN-based models have shown strong performance in fetal head analysis, most existing methods focus on biometric measurements (e.g., head or abdominal circumference), leaving internal abdominal organ segmentation largely underexplored. Recently, foundation models like the Segment Anything Model (SAM) have emerged as flexible alternatives, enabling zero- or few-shot segmentation. Yet, their performance on fetal US remains poorly understood, and the need for adaptation is still an open question.</div></div><div><h3>Methods:</h3><div>We compare two segmentation strategies: (1) task-specific CNNs, including UNet, Attention UNet, nnUNet, DeepLabv3+, and their focal-loss variants; and (2) adapted SAM-based models. The latter includes zero-shot variants (SAMPoint, SAMBBox), pre-trained models (SAM Med2DBBox, MedSAMBBox), and adapted configurations, including lightweight fine-tuned models (SAM-LoRA, MedSAM-FrozenEncoder) and SAM Med2D variants with adapted layers. Experiments are conducted on a curated dataset of fetal abdominal US images with manual segmentations of the liver, stomach, artery, and umbilical vein. Performance is evaluated using Dice Similarity Coefficient, Intersection over Union, and precision. Statistical significance is assessed via pairwise Friedman chi-square tests.</div></div><div><h3>Results & Conclusions:</h3><div>Zero-shot SAM variants performed poorly, particularly on small or low-contrast structures. In contrast, adapted SAM models consistently outperformed CNNs, reaching DSC scores up to 0.90 (liver) and 0.80 (artery). Prompt-based interaction enables semi-automated, human-in-the-loop workflows, supporting clinical applicability.</div></div>","PeriodicalId":10624,"journal":{"name":"Computer methods and programs in biomedicine","volume":"275 ","pages":"Article 109178"},"PeriodicalIF":4.8,"publicationDate":"2026-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145600477","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

The ICeX framework: Brain age estimation from thalamic nuclei with conformalized and eXplainable random forest regression ICeX框架：丘脑核的脑年龄估计与符合化和可解释的随机森林回归。

IF 4.8 2区医学 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computer methods and programs in biomedicine

Pub Date : 2026-02-01 Epub Date: 2025-11-01 DOI: 10.1016/j.cmpb.2025.109140

Alessia Sarica , Chiara Camastra , Assunta Pelagi , Fulvia Arcuri , Fabiana Novellino , Aldo Quattrone , Andrea Quattrone

Background and Objective

Accurate estimation of brain age is essential to identify deviations from typical aging trajectories, which may signal early neurodegenerative, psychiatric, or cognitive dysfunctions. Traditional models often rely on global brain features, potentially overlooking subtle region-specific alterations. This study introduces a novel framework for brain age prediction based on thalamic nuclei volumes—anatomically and functionally distinct structures known to be sensitive to aging. The objective is to enhance interpretability and reliability by integrating feature attribution and uncertainty quantification.

Methods

A dataset of 630 healthy young adults from the Human Connectome Project was used to train a Random Forest Regression model. Thalamic nuclei volumes were extracted from structural MRI scans. To enhance transparency, the model was coupled with SHAP-based feature attribution and Conformal Prediction to generate subject-specific prediction intervals. These components were combined into a unified approach called Individual Conformalized Explanations (ICeX), providing insights into how individual features influence both predicted brain age and its associated uncertainty.

Results

The model achieved a mean absolute error of 2.77 years and a 90.77 % coverage, with an average prediction interval width of 11.03 years. Key contributors to prediction accuracy included the left Lateral Geniculate, left Paratenial, and right Ventromedial nuclei. ICeX enabled a detailed understanding of how each feature influenced both prediction and uncertainty at the individual level.

Conclusions

The proposed framework provides reliable and interpretable brain age predictions by combining region-specific biomarkers with robust uncertainty quantification. ICeX offers clinicians and researchers a powerful tool for exploring individual aging trajectories with both precision and transparency, supporting future applications in early detection and personalized intervention strategies for age-related neurological conditions.

背景和目的：准确估计脑年龄对于识别典型衰老轨迹的偏差至关重要，这可能是早期神经退行性、精神或认知功能障碍的信号。传统的模型通常依赖于整体的大脑特征，可能忽略了细微的特定区域的变化。本研究提出了一种基于丘脑核体积的大脑年龄预测的新框架，丘脑核体积是一种在解剖学和功能上不同的结构，已知对衰老敏感。目标是通过整合特征属性和不确定性量化来提高可解释性和可靠性。方法：使用人类连接组计划中630名健康年轻人的数据集训练随机森林回归模型。从结构MRI扫描中提取丘脑核体积。为了提高透明度，将模型与基于shap的特征属性和保形预测相结合，生成特定于受试者的预测区间。这些成分被组合成一种统一的方法，称为个体归化解释（ICeX），为个体特征如何影响预测的大脑年龄及其相关的不确定性提供了见解。结果：模型平均绝对误差为2.77年，覆盖率为90.77%，平均预测区间宽度为11.03年。预测准确性的关键因素包括左侧膝状外侧核、左侧腱旁核和右侧腹内侧核。ICeX能够在个体层面上详细了解每个特征如何影响预测和不确定性。结论：该框架通过结合区域特异性生物标志物和稳健的不确定性量化，提供了可靠且可解释的脑年龄预测。ICeX为临床医生和研究人员提供了一个强大的工具，可以精确和透明地探索个体衰老轨迹，支持未来在与年龄相关的神经系统疾病的早期检测和个性化干预策略中的应用。

