Intelligence-based medicine最新文献_第4页

Improving CNN interpretability and evaluation via alternating training and regularization in chest CT scans 通过交替训练和正则化在胸部CT扫描中提高CNN的可解释性和评价

Intelligence-based medicine

Pub Date : 2025-01-01 DOI: 10.1016/j.ibmed.2025.100211

Rodrigo Ramos-Díaz , Jesús García-Ramírez , Jimena Olveres , Boris Escalante-Ramírez

Interpretable machine learning is an emerging trend that holds significant importance, considering the growing impact of machine learning systems on society and human lives. Many interpretability methods are applied in CNN after training to provide deeper insights into the outcomes, but only a few have tried to promote interpretability during training. The aim of this experimental study is to investigate the interpretability of CNN. This research was applied to chest computed tomography scans, as understanding CNN predictions has particular importance in the automatic classification of medical images. We attempted to implement a CNN technique aimed at improving interpretability by relating filters in the last convolutional to specific output classes. Variations of such a technique were explored and assessed using chest CT images for classification based on the presence of lungs and lesions. A search was conducted to optimize the specific hyper-parameters necessary for the evaluated strategies. A novel strategy is proposed employing transfer learning and regularization. Models obtained with this strategy and the optimized hyperparameters were statistically compared to standard models, demonstrating greater interpretability without a significant loss in predictive accuracy. We achieved CNN models with improved interpretability, which is crucial for the development of more explainable and reliable AI systems.

考虑到机器学习系统对社会和人类生活的影响越来越大，可解释的机器学习是一种具有重要意义的新兴趋势。许多可解释性方法在训练后应用于CNN，以提供对结果的更深入的了解，但只有少数方法试图在训练过程中提高可解释性。本实验研究的目的是探讨CNN的可解释性。这项研究应用于胸部计算机断层扫描，因为理解CNN预测在医学图像的自动分类中特别重要。我们试图实现一种CNN技术，旨在通过将最后一个卷积中的过滤器与特定的输出类关联来提高可解释性。这种技术的变化被探索和评估使用胸部CT图像进行分类基于肺和病变的存在。进行搜索以优化评估策略所需的特定超参数。提出了一种利用迁移学习和正则化的新策略。用该策略获得的模型和优化的超参数与标准模型进行统计比较，显示出更大的可解释性，而不会显著降低预测精度。我们实现了具有改进可解释性的CNN模型，这对于开发更具可解释性和可靠性的人工智能系统至关重要。

{"title":"Improving CNN interpretability and evaluation via alternating training and regularization in chest CT scans","authors":"Rodrigo Ramos-Díaz , Jesús García-Ramírez , Jimena Olveres , Boris Escalante-Ramírez","doi":"10.1016/j.ibmed.2025.100211","DOIUrl":"10.1016/j.ibmed.2025.100211","url":null,"abstract":"<div><div>Interpretable machine learning is an emerging trend that holds significant importance, considering the growing impact of machine learning systems on society and human lives. Many interpretability methods are applied in CNN after training to provide deeper insights into the outcomes, but only a few have tried to promote interpretability during training. The aim of this experimental study is to investigate the interpretability of CNN. This research was applied to chest computed tomography scans, as understanding CNN predictions has particular importance in the automatic classification of medical images. We attempted to implement a CNN technique aimed at improving interpretability by relating filters in the last convolutional to specific output classes. Variations of such a technique were explored and assessed using chest CT images for classification based on the presence of lungs and lesions. A search was conducted to optimize the specific hyper-parameters necessary for the evaluated strategies. A novel strategy is proposed employing transfer learning and regularization. Models obtained with this strategy and the optimized hyperparameters were statistically compared to standard models, demonstrating greater interpretability without a significant loss in predictive accuracy. We achieved CNN models with improved interpretability, which is crucial for the development of more explainable and reliable AI systems.</div></div>","PeriodicalId":73399,"journal":{"name":"Intelligence-based medicine","volume":"11 ","pages":"Article 100211"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143388250","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

A multimodal machine learning model for bipolar disorder mania classification: Insights from acoustic, linguistic, and visual cues 双相情感障碍躁狂分类的多模态机器学习模型：来自声学、语言和视觉线索的见解

Intelligence-based medicine

Pub Date : 2025-01-01 DOI: 10.1016/j.ibmed.2025.100223

Kiruthiga Devi Murugavel , Parthasarathy R , Sandeep Kumar Mathivanan , Saravanan Srinivasan , Basu Dev Shivahare , Mohd Asif Shah