{"title":"The ICeX framework: Brain age estimation from thalamic nuclei with conformalized and eXplainable random forest regression","authors":"Alessia Sarica , Chiara Camastra , Assunta Pelagi , Fulvia Arcuri , Fabiana Novellino , Aldo Quattrone , Andrea Quattrone","doi":"10.1016/j.cmpb.2025.109140","DOIUrl":"10.1016/j.cmpb.2025.109140","url":null,"abstract":"<div><h3>Background and Objective</h3><div>Accurate estimation of brain age is essential to identify deviations from typical aging trajectories, which may signal early neurodegenerative, psychiatric, or cognitive dysfunctions. Traditional models often rely on global brain features, potentially overlooking subtle region-specific alterations. This study introduces a novel framework for brain age prediction based on thalamic nuclei volumes—anatomically and functionally distinct structures known to be sensitive to aging. The objective is to enhance interpretability and reliability by integrating feature attribution and uncertainty quantification.</div></div><div><h3>Methods</h3><div>A dataset of 630 healthy young adults from the Human Connectome Project was used to train a Random Forest Regression model. Thalamic nuclei volumes were extracted from structural MRI scans. To enhance transparency, the model was coupled with SHAP-based feature attribution and Conformal Prediction to generate subject-specific prediction intervals. These components were combined into a unified approach called Individual Conformalized Explanations (ICeX), providing insights into how individual features influence both predicted brain age and its associated uncertainty.</div></div><div><h3>Results</h3><div>The model achieved a mean absolute error of 2.77 years and a 90.77 % coverage, with an average prediction interval width of 11.03 years. Key contributors to prediction accuracy included the left Lateral Geniculate, left Paratenial, and right Ventromedial nuclei. ICeX enabled a detailed understanding of how each feature influenced both prediction and uncertainty at the individual level.</div></div><div><h3>Conclusions</h3><div>The proposed framework provides reliable and interpretable brain age predictions by combining region-specific biomarkers with robust uncertainty quantification. ICeX offers clinicians and researchers a powerful tool for exploring individual aging trajectories with both precision and transparency, supporting future applications in early detection and personalized intervention strategies for age-related neurological conditions.</div></div>","PeriodicalId":10624,"journal":{"name":"Computer methods and programs in biomedicine","volume":"274 ","pages":"Article 109140"},"PeriodicalIF":4.8,"publicationDate":"2026-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145457852","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Revealing EEG signatures of intervention in disorder of consciousness using artificial intelligence: methodology and feasibility 利用人工智能揭示意识障碍干预的脑电图特征：方法与可行性

IF 4.8 2区医学 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computer methods and programs in biomedicine

Pub Date : 2026-02-01 Epub Date: 2025-11-09 DOI: 10.1016/j.cmpb.2025.109159

Davide Borra , Valentina Bonsangue , Sofia Straudi , Elisa Magosso

Background and Objective

Electroencephalography (EEG) is a crucial tool for monitoring recovery in patients with disorders of consciousness (DOC) after therapeutic interventions. It helps in identifying the neural correlates and in guiding the development of personalized treatments. Spectrum power measures are widely employed. However, these measures are manually handcrafted, not patient-specific, and not tailored to the specific intervention.

Methods

To address these limitations, we propose an explainable artificial intelligence (XAI) framework designed to automatically uncover the most salient frequency-domain EEG signatures in an intervention- and patient-specific manner. The framework integrates an interpretable convolutional neural network, which is capable of learning interpretable frequency-domain EEG features, with an explanation technique, which quantifies the relevance of the learned spectral features. This approach enables the automatic tracking of patient-specific spectral EEG changes and refines the analysis toward neural features that are more closely associated with key clinical variables.

Results

We showcase the potential of our approach by applying it to EEG signals collected from patients in a minimally conscious state following an intervention based on transcranial direct current stimulation. The XAI results reveal a prominent role of alpha-band EEG oscillations in DOC intervention, supporting evidence that functional improvements due to intervention are associated with an increase in alpha-band spectral content.

Conclusions

Our XAI-driven analysis offers a robust, individualized, and transparent alternative (or complement) to conventional EEG analyses, thereby enhancing the EEG characterization of DOC patients.

背景与目的脑电图（EEG）是监测意识障碍（DOC）患者治疗干预后恢复情况的重要工具。它有助于识别神经关联并指导个性化治疗的发展。频谱功率测量被广泛应用。然而，这些措施是手工制作的，不是针对特定患者的，也不是针对特定干预措施量身定制的。为了解决这些限制，我们提出了一个可解释的人工智能（XAI）框架，旨在以干预和患者特定的方式自动发现最显著的频域EEG特征。该框架将可解释的卷积神经网络（能够学习可解释的频域EEG特征）与解释技术（量化学习到的频谱特征的相关性）相结合。这种方法能够自动跟踪患者特定的频谱脑电图变化，并将分析细化到与关键临床变量更密切相关的神经特征。结果：我们将该方法应用于经颅直流电刺激干预后的最低意识状态患者的脑电图信号，展示了该方法的潜力。XAI结果揭示了α波段脑电图振荡在DOC干预中的突出作用，支持了干预导致的功能改善与α波段频谱含量增加相关的证据。结论基于ai驱动的分析为传统脑电图分析提供了一种可靠的、个性化的、透明的替代（或补充）方法，从而增强了DOC患者的脑电图特征。

{"title":"Revealing EEG signatures of intervention in disorder of consciousness using artificial intelligence: methodology and feasibility","authors":"Davide Borra , Valentina Bonsangue , Sofia Straudi , Elisa Magosso","doi":"10.1016/j.cmpb.2025.109159","DOIUrl":"10.1016/j.cmpb.2025.109159","url":null,"abstract":"<div><h3>Background and Objective</h3><div>Electroencephalography (EEG) is a crucial tool for monitoring recovery in patients with disorders of consciousness (DOC) after therapeutic interventions. It helps in identifying the neural correlates and in guiding the development of personalized treatments. Spectrum power measures are widely employed. However, these measures are manually handcrafted, not patient-specific, and not tailored to the specific intervention.</div></div><div><h3>Methods</h3><div>To address these limitations, we propose an explainable artificial intelligence (XAI) framework designed to automatically uncover the most salient frequency-domain EEG signatures in an intervention- and patient-specific manner. The framework integrates an interpretable convolutional neural network, which is capable of learning interpretable frequency-domain EEG features, with an explanation technique, which quantifies the relevance of the learned spectral features. This approach enables the automatic tracking of patient-specific spectral EEG changes and refines the analysis toward neural features that are more closely associated with key clinical variables.</div></div><div><h3>Results</h3><div>We showcase the potential of our approach by applying it to EEG signals collected from patients in a minimally conscious state following an intervention based on transcranial direct current stimulation. The XAI results reveal a prominent role of alpha-band EEG oscillations in DOC intervention, supporting evidence that functional improvements due to intervention are associated with an increase in alpha-band spectral content.</div></div><div><h3>Conclusions</h3><div>Our XAI-driven analysis offers a robust, individualized, and transparent alternative (or complement) to conventional EEG analyses, thereby enhancing the EEG characterization of DOC patients.</div></div>","PeriodicalId":10624,"journal":{"name":"Computer methods and programs in biomedicine","volume":"275 ","pages":"Article 109159"},"PeriodicalIF":4.8,"publicationDate":"2026-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145682464","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Building and validating a machine learning model to predict coronary heart disease risk based on non-invasive indicators 建立并验证基于非侵入性指标预测冠心病风险的机器学习模型

IF 4.8 2区医学 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computer methods and programs in biomedicine

Pub Date : 2026-02-01 Epub Date: 2025-11-30 DOI: 10.1016/j.cmpb.2025.109186

Bo Wu , Kang Huang , Xin Hong , Shuai-shuai Zhao , Chuan Yuan , Xiao-gui Li , Cheng-tao Peng , Ya-hui Li , Qi-cai Wu , Xue-liang Zhou

Background

Coronary heart disease (CHD) remains a leading global cause of death. Early identification of high-risk individuals and timely intervention are crucial. This study developed and evaluated a predictive CHD risk model using machine learning (ML) techniques.