Mood fluctuations that can vary from manic to depressive states are a symptom of a disease known as bipolar disorder, which affects mental health. Interviews with patients and gathering information from their families are essential steps in the diagnostic process for bipolar disorder. Automated approaches for treating bipolar disorder are also being explored. In mental health prevention and care, machine learning techniques (ML) are increasingly used to detect and treat diseases. With frequently analyzed human behaviour patterns, identified symptoms, and risk factors as various parameters of the dataset, predictions can be made for improving traditional diagnosis methods. In this study, A Multimodal Fusion System was developed based on an auditory, linguistic, and visual patient recording as an input dataset for a three-stage mania classification decision system. Deep Denoising Autoencoders (DDAEs) are introduced to learn common representations across five modalities: acoustic characteristics, eye gaze, facial landmarks, head posture, and Facial Action Units (FAUs). This is done in particular for the audio-visual modality. The distributed representations and the transient information during each recording session are eventually encoded into Fisher Vectors (FVs), which capture the representations. Once the Fisher Vectors (FVs) and document embeddings are integrated, a Multi-Task Neural Network is used to perform the classification task, while mitigating overfitting issues caused by the limited size of the bipolar disorder dataset. The study introduces Deep Denoising Autoencoders (DDAEs) for cross-modal representation learning and utilizes Fisher Vectors with Multi-Task Neural Networks, enhancing diagnostic accuracy while highlighting the benefits of multimodal fusion for mental health diagnostics. Achieving an unweighted average recall score of 64.8 %, with the highest AUC-ROC of 0.85 & less interface time of 6.5 ms/sample scores the effectiveness of integrating multiple modalities in improving system performance and advancing feature representation and model interpretability.

从躁狂到抑郁状态的情绪波动是一种被称为双相情感障碍的疾病的症状，这种疾病会影响心理健康。在双相情感障碍的诊断过程中，与患者面谈和从其家庭收集信息是必不可少的步骤。治疗双相情感障碍的自动化方法也在探索中。在心理健康预防和护理中，机器学习技术（ML）越来越多地用于检测和治疗疾病。通过频繁分析人类行为模式、识别症状和风险因素作为数据集的各种参数，可以对改进传统诊断方法进行预测。在这项研究中，一个多模态融合系统是基于听觉、语言和视觉患者记录作为三阶段躁狂分类决策系统的输入数据集而开发的。引入深度去噪自动编码器（DDAEs）来学习五种模式的常见表示：声学特征、眼睛注视、面部标志、头部姿势和面部动作单位（FAUs）。这尤其适用于视听方式。每个记录过程中的分布式表示和瞬态信息最终被编码成捕获表示的Fisher向量（FVs）。一旦将Fisher向量（FVs）和文档嵌入集成在一起，就可以使用多任务神经网络来执行分类任务，同时减轻由双相情感障碍数据集有限大小引起的过拟合问题。该研究引入了用于跨模态表示学习的深度去噪自动编码器（DDAEs），并利用Fisher向量与多任务神经网络，提高了诊断准确性，同时突出了多模态融合对心理健康诊断的好处。实现了64.8%的未加权平均召回分数，最高AUC-ROC为0.85 &；6.5 ms/样本的接口时间较短，对集成多种模式在提高系统性能、提高特征表示和模型可解释性方面的有效性进行了评分。

{"title":"A multimodal machine learning model for bipolar disorder mania classification: Insights from acoustic, linguistic, and visual cues","authors":"Kiruthiga Devi Murugavel , Parthasarathy R , Sandeep Kumar Mathivanan , Saravanan Srinivasan , Basu Dev Shivahare , Mohd Asif Shah","doi":"10.1016/j.ibmed.2025.100223","DOIUrl":"10.1016/j.ibmed.2025.100223","url":null,"abstract":"<div><div>Mood fluctuations that can vary from manic to depressive states are a symptom of a disease known as bipolar disorder, which affects mental health. Interviews with patients and gathering information from their families are essential steps in the diagnostic process for bipolar disorder. Automated approaches for treating bipolar disorder are also being explored. In mental health prevention and care, machine learning techniques (ML) are increasingly used to detect and treat diseases. With frequently analyzed human behaviour patterns, identified symptoms, and risk factors as various parameters of the dataset, predictions can be made for improving traditional diagnosis methods. In this study, A Multimodal Fusion System was developed based on an auditory, linguistic, and visual patient recording as an input dataset for a three-stage mania classification decision system. Deep Denoising Autoencoders (DDAEs) are introduced to learn common representations across five modalities: acoustic characteristics, eye gaze, facial landmarks, head posture, and Facial Action Units (FAUs). This is done in particular for the audio-visual modality. The distributed representations and the transient information during each recording session are eventually encoded into Fisher Vectors (FVs), which capture the representations. Once the Fisher Vectors (FVs) and document embeddings are integrated, a Multi-Task Neural Network is used to perform the classification task, while mitigating overfitting issues caused by the limited size of the bipolar disorder dataset. The study introduces Deep Denoising Autoencoders (DDAEs) for cross-modal representation learning and utilizes Fisher Vectors with Multi-Task Neural Networks, enhancing diagnostic accuracy while highlighting the benefits of multimodal fusion for mental health diagnostics. Achieving an unweighted average recall score of 64.8 %, with the highest AUC-ROC of 0.85 & less interface time of 6.5 ms/sample scores the effectiveness of integrating multiple modalities in improving system performance and advancing feature representation and model interpretability.</div></div>","PeriodicalId":73399,"journal":{"name":"Intelligence-based medicine","volume":"11 ","pages":"Article 100223"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143792053","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Early prediction of sepsis using an XGBoost model with single time-point non-invasive vital signs and its correlation with C-reactive protein and procalcitonin: A multi-center study 基于单时间点无创生命体征的XGBoost模型早期预测脓毒症及其与c反应蛋白和降钙素原的相关性：一项多中心研究

Intelligence-based medicine

Pub Date : 2025-01-01 DOI: 10.1016/j.ibmed.2025.100242

Albert C. Yang , Wei-Ming Ma , Dung-Hung Chiang , Yi-Ze Liao , Hsien-Yung Lai , Shu-Chuan Lin , Mei-Chin Liu , Kai-Ting Wen , Tzong-Huei Lin , Wen-Xiang Tsai , Jun-Ding Zhu , Ting-Yu Chen , Hung-Fu Lee , Pei-Hung Liao , Huey-Wen Yien , Chien-Ying Wang

We aimed to develop an early warning system to predict sepsis based solely on single time-point and non-invasive vital signs, and to evaluate its correlation with related biomarkers, namely C-reactive protein (CRP) and Procalcitonin (PCT). We utilized retrospective data from Physionet and four medical centers in Taiwan, encompassing a total of 46,184 Intensive Care Unit (ICU) patients, to develop and validate a machine learning algorithm based on XGBoost for predicting sepsis. The model was specifically designed to use non-invasive vital signs captured at a single time point, The correlation between sepsis AI prediction model and levels of CRP and PCT was evaluated. The developed model demonstrated balanced performance across various datasets, with an average recall of 0.908 and precision of 0.577. The model's performance was further validated by the independent dataset from Cheng-Hsin General Hospital (recall: 0.986, precision: 0.585). Temperature, systolic blood pressure, and respiration rate were the top contributing predictors in the model. A significant correlation was observed between the model's sepsis predictions and elevated CRP levels, while PCT showed a less consistent pattern. Our approach, combining AI algorithms with vital sign data and its clinical relevance to CRP level, offers a more precise and timely sepsis detection, with the potential to improve care in emergency and critical care settings.

我们的目标是开发一种仅基于单一时间点和无创生命体征的脓毒症预警系统，并评估其与相关生物标志物，即c反应蛋白（CRP）和降钙素原（PCT）的相关性。我们利用来自Physionet和台湾四家医疗中心的回顾性数据，包括46,184名重症监护病房（ICU）患者，开发并验证了基于XGBoost的机器学习算法，用于预测败血症。该模型专门设计用于使用在单个时间点捕获的无创生命体征，评估脓毒症AI预测模型与CRP和PCT水平的相关性。开发的模型在各种数据集上表现出平衡的性能，平均召回率为0.908，精度为0.577。通过独立数据集验证模型的有效性（召回率：0.986，精度：0.585）。温度、收缩压和呼吸速率是模型中最重要的预测因子。在模型的脓毒症预测与CRP水平升高之间观察到显著的相关性，而PCT表现出不太一致的模式。我们的方法将人工智能算法与生命体征数据及其与CRP水平的临床相关性相结合，提供了更精确和及时的败血症检测，有可能改善急诊和重症监护环境的护理。