Methods

The Behavioral Risk Factor Surveillance System (BRFSS) data were randomly split into a training set and an internal validation set in a 7:3 ratio. Variable screening was performed using univariate and multivariate logistic regression analyses. Subsequently, predictive models were developed using eight machine learning algorithms. Model performance on the internal validation set was evaluated using the area under the curve (AUC), sensitivity, specificity, and accuracy, and the optimal model was selected based on these metrics. The National Health and Nutrition Examination Survey (NHANES) dataset was used for external validation of the optimal model. Shapley Additive exPlanations (SHAP) analysis was employed to visualize the importance of features.

Results

Eight machine learning models were developed based on 12 clinical features. Among these models, the Light Gradient Boosting Machine (LightGBM) demonstrated the best performance, with an internal-validation cohort AUC of 0.825 (95% CI 0.821–0.829), sensitivity of 0.800, and specificity of 0.700, significantly outperforming the other models. The external-validation cohort achieved an AUC of 0.851 (95% CI 0.835–0.867). SHAP analysis identified age, sex, hypertension, and dyslipidaemia as key risk factors. A web-based calculator was developed based on the LightGBM model to predict CHD.

Conclusion

LightGBM-based prediction model exhibits high accuracy in assessing CHD risk and holds promise as an effective tool for the early screening and prevention of CHD.

背景：冠心病（CHD）仍然是全球主要的死亡原因。早期识别高危人群并及时干预至关重要。本研究利用机器学习（ML）技术开发并评估了预测冠心病风险模型。方法将行为风险因素监测系统（BRFSS）数据按7:3的比例随机分为训练集和内部验证集。采用单因素和多因素logistic回归分析进行变量筛选。随后，使用八种机器学习算法开发了预测模型。使用曲线下面积（AUC）、灵敏度、特异性和准确性评估模型在内部验证集上的性能，并根据这些指标选择最佳模型。采用国家健康与营养调查（NHANES）数据集对优化模型进行外部验证。Shapley加性解释（SHAP）分析用于可视化特征的重要性。结果基于12个临床特征建立了8个机器学习模型。其中，光梯度增强机（Light Gradient Boosting Machine, LightGBM）表现最好，其内部验证队列AUC为0.825 (95% CI 0.821-0.829)，灵敏度为0.800，特异性为0.700，显著优于其他模型。外部验证队列的AUC为0.851 （95% CI 0.835-0.867）。SHAP分析确定年龄、性别、高血压和血脂异常是关键的危险因素。基于LightGBM模型开发了一个基于网络的计算器来预测冠心病。结论基于lightgbm的冠心病风险预测模型具有较高的准确性，有望成为早期筛查和预防冠心病的有效工具。

{"title":"Building and validating a machine learning model to predict coronary heart disease risk based on non-invasive indicators","authors":"Bo Wu , Kang Huang , Xin Hong , Shuai-shuai Zhao , Chuan Yuan , Xiao-gui Li , Cheng-tao Peng , Ya-hui Li , Qi-cai Wu , Xue-liang Zhou","doi":"10.1016/j.cmpb.2025.109186","DOIUrl":"10.1016/j.cmpb.2025.109186","url":null,"abstract":"<div><h3>Background</h3><div>Coronary heart disease (CHD) remains a leading global cause of death. Early identification of high-risk individuals and timely intervention are crucial. This study developed and evaluated a predictive CHD risk model using machine learning (ML) techniques.</div></div><div><h3>Methods</h3><div>The Behavioral Risk Factor Surveillance System (BRFSS) data were randomly split into a training set and an internal validation set in a 7:3 ratio. Variable screening was performed using univariate and multivariate logistic regression analyses. Subsequently, predictive models were developed using eight machine learning algorithms. Model performance on the internal validation set was evaluated using the area under the curve (AUC), sensitivity, specificity, and accuracy, and the optimal model was selected based on these metrics. The National Health and Nutrition Examination Survey (NHANES) dataset was used for external validation of the optimal model. Shapley Additive exPlanations (SHAP) analysis was employed to visualize the importance of features.</div></div><div><h3>Results</h3><div>Eight machine learning models were developed based on 12 clinical features. Among these models, the Light Gradient Boosting Machine (LightGBM) demonstrated the best performance, with an internal-validation cohort AUC of 0.825 (95% CI 0.821–0.829), sensitivity of 0.800, and specificity of 0.700, significantly outperforming the other models. The external-validation cohort achieved an AUC of 0.851 (95% CI 0.835–0.867). SHAP analysis identified age, sex, hypertension, and dyslipidaemia as key risk factors. A web-based calculator was developed based on the LightGBM model to predict CHD.</div></div><div><h3>Conclusion</h3><div>LightGBM-based prediction model exhibits high accuracy in assessing CHD risk and holds promise as an effective tool for the early screening and prevention of CHD.</div></div>","PeriodicalId":10624,"journal":{"name":"Computer methods and programs in biomedicine","volume":"275 ","pages":"Article 109186"},"PeriodicalIF":4.8,"publicationDate":"2026-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145682575","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

FRET-SAM: SAM_Med2D-based automatic FRET two-hybrid analysis FRET- sam：基于sam_med2d的自动FRET双混合分析。

IF 4.8 2区医学 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computer methods and programs in biomedicine

Pub Date : 2026-02-01 Epub Date: 2025-12-13 DOI: 10.1016/j.cmpb.2025.109208

Jingzhen Wang , Yanling Xu , Beini Sun, Zhiqiang Wei, Rumeng Qu, Fengting Wang, Zhengfei Zhuang, Min Hu, Tongsheng Chen

Background and Objective:

The fluorescence resonance energy transfer (FRET) two-hybrid assay enables quantification of the stoichiometry and binding affinity of protein interactions directly in living cells, but its broader application remains constrained by labor-intensive manual image analysis and high computational complexity. This study leverages deep learning to accurately extract FRET two-hybrid image signals and automate the FRET two-hybrid analysis process, thereby eliminating subjective bias and enhancing the method’s efficiency and accuracy.