{"title":"Early prediction of sepsis using an XGBoost model with single time-point non-invasive vital signs and its correlation with C-reactive protein and procalcitonin: A multi-center study","authors":"Albert C. Yang , Wei-Ming Ma , Dung-Hung Chiang , Yi-Ze Liao , Hsien-Yung Lai , Shu-Chuan Lin , Mei-Chin Liu , Kai-Ting Wen , Tzong-Huei Lin , Wen-Xiang Tsai , Jun-Ding Zhu , Ting-Yu Chen , Hung-Fu Lee , Pei-Hung Liao , Huey-Wen Yien , Chien-Ying Wang","doi":"10.1016/j.ibmed.2025.100242","DOIUrl":"10.1016/j.ibmed.2025.100242","url":null,"abstract":"<div><div>We aimed to develop an early warning system to predict sepsis based solely on single time-point and non-invasive vital signs, and to evaluate its correlation with related biomarkers, namely C-reactive protein (CRP) and Procalcitonin (PCT). We utilized retrospective data from Physionet and four medical centers in Taiwan, encompassing a total of 46,184 Intensive Care Unit (ICU) patients, to develop and validate a machine learning algorithm based on XGBoost for predicting sepsis. The model was specifically designed to use non-invasive vital signs captured at a single time point, The correlation between sepsis AI prediction model and levels of CRP and PCT was evaluated. The developed model demonstrated balanced performance across various datasets, with an average recall of 0.908 and precision of 0.577. The model's performance was further validated by the independent dataset from Cheng-Hsin General Hospital (recall: 0.986, precision: 0.585). Temperature, systolic blood pressure, and respiration rate were the top contributing predictors in the model. A significant correlation was observed between the model's sepsis predictions and elevated CRP levels, while PCT showed a less consistent pattern. Our approach, combining AI algorithms with vital sign data and its clinical relevance to CRP level, offers a more precise and timely sepsis detection, with the potential to improve care in emergency and critical care settings.</div></div>","PeriodicalId":73399,"journal":{"name":"Intelligence-based medicine","volume":"11 ","pages":"Article 100242"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143747482","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

An integrated machine learning based adaptive error minimization framework for Alzheimer's stage identification 基于集成机器学习的阿尔茨海默氏症阶段识别自适应误差最小化框架

Intelligence-based medicine

Pub Date : 2025-01-01 DOI: 10.1016/j.ibmed.2025.100243

Fahima Hossain, Rajib Kumar Halder, Mohammed Nasir Uddin

Alzheimer's disease (AD) is a degenerative neurological condition that impairs cognitive functioning. Early detection is critical for slowing disease progression and limiting brain damage. Although machine learning and deep learning models help identify Alzheimer's disease, their accuracy and efficiency are widely questioned. This study provides an integrated system for classifying four AD phases from 6400 MRI scans using pre-trained neural networks and machine learning classifiers. Preprocessing steps include noise removal, image enhancement (AGCWD, Bilateral Filter), and segmentation. Intensity normalization and data augmentation methods are applied to improve model generalization. Two models are developed: the first employs pre-trained neural net-works (VGG16, VGG19, DenseNet201, ResNet50, EfficientNetV7, InceptionV3, InceptionResNetV2, and MobileNet) for both feature extraction and classification. In contrast, the second integrates features from these networks with machine learning classifiers (XGBoost, Random Forest, SVM, KNN, Gradient Boosting, AdaBoost, Decision Tree, Linear Discriminant Analysis, Logistic Regression, and Multilayer Perceptron). The second model incorporates an adaptive error minimization sys-tem for enhanced accuracy. VGG16 achieved the highest accuracy (99.61 % training and 97.94 % testing), whereas VGG19+MLP with adaptive error minimization achieved 97.08 %, exhibiting superior AD classification ability.

阿尔茨海默病（AD）是一种退化的神经系统疾病，损害认知功能。早期发现对于减缓疾病进展和限制脑损伤至关重要。虽然机器学习和深度学习模型有助于识别阿尔茨海默病，但它们的准确性和效率受到广泛质疑。本研究提供了一个集成系统，使用预训练的神经网络和机器学习分类器从6400个MRI扫描中对四个AD阶段进行分类。预处理步骤包括去噪、图像增强（AGCWD，双边滤波）和分割。采用强度归一化和数据增强方法提高模型泛化能力。开发了两个模型：第一个模型使用预训练的神经网络（VGG16, VGG19, DenseNet201, ResNet50, EfficientNetV7, InceptionV3， InceptionResNetV2和MobileNet）进行特征提取和分类。相比之下，第二种将这些网络的特征与机器学习分类器（XGBoost、随机森林、SVM、KNN、梯度增强、AdaBoost、决策树、线性判别分析、逻辑回归和多层感知器）集成在一起。第二种模型采用自适应误差最小化系统来提高精度。VGG16的准确率最高（训练准确率为99.61%，测试准确率为97.94%），而VGG19+自适应误差最小化的MLP准确率为97.08%，表现出更强的AD分类能力。

{"title":"An integrated machine learning based adaptive error minimization framework for Alzheimer's stage identification","authors":"Fahima Hossain, Rajib Kumar Halder, Mohammed Nasir Uddin","doi":"10.1016/j.ibmed.2025.100243","DOIUrl":"10.1016/j.ibmed.2025.100243","url":null,"abstract":"<div><div>Alzheimer's disease (AD) is a degenerative neurological condition that impairs cognitive functioning. Early detection is critical for slowing disease progression and limiting brain damage. Although machine learning and deep learning models help identify Alzheimer's disease, their accuracy and efficiency are widely questioned. This study provides an integrated system for classifying four AD phases from 6400 MRI scans using pre-trained neural networks and machine learning classifiers. Preprocessing steps include noise removal, image enhancement (AGCWD, Bilateral Filter), and segmentation. Intensity normalization and data augmentation methods are applied to improve model generalization. Two models are developed: the first employs pre-trained neural net-works (VGG16, VGG19, DenseNet201, ResNet50, EfficientNetV7, InceptionV3, InceptionResNetV2, and MobileNet) for both feature extraction and classification. In contrast, the second integrates features from these networks with machine learning classifiers (XGBoost, Random Forest, SVM, KNN, Gradient Boosting, AdaBoost, Decision Tree, Linear Discriminant Analysis, Logistic Regression, and Multilayer Perceptron). The second model incorporates an adaptive error minimization sys-tem for enhanced accuracy. VGG16 achieved the highest accuracy (99.61 % training and 97.94 % testing), whereas VGG19+MLP with adaptive error minimization achieved 97.08 %, exhibiting superior AD classification ability.</div></div>","PeriodicalId":73399,"journal":{"name":"Intelligence-based medicine","volume":"11 ","pages":"Article 100243"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143807121","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

Using convolutional network in graphical model detection of autism disorders with fuzzy inference systems 基于模糊推理系统的卷积网络在自闭症障碍图形模型检测中的应用

Intelligence-based medicine

Pub Date : 2025-01-01 DOI: 10.1016/j.ibmed.2025.100213

S. Rajaprakash , C. Bagath Basha , C. Sunitha Ram , I. Ameethbasha , V. Subapriya , R. Sofia

Autism spectrum disorder (ASD) study faces several challenges, including variations in brain connectivity patterns, small sample sizes, and data heterogeneity detection by magnetic resonance imaging (MRI). These issues make it challenging to identify consistent imaging modalities. Researchers have explored improved analysis techniques to solve the above problem via multimodal imaging and graph-based methods. Therefore, it is better to understand ASD neurology. The current techniques focus mainly on pairwise comparisons between individuals and often overlook features and individual characteristics. To overcome these limitations, in the proposed novel method, a multiscale enhanced graph with a convolutional network is used for ASD detection.

This work integrates non-imaging phenotypic data (from brain imaging data) with functional connectivity data (from Functional magnetic resonance images). In this approach, the population graph represents all individuals as vertices. The phenotypic data were used to calculate the weight between vertices in the graph using the fuzzy inference system. Fuzzy if-then rules, is used to determine the similarity between the phenotypic data. Each vertice connects feature vectors derived from the image data. The vertices and weights of each edge are used to incorporate phenotypic information. A random walk with a fuzzy MSE-GCN framework employs multiple parallel GCN layer embeddings. The outputs from these layers are joined in a completely linked layer to detect ASD efficiently. We assessed the performance of this background by the ABIDE data set and utilized recursive feature elimination and a multilayer perceptron for feature selection. This method achieved an accuracy rate of 87 % better than the current study.