Methods:

Based on the Segment Anything Model (SAM), we developed FRET-SAM, an optimized analysis method adapting SAM_Med2D’s structure for automated regions of interest (ROI) selection and fluorescence signal extraction in FRET two-hybrid images. A comprehensive FRET image dataset was established, including six model plasmids (C4Y, C10Y, C40Y, C80Y, C32V and CVC) and three functional FRET pairs (Bcl-XL-CFP/Bak-YFP, EGFR-CFP/Grb2-YFP and RAF-CFP/RAS-YFP), for model training and validation. Model segmentation performance was assessed by comparing its mean pixel accuracy (MPA), mean intersection over union (MIoU), and Dice coefficient against the original SAM_Med2D model. To assess protein interaction results, FRET-SAM-derived values were compared to established literature values, using relative error as a key metric of consistency.

Results:

The FRET-SAM model exhibited enhanced segmentation accuracy, with MPA, MIoU, and Dice coefficient increasing by 2.88%, 2.36%, and 2.19%, respectively, compared to the original SAM_Med2D model. Validation experiments demonstrated high consistency between FRET-SAM-derived results and literature values, with all plasmid models exhibiting relative errors that were individually calculated and confirmed to be under 5%. Furthermore, FRET-SAM exhibited robust drug screening potential in three biomedical case studies: (1) EGFR-Grb2-targeted lung cancer intervention (gefitinib), (2) RAS-RAF-mediated hepatocellular carcinoma suppression (sorafenib), and (3) Bcl-XL inhibitors discovery (A-1331852). Mechanistic studies confirmed its ability to resolve drug-target interactions.

Conclusions:

By enabling automated analysis of FRET images, FRET-SAM significantly enhances the efficiency and accuracy of FRET two-hybrid assays, while eliminating subjective bias. The capability of FRET-SAM to resolve drug-target interactions establishes it as a promising tool for drug discovery.

背景和目的：荧光共振能量转移（FRET）双杂交分析能够直接定量测定活细胞中蛋白质相互作用的化学计量学和结合亲和力，但其广泛应用仍然受到劳动密集型人工图像分析和高计算复杂性的限制。本研究利用深度学习准确提取FRET双混合图像信号，实现FRET双混合分析过程的自动化，从而消除了主观偏差，提高了方法的效率和准确性。方法：基于片段任意模型（SAM），利用SAM_Med2D的结构，开发了一种用于自动选择感兴趣区域（ROI）和提取荧光信号的优化分析方法——FRET-SAM。建立了完整的FRET图像数据集，包括6个模型质粒（C4Y、C10Y、C40Y、C80Y、C32V和CVC）和3个功能FRET对（Bcl-XL-CFP/ bank - yfp、EGFR-CFP/Grb2-YFP和RAF-CFP/RAS-YFP），用于模型训练和验证。通过对比SAM_Med2D模型的平均像素精度（MPA）、平均交集比（MIoU）和Dice系数来评估模型的分割性能。为了评估蛋白质相互作用结果，将fret - sam衍生值与已建立的文献值进行比较，使用相对误差作为一致性的关键指标。结果：与原始SAM_Med2D模型相比，FRET-SAM模型的MPA、MIoU和Dice系数分别提高了2.88%、2.36%和2.19%，分割精度有所提高。验证实验表明，fret - sam衍生的结果与文献值高度一致，所有质粒模型都显示出单独计算并确认在5%以下的相对误差。此外，FRET-SAM在三个生物医学案例研究中显示出强大的药物筛选潜力：(1)egfr - grb2靶向肺癌干预（吉非替尼），(2)ras - raf介导的肝细胞癌抑制（索拉非尼），以及(3)Bcl-XL抑制剂发现（A-1331852）。机制研究证实了其解决药物-靶标相互作用的能力。结论：通过自动分析FRET图像，FRET- sam显著提高了FRET双杂交分析的效率和准确性，同时消除了主观偏差。FRET-SAM解决药物-靶标相互作用的能力使其成为药物发现的有前途的工具。

{"title":"FRET-SAM: SAM_Med2D-based automatic FRET two-hybrid analysis","authors":"Jingzhen Wang , Yanling Xu , Beini Sun, Zhiqiang Wei, Rumeng Qu, Fengting Wang, Zhengfei Zhuang, Min Hu, Tongsheng Chen","doi":"10.1016/j.cmpb.2025.109208","DOIUrl":"10.1016/j.cmpb.2025.109208","url":null,"abstract":"<div><h3>Background and Objective:</h3><div>The fluorescence resonance energy transfer (FRET) two-hybrid assay enables quantification of the stoichiometry and binding affinity of protein interactions directly in living cells, but its broader application remains constrained by labor-intensive manual image analysis and high computational complexity. This study leverages deep learning to accurately extract FRET two-hybrid image signals and automate the FRET two-hybrid analysis process, thereby eliminating subjective bias and enhancing the method’s efficiency and accuracy.</div></div><div><h3>Methods:</h3><div>Based on the Segment Anything Model (SAM), we developed FRET-SAM, an optimized analysis method adapting SAM_Med2D’s structure for automated regions of interest (ROI) selection and fluorescence signal extraction in FRET two-hybrid images. A comprehensive FRET image dataset was established, including six model plasmids (C4Y, C10Y, C40Y, C80Y, C32V and CVC) and three functional FRET pairs (Bcl-XL-CFP/Bak-YFP, EGFR-CFP/Grb2-YFP and RAF-CFP/RAS-YFP), for model training and validation. Model segmentation performance was assessed by comparing its mean pixel accuracy (MPA), mean intersection over union (MIoU), and Dice coefficient against the original SAM_Med2D model. To assess protein interaction results, FRET-SAM-derived values were compared to established literature values, using relative error as a key metric of consistency.</div></div><div><h3>Results:</h3><div>The FRET-SAM model exhibited enhanced segmentation accuracy, with MPA, MIoU, and Dice coefficient increasing by 2.88%, 2.36%, and 2.19%, respectively, compared to the original SAM_Med2D model. Validation experiments demonstrated high consistency between FRET-SAM-derived results and literature values, with all plasmid models exhibiting relative errors that were individually calculated and confirmed to be under 5%. Furthermore, FRET-SAM exhibited robust drug screening potential in three biomedical case studies: (1) EGFR-Grb2-targeted lung cancer intervention (gefitinib), (2) RAS-RAF-mediated hepatocellular carcinoma suppression (sorafenib), and (3) Bcl-XL inhibitors discovery (A-1331852). Mechanistic studies confirmed its ability to resolve drug-target interactions.</div></div><div><h3>Conclusions:</h3><div>By enabling automated analysis of FRET images, FRET-SAM significantly enhances the efficiency and accuracy of FRET two-hybrid assays, while eliminating subjective bias. The capability of FRET-SAM to resolve drug-target interactions establishes it as a promising tool for drug discovery.</div></div>","PeriodicalId":10624,"journal":{"name":"Computer methods and programs in biomedicine","volume":"275 ","pages":"Article 109208"},"PeriodicalIF":4.8,"publicationDate":"2026-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145767252","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