自闭症谱系障碍（ASD）的研究面临着一些挑战，包括脑连接模式的差异、小样本量和磁共振成像（MRI）数据异质性检测。这些问题使得确定一致的成像模式具有挑战性。研究人员已经探索了改进的分析技术，通过多模态成像和基于图的方法来解决上述问题。因此，更好地了解ASD神经学。目前的技术主要集中在个体之间的两两比较，往往忽略了特征和个体特征。为了克服这些局限性，本文提出了一种基于卷积网络的多尺度增强图检测ASD的新方法。这项工作整合了非成像表型数据（来自脑成像数据）和功能连接数据（来自功能磁共振图像）。在这种方法中，总体图将所有个体表示为顶点。利用表型数据，利用模糊推理系统计算图中顶点之间的权重。模糊if-then规则用于确定表型数据之间的相似性。每个顶点连接来自图像数据的特征向量。每个边的顶点和权值被用来合并表型信息。模糊MSE-GCN框架的随机漫步采用多个并行GCN层嵌入。这些层的输出连接在一个完全链接的层中，以有效地检测ASD。我们通过ABIDE数据集评估了该背景的性能，并利用递归特征消除和多层感知器进行特征选择。该方法的准确率比目前的研究提高了87%。

{"title":"Using convolutional network in graphical model detection of autism disorders with fuzzy inference systems","authors":"S. Rajaprakash , C. Bagath Basha , C. Sunitha Ram , I. Ameethbasha , V. Subapriya , R. Sofia","doi":"10.1016/j.ibmed.2025.100213","DOIUrl":"10.1016/j.ibmed.2025.100213","url":null,"abstract":"<div><div>Autism spectrum disorder (ASD) study faces several challenges, including variations in brain connectivity patterns, small sample sizes, and data heterogeneity detection by magnetic resonance imaging (MRI). These issues make it challenging to identify consistent imaging modalities. Researchers have explored improved analysis techniques to solve the above problem via multimodal imaging and graph-based methods. Therefore, it is better to understand ASD neurology. The current techniques focus mainly on pairwise comparisons between individuals and often overlook features and individual characteristics. To overcome these limitations, in the proposed novel method, a multiscale enhanced graph with a convolutional network is used for ASD detection.</div><div>This work integrates non-imaging phenotypic data (from brain imaging data) with functional connectivity data (from Functional magnetic resonance images). In this approach, the population graph represents all individuals as vertices. The phenotypic data were used to calculate the weight between vertices in the graph using the fuzzy inference system. Fuzzy if-then rules, is used to determine the similarity between the phenotypic data. Each vertice connects feature vectors derived from the image data. The vertices and weights of each edge are used to incorporate phenotypic information. A random walk with a fuzzy MSE-GCN framework employs multiple parallel GCN layer embeddings. The outputs from these layers are joined in a completely linked layer to detect ASD efficiently. We assessed the performance of this background by the ABIDE data set and utilized recursive feature elimination and a multilayer perceptron for feature selection. This method achieved an accuracy rate of 87 % better than the current study.</div></div>","PeriodicalId":73399,"journal":{"name":"Intelligence-based medicine","volume":"11 ","pages":"Article 100213"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143350535","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Predicting the effect of Bevacizumab therapy in ovarian cancer from H&E whole slide images using transformer model 利用变压器模型从H&E全幻灯片图像预测贝伐单抗治疗卵巢癌的效果

Intelligence-based medicine

Pub Date : 2025-01-01 DOI: 10.1016/j.ibmed.2025.100231

Md Shakhawat Hossain , Munim Ahmed , Md Sahilur Rahman , MM Mahbubul Syeed , Mohammad Faisal Uddin