A novel multimodal diagnostic framework integrating hyperspectral imaging and deep learning for predicting RET gene mutations in medullary thyroid carcinoma 结合高光谱成像和深度学习预测甲状腺髓样癌RET基因突变的新型多模态诊断框架。

IF 4.8 2区医学 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computer methods and programs in biomedicine

Pub Date : 2026-02-01 Epub Date: 2025-12-13 DOI: 10.1016/j.cmpb.2025.109207

Zhenpeng Yang , Peng Su , Yuyang Zhang , Haitao Zheng , Yupeng Deng , Xiangfeng Lin , Changyuan Ding , Wei Li , Weili Liang , Bin Lv

Background and objective

Medullary thyroid carcinoma (MTC) is an aggressive malignancy driven predominantly by activating mutations in the RET proto-oncogene. Conventional genotyping using polymerase chain reaction (PCR) or next-generation sequencing (NGS) is often hampered by burdensome costs and prolonged turnaround times, hindering timely clinical decision-making.

Methods

We developed a rapid, cost-effective, multimodal deep-learning framework to predict RET mutations from standard H&E-stained slides. Our approach leverages hyperspectral imaging and integrates a 1D-CNN-LSTM network for spectral analysis with a Swin Transformer for spatial feature extraction. A cross-modal attention mechanism effectively fuses these representations. The model was trained and validated on 82 MTC cases from Qilu Hospital and externally tested on independent cohorts from two additional centers (n = 60).

Results

The proposed framework achieved an overall accuracy of 89.5 %, with a sensitivity of 90.2 % and specificity of 88.6 % for RET mutation classification. External validation confirmed robust generalizability, with performance surpassing single-modality benchmarks by 7.0–19.5 %.

Conclusions

This study presents a non-invasive and efficient alternative for predicting RET mutations in MTC, demonstrating the potential of hyperspectral imaging and integrated deep learning to advance precision oncology.

背景和目的：甲状腺髓样癌（MTC）是一种主要由RET原癌基因激活突变驱动的侵袭性恶性肿瘤。使用聚合酶链反应（PCR）或下一代测序（NGS）的传统基因分型常常受到沉重的成本和较长的周转时间的阻碍，阻碍了及时的临床决策。方法：我们开发了一种快速、经济、多模式的深度学习框架，从标准h&e染色的载玻片中预测RET突变。我们的方法利用高光谱成像，并将用于光谱分析的1D-CNN-LSTM网络与用于空间特征提取的Swin变压器集成在一起。跨模态注意机制有效地融合了这些表征。该模型在齐鲁医院的82例MTC病例中进行了训练和验证，并在另外两个中心的独立队列中进行了外部测试（n = 60）。结果：该框架对RET突变分类的总体准确率为89.5%，敏感性为90.2%，特异性为88.6%。外部验证证实了鲁棒的泛化性，性能优于单模态基准7.0- 19.5%。结论：本研究提出了一种非侵入性和有效的预测MTC RET突变的替代方法，展示了高光谱成像和集成深度学习在推进精准肿瘤学方面的潜力。

{"title":"A novel multimodal diagnostic framework integrating hyperspectral imaging and deep learning for predicting RET gene mutations in medullary thyroid carcinoma","authors":"Zhenpeng Yang , Peng Su , Yuyang Zhang , Haitao Zheng , Yupeng Deng , Xiangfeng Lin , Changyuan Ding , Wei Li , Weili Liang , Bin Lv","doi":"10.1016/j.cmpb.2025.109207","DOIUrl":"10.1016/j.cmpb.2025.109207","url":null,"abstract":"<div><h3>Background and objective</h3><div>Medullary thyroid carcinoma (MTC) is an aggressive malignancy driven predominantly by activating mutations in the RET proto-oncogene. Conventional genotyping using polymerase chain reaction (PCR) or next-generation sequencing (NGS) is often hampered by burdensome costs and prolonged turnaround times, hindering timely clinical decision-making.</div></div><div><h3>Methods</h3><div>We developed a rapid, cost-effective, multimodal deep-learning framework to predict RET mutations from standard H&E-stained slides. Our approach leverages hyperspectral imaging and integrates a 1D-CNN-LSTM network for spectral analysis with a Swin Transformer for spatial feature extraction. A cross-modal attention mechanism effectively fuses these representations. The model was trained and validated on 82 MTC cases from Qilu Hospital and externally tested on independent cohorts from two additional centers (<em>n</em> = 60).</div></div><div><h3>Results</h3><div>The proposed framework achieved an overall accuracy of 89.5 %, with a sensitivity of 90.2 % and specificity of 88.6 % for RET mutation classification. External validation confirmed robust generalizability, with performance surpassing single-modality benchmarks by 7.0–19.5 %.</div></div><div><h3>Conclusions</h3><div>This study presents a non-invasive and efficient alternative for predicting RET mutations in MTC, demonstrating the potential of hyperspectral imaging and integrated deep learning to advance precision oncology.</div></div>","PeriodicalId":10624,"journal":{"name":"Computer methods and programs in biomedicine","volume":"275 ","pages":"Article 109207"},"PeriodicalIF":4.8,"publicationDate":"2026-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145780510","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Explainable feature selection combining particle swarm optimisation with adaptive LASSO for MRI radiogenomics: Predicting HPV status in oropharyngeal cancer 结合粒子群优化和自适应LASSO的MRI放射基因组学的可解释特征选择：预测口咽癌中的HPV状态。

IF 4.8 2区医学 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computer methods and programs in biomedicine

Pub Date : 2026-02-01 Epub Date: 2025-12-11 DOI: 10.1016/j.cmpb.2025.109204

Milad Ahmadian , Zuhir Bodalal , Mary Adib , Seyed Sahand Mohammadi Ziabari , Paula Bos , Roland M. Martens , Georgios Agrotis , Conchita Vens , Luc Karssemakers , Abrahim Al-Mamgani , Pim de Graaf , Bas Jasperse , Ruud H Brakenhoff , C René Leemans , Regina G.H. Beets-Tan , Michiel W.M. van den Brekel , Jonas A. Castelijns

Background

Radiogenomic modelling faces a significant challenge due to the high-dimensional nature of quantitative radiomic features and limited sample sizes. Feature selection is therefore essential to eliminate irrelevant features and mitigate overfitting. Particle swarm optimisation (PSO) has shown promise for effectively navigating large feature spaces, yet its effectiveness in radiogenomics remains unexplored. This study investigates the value of PSO-based methods, both independently and in combination with other advanced techniques, for MRI-based prediction of human papillomavirus (HPV) status in oropharyngeal squamous cell carcinoma (OPSCC).

Materials and methods

Baseline contrast-enhanced T1-weighted MR scans from two centres were analysed: 153 patients in an internal cohort (randomly split into 80 % for training and 20 % for testing) and 157 patients in an external validation cohort. Radiomic features were extracted from manually segmented tumours and multiple feature selection methods, including PSO and its ensembles, filter-based methods, wrapper-based approaches, and shrinkage techniques were evaluated. Performance was measured and compared using the area under the receiver operating characteristic curve (AUC).

Results

PSO alone had a reasonable predictive power on the internal test set (AUC = 0.76, 95 % CI: 0.57–0.92, p = 0.092). When combined with adaptive LASSO using Shapley values, PSO’s performance improved (AUC = 0.81, 95 % CI: 0.61–0.94, p = 0.023). Recursive feature elimination (RFE) selected the most relevant features (AUC = 0.91, 95 % CI: 0.79–1.00, p < 0.001). Despite this, RFE failed to generalise well to the external cohort (AUC = 0.52, 95 % CI: 0.42–0.60, p = 1). Meanwhile, the PSO–adaptive LASSO combination maintained a robust AUC = 0.78 (95 % CI: 0.70–0.85, p < 0.001), indicating superior generalisability.

Conclusions

The explainable PSO–adaptive LASSO feature selection method provides generalisable radiogenomic signatures associated with HPV status in OPSCC, outperforming other feature selection approaches. This combination may serve as a robust strategy for developing transferable models in radiogenomics.

背景：由于定量放射组学特征的高维性质和有限的样本量，放射基因组学建模面临着重大挑战。因此，特征选择对于消除不相关特征和减轻过度拟合至关重要。粒子群优化（PSO）已经显示出有效导航大特征空间的希望，但其在放射基因组学中的有效性仍有待探索。本研究探讨了基于pso的方法，无论是独立的还是与其他先进技术相结合，在基于mri预测口咽鳞状细胞癌（OPSCC）中人乳头瘤病毒（HPV）状态的价值。材料和方法：分析了来自两个中心的基线对比增强t1加权MR扫描：153名患者在内部队列（随机分为80%用于训练和20%用于测试）和157名患者在外部验证队列。从人工分割的肿瘤中提取放射学特征，并评估了多种特征选择方法，包括PSO及其集合、基于过滤器的方法、基于包装的方法和收缩技术。使用接收器工作特性曲线（AUC）下的面积来测量和比较性能。结果：PSO单独在内部测试集上具有合理的预测能力（AUC = 0.76, 95% CI: 0.57-0.92, p = 0.092）。当使用Shapley值与自适应LASSO结合使用时，PSO的性能得到改善（AUC = 0.81, 95% CI: 0.61-0.94, p = 0.023）。递归特征消除（RFE）选择了最相关的特征（AUC = 0.91, 95% CI: 0.79-1.00, p < 0.001）。尽管如此，RFE未能很好地推广到外部队列（AUC = 0.52, 95% CI: 0.42-0.60, p = 1）。同时，pso -自适应LASSO组合保持稳健的AUC = 0.78 (95% CI: 0.70-0.85, p < 0.001)，表明具有较好的通用性。结论：可解释的pso自适应LASSO特征选择方法提供了与OPSCC中HPV状态相关的通用放射基因组特征，优于其他特征选择方法。这种组合可以作为在放射基因组学中开发可转移模型的有力策略。

{"title":"Explainable feature selection combining particle swarm optimisation with adaptive LASSO for MRI radiogenomics: Predicting HPV status in oropharyngeal cancer","authors":"Milad Ahmadian , Zuhir Bodalal , Mary Adib , Seyed Sahand Mohammadi Ziabari , Paula Bos , Roland M. Martens , Georgios Agrotis , Conchita Vens , Luc Karssemakers , Abrahim Al-Mamgani , Pim de Graaf , Bas Jasperse , Ruud H Brakenhoff , C René Leemans , Regina G.H. Beets-Tan , Michiel W.M. van den Brekel , Jonas A. Castelijns","doi":"10.1016/j.cmpb.2025.109204","DOIUrl":"10.1016/j.cmpb.2025.109204","url":null,"abstract":"<div><h3>Background</h3><div>Radiogenomic modelling faces a significant challenge due to the high-dimensional nature of quantitative radiomic features and limited sample sizes. Feature selection is therefore essential to eliminate irrelevant features and mitigate overfitting. Particle swarm optimisation (PSO) has shown promise for effectively navigating large feature spaces, yet its effectiveness in radiogenomics remains unexplored. This study investigates the value of PSO-based methods, both independently and in combination with other advanced techniques, for MRI-based prediction of human papillomavirus (HPV) status in oropharyngeal squamous cell carcinoma (OPSCC).</div></div><div><h3>Materials and methods</h3><div>Baseline contrast-enhanced T1-weighted MR scans from two centres were analysed: 153 patients in an internal cohort (randomly split into 80 % for training and 20 % for testing) and 157 patients in an external validation cohort. Radiomic features were extracted from manually segmented tumours and multiple feature selection methods, including PSO and its ensembles, filter-based methods, wrapper-based approaches, and shrinkage techniques were evaluated. Performance was measured and compared using the area under the receiver operating characteristic curve (AUC).</div></div><div><h3>Results</h3><div>PSO alone had a reasonable predictive power on the internal test set (AUC = 0.76, 95 % CI: 0.57–0.92, <em>p</em> = 0.092). When combined with adaptive LASSO using Shapley values, PSO’s performance improved (AUC = 0.81, 95 % CI: 0.61–0.94, <em>p</em> = 0.023). Recursive feature elimination (RFE) selected the most relevant features (AUC = 0.91, 95 % CI: 0.79–1.00, <em>p</em> < 0.001). Despite this, RFE failed to generalise well to the external cohort (AUC = 0.52, 95 % CI: 0.42–0.60, <em>p</em> = 1). Meanwhile, the PSO–adaptive LASSO combination maintained a robust AUC = 0.78 (95 % CI: 0.70–0.85, <em>p</em> < 0.001), indicating superior generalisability.</div></div><div><h3>Conclusions</h3><div>The explainable PSO–adaptive LASSO feature selection method provides generalisable radiogenomic signatures associated with HPV status in OPSCC, outperforming other feature selection approaches. This combination may serve as a robust strategy for developing transferable models in radiogenomics.</div></div>","PeriodicalId":10624,"journal":{"name":"Computer methods and programs in biomedicine","volume":"275 ","pages":"Article 109204"},"PeriodicalIF":4.8,"publicationDate":"2026-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145780469","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Multiscale simulation and parallel space–time adaptivity of calcium sparks in cardiac myocytes 心肌细胞钙火花的多尺度模拟及平行时空适应性研究。

IF 4.8 2区医学 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computer methods and programs in biomedicine

Pub Date : 2026-02-01 Epub Date: 2025-11-07 DOI: 10.1016/j.cmpb.2025.109154

Wilhelm Neubert , Martin Falcke , Nagaiah Chamakuri

Background and Objective:

Calcium serves as the bidirectional link between the heart’s electrical excitation and contraction. Electrical excitation induces an influx of Calcium across the sarcolemma and T-tubular membrane, triggering calcium release from the sarcoplasmic reticulum. Calcium sparks, the fundamental events of calcium release from the SR, are initiated in specialized microdomains where Ryanodine Receptors and L-type calcium channels co-locate. The spatial heterogeneity of Calcium release and the random occurrence of strong release fluxes render simulations challenging. Developing mathematical models and efficient simulations of detailed calcium spark models is crucial to understanding heart function. In this paper, we introduce space–time adaptivity within a parallel computing framework into the multiscale simulation of calcium sparks in cardiac myocytes to improve the stability and performance of these simulations.

Methods:

We model intracellular calcium concentrations in both the cytoplasm and the SR domains using a set of coupled reaction–diffusion equations. Spatial grid adaptivity is implemented through multilevel finite element methods to account for the spatial heterogeneity of intracellular

{Ca}^{2 +}

release. Rosenbrock-type techniques handle small time steps for simulating stochastic channel opening and closing in the

{Ca}^{2 +}

release units (CRUs).

Results:

Our test cases demonstrate the superior efficiency of the space–time adaptive approach in optimizing computational resources. The parallel space–time adaptive method accelerates simulations of calcium sparks by a factor of 16.07.

Conclusions:

The efficiency and speed gains in Calcium spark simulations are significant and enable modeling based research into previously difficult to tackle questions with regard to sub-micrometer scale models, e.g with respect to local interactions between the Sodium Calcium Exchanger and RyR clusters.

背景与目的：钙在心脏的电兴奋和电收缩之间起着双向联系的作用。电兴奋诱导钙通过肌膜和t管膜流入，触发钙从肌浆网释放。钙火花是钙从SR释放的基本事件，在Ryanodine受体和l型钙通道共同定位的特定微域中启动。钙释放的空间异质性和强释放通量的随机发生使得模拟具有挑战性。建立数学模型和有效模拟详细的钙火花模型对了解心脏功能至关重要。本文在并行计算框架下，将时空自适应引入心肌细胞钙火花的多尺度模拟中，以提高模拟的稳定性和性能。方法：我们使用一组耦合反应-扩散方程来模拟细胞质和SR结构域的细胞内钙浓度。空间网格适应性是通过多层次的有限元方法来实现的，以解释细胞内Ca2+释放的空间异质性。rosenbrock型技术处理小时间步模拟Ca2+释放单元（cru）中的随机通道打开和关闭。结果：我们的测试用例证明了时空自适应方法在优化计算资源方面的优越效率。平行时空自适应方法将钙火花的模拟速度提高了16.07倍。结论：钙火花模拟的效率和速度的提高是显著的，并且使基于建模的研究能够解决以前难以解决的关于亚微米尺度模型的问题，例如关于钠钙交换器和RyR簇之间的局部相互作用。

{"title":"Multiscale simulation and parallel space–time adaptivity of calcium sparks in cardiac myocytes","authors":"Wilhelm Neubert , Martin Falcke , Nagaiah Chamakuri","doi":"10.1016/j.cmpb.2025.109154","DOIUrl":"10.1016/j.cmpb.2025.109154","url":null,"abstract":"<div><h3>Background and Objective:</h3><div>Calcium serves as the bidirectional link between the heart’s electrical excitation and contraction. Electrical excitation induces an influx of Calcium across the sarcolemma and T-tubular membrane, triggering calcium release from the sarcoplasmic reticulum. Calcium sparks, the fundamental events of calcium release from the SR, are initiated in specialized microdomains where Ryanodine Receptors and L-type calcium channels co-locate. The spatial heterogeneity of Calcium release and the random occurrence of strong release fluxes render simulations challenging. Developing mathematical models and efficient simulations of detailed calcium spark models is crucial to understanding heart function. In this paper, we introduce space–time adaptivity within a parallel computing framework into the multiscale simulation of calcium sparks in cardiac myocytes to improve the stability and performance of these simulations.</div></div><div><h3>Methods:</h3><div>We model intracellular calcium concentrations in both the cytoplasm and the SR domains using a set of coupled reaction–diffusion equations. Spatial grid adaptivity is implemented through multilevel finite element methods to account for the spatial heterogeneity of intracellular <span><math><msup><mrow><mtext>Ca</mtext></mrow><mrow><mn>2</mn><mo>+</mo></mrow></msup></math></span> release. Rosenbrock-type techniques handle small time steps for simulating stochastic channel opening and closing in the <span><math><msup><mrow><mtext>Ca</mtext></mrow><mrow><mn>2</mn><mo>+</mo></mrow></msup></math></span> release units (CRUs).</div></div><div><h3>Results:</h3><div>Our test cases demonstrate the superior efficiency of the space–time adaptive approach in optimizing computational resources. The parallel space–time adaptive method accelerates simulations of calcium sparks by a factor of 16.07.</div></div><div><h3>Conclusions:</h3><div>The efficiency and speed gains in Calcium spark simulations are significant and enable modeling based research into previously difficult to tackle questions with regard to sub-micrometer scale models, e.g with respect to local interactions between the Sodium Calcium Exchanger and RyR clusters.</div></div>","PeriodicalId":10624,"journal":{"name":"Computer methods and programs in biomedicine","volume":"274 ","pages":"Article 109154"},"PeriodicalIF":4.8,"publicationDate":"2026-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145494854","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Drug repurposing through pathway perturbation dynamics: A systems biology approach for precision oncology 通过途径微扰动力学的药物再利用：精确肿瘤学的系统生物学方法。

IF 4.8 2区医学 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computer methods and programs in biomedicine

Pub Date : 2026-02-01 Epub Date: 2025-11-20 DOI: 10.1016/j.cmpb.2025.109177

Xianbin Li , Wuxiang Ruan , Guoan Lu , Dingcheng Ban , Luming Tian , Binbin Wang , Tao Liu , Guodao Zhang , Chunping Wang , Jie Lin

Background and objective

Drug repurposing offers a cost-efficient strategy to discover new therapeutic applications for approved drugs. While current computational strategies prioritize candidates by targeting disease-related pathways, they often fail to quantitatively model pathway perturbation dynamics—a critical gap that limits mechanistic interpretability.

Methods

To address this issue, we presented PathPertDrug, a novel framework that systematically identifies cancer drug candidates by quantifying functional antagonism between drug-induced and disease-associated pathway perturbations (activation/ inhibition). By integrating drug-induced gene expression, disease-related gene expression, and pathway information, PathPertDrug evaluated pathway-level functional reversals, enabling precise prediction of drug-disease associations.

Results

Our method demonstrated superior predictive accuracy and robustness across pan-cancer benchmarks. It achieved a higher median AUROC (0.62 vs. 0.42–0.53) and a substantial improvement in AUPR (3–23 %) over existing methods. The consistent AUPR enhancement, particularly under class imbalance, underscores the robustness of our model in reliably prioritizing true positive associations. Validated by the comparative toxicogenmics database, PathPertDrug rediscovered 83 % of literature-supported cancer drugs (e.g., fulvestrant (Fulvestrant) for colorectal cancer) and predicted novel candidates (e.g., rifabutin–lung cancer).

Conclusions

This pathway-centric approach bridged mechanistic insights with translational applications, providing a paradigm shift for precision oncology drug discovery.

背景和目的：药物再利用为发现已批准药物的新治疗应用提供了一种具有成本效益的策略。虽然目前的计算策略通过针对疾病相关的途径来优先考虑候选者，但它们往往无法定量地模拟途径扰动动力学——这是一个限制机制可解释性的关键缺陷。方法：为了解决这个问题，我们提出了PathPertDrug，这是一个新的框架，通过量化药物诱导和疾病相关途径扰动（激活/抑制）之间的功能拮抗，系统地识别癌症候选药物。通过整合药物诱导的基因表达、疾病相关的基因表达和通路信息，PathPertDrug评估了通路水平的功能逆转，从而能够精确预测药物与疾病的关联。结果：我们的方法在泛癌症基准中表现出卓越的预测准确性和稳健性。与现有方法相比，该方法获得了更高的中位AUROC (0.62 vs. 0.42-0.53)，并且AUPR（3- 23%）得到了显著改善。一致的AUPR增强，特别是在阶级不平衡的情况下，强调了我们的模型在可靠地优先考虑真正关联方面的稳健性。通过比较毒理学数据库的验证，PathPertDrug重新发现了83%的文献支持的癌症药物（例如，用于结直肠癌的氟维司汀（fulvestrant）），并预测了新的候选药物（例如，利法布汀-肺癌）。结论：这种以途径为中心的方法将机制见解与转化应用联系起来，为精确的肿瘤药物发现提供了范式转变。

{"title":"Drug repurposing through pathway perturbation dynamics: A systems biology approach for precision oncology","authors":"Xianbin Li , Wuxiang Ruan , Guoan Lu , Dingcheng Ban , Luming Tian , Binbin Wang , Tao Liu , Guodao Zhang , Chunping Wang , Jie Lin","doi":"10.1016/j.cmpb.2025.109177","DOIUrl":"10.1016/j.cmpb.2025.109177","url":null,"abstract":"<div><h3>Background and objective</h3><div>Drug repurposing offers a cost-efficient strategy to discover new therapeutic applications for approved drugs. While current computational strategies prioritize candidates by targeting disease-related pathways, they often fail to quantitatively model pathway perturbation dynamics—a critical gap that limits mechanistic interpretability.</div></div><div><h3>Methods</h3><div>To address this issue, we presented PathPertDrug, a novel framework that systematically identifies cancer drug candidates by quantifying functional antagonism between drug-induced and disease-associated pathway perturbations (activation/ inhibition). By integrating drug-induced gene expression, disease-related gene expression, and pathway information, PathPertDrug evaluated pathway-level functional reversals, enabling precise prediction of drug-disease associations.</div></div><div><h3>Results</h3><div>Our method demonstrated superior predictive accuracy and robustness across pan-cancer benchmarks. It achieved a higher median AUROC (0.62 vs. 0.42–0.53) and a substantial improvement in AUPR (3–23 %) over existing methods. The consistent AUPR enhancement, particularly under class imbalance, underscores the robustness of our model in reliably prioritizing true positive associations. Validated by the comparative toxicogenmics database, PathPertDrug rediscovered 83 % of literature-supported cancer drugs (e.g., fulvestrant (Fulvestrant) for colorectal cancer) and predicted novel candidates (e.g., rifabutin–lung cancer).</div></div><div><h3>Conclusions</h3><div>This pathway-centric approach bridged mechanistic insights with translational applications, providing a paradigm shift for precision oncology drug discovery.</div></div>","PeriodicalId":10624,"journal":{"name":"Computer methods and programs in biomedicine","volume":"274 ","pages":"Article 109177"},"PeriodicalIF":4.8,"publicationDate":"2026-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145602799","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0