Ovarian cancer (OC) ranks fifth in all cancer-related fatalities in women. Epithelial ovarian cancer (EOC) is a subclass of OC, accounting for 95 % of all patients. Conventional treatment for EOC is debulking surgery with adjuvant Chemotherapy; however, in 70 % of cases, this leads to progressive resistance and tumor recurrence. The United States Food and Drug Administration (FDA) recently approved Bevacizumab therapy for EOC patients. Bevacizumab improved survival and decreased recurrence in 30 % of cases, while the rest reported side effects, which include severe hypertension (27 %), thrombocytopenia (26 %), bleeding issues (39 %), heart problems (11 %), kidney problems (7 %), intestinal perforation and delayed wound healing. Moreover, it is costly; single-cycle Bevacizumab therapy costs approximately $3266. Therefore, selecting patients for this therapy is critical due to the high cost, probable adverse effects and small beneficiaries. Several methods were proposed previously; however, they failed to attain adequate accuracy. We present an AI-driven method to predict the effect from H&E whole slide image (WSI) produced from a patient's biopsy. We trained multiple CNN and transformer models using 10 × and 20 × images to predict the effect. Finally, the Data Efficient Image Transformer (DeiT) model was selected considering its high accuracy, interoperability and time efficiency. The proposed method achieved 96.60 % test accuracy and 93 % accuracy in 5-fold cross-validation and can predict the effect in less than 30 s. This method outperformed the state-of-the-art test accuracy (85.10 %) by 11 % and cross-validation accuracy (88.2 %) by 5 %. High accuracy and low prediction time ensured the efficacy of the proposed method.

卵巢癌（OC）在所有与癌症相关的女性死亡中排名第五。上皮性卵巢癌（EOC）是卵巢癌的一个亚类，占所有患者的95%。EOC的常规治疗是减体积手术加辅助化疗；然而，在70%的病例中，这导致了逐渐的耐药性和肿瘤复发。美国食品和药物管理局（FDA）最近批准了贝伐单抗治疗EOC患者。在30%的病例中，贝伐单抗提高了生存率并降低了复发率，而其余的病例报告了副作用，包括严重高血压（27%）、血小板减少（26%）、出血问题（39%）、心脏问题（11%）、肾脏问题（7%）、肠穿孔和伤口愈合延迟。此外，它是昂贵的；单周期贝伐单抗治疗费用约为3266美元。因此，选择患者进行这种治疗是至关重要的，因为成本高，可能的不良反应和小的受益者。之前提出了几种方法；然而，他们未能达到足够的准确性。我们提出了一种人工智能驱动的方法来预测从患者活检产生的H&；E全幻灯片图像（WSI）的效果。我们使用10 ×和20 ×图像训练多个CNN和transformer模型来预测效果。最后，考虑到数据高效图像转换器（Data Efficient Image Transformer, DeiT）模型具有较高的精度、互操作性和时间效率，选择了DeiT模型。该方法的试验准确度为96.60%，5次交叉验证准确度为93%，可在30 s内预测效果。该方法比目前最先进的测试准确度（85.10%）提高了11%，交叉验证准确度（88.2%）提高了5%。较高的预测精度和较短的预测时间保证了该方法的有效性。

{"title":"Predicting the effect of Bevacizumab therapy in ovarian cancer from H&E whole slide images using transformer model","authors":"Md Shakhawat Hossain , Munim Ahmed , Md Sahilur Rahman , MM Mahbubul Syeed , Mohammad Faisal Uddin","doi":"10.1016/j.ibmed.2025.100231","DOIUrl":"10.1016/j.ibmed.2025.100231","url":null,"abstract":"<div><div>Ovarian cancer (OC) ranks fifth in all cancer-related fatalities in women. Epithelial ovarian cancer (EOC) is a subclass of OC, accounting for 95 % of all patients. Conventional treatment for EOC is debulking surgery with adjuvant Chemotherapy; however, in 70 % of cases, this leads to progressive resistance and tumor recurrence. The United States Food and Drug Administration (FDA) recently approved Bevacizumab therapy for EOC patients. Bevacizumab improved survival and decreased recurrence in 30 % of cases, while the rest reported side effects, which include severe hypertension (27 %), thrombocytopenia (26 %), bleeding issues (39 %), heart problems (11 %), kidney problems (7 %), intestinal perforation and delayed wound healing. Moreover, it is costly; single-cycle Bevacizumab therapy costs approximately $3266. Therefore, selecting patients for this therapy is critical due to the high cost, probable adverse effects and small beneficiaries. Several methods were proposed previously; however, they failed to attain adequate accuracy. We present an AI-driven method to predict the effect from H&E whole slide image (WSI) produced from a patient's biopsy. We trained multiple CNN and transformer models using 10 × and 20 × images to predict the effect. Finally, the Data Efficient Image Transformer (DeiT) model was selected considering its high accuracy, interoperability and time efficiency. The proposed method achieved 96.60 % test accuracy and 93 % accuracy in 5-fold cross-validation and can predict the effect in less than 30 s. This method outperformed the state-of-the-art test accuracy (85.10 %) by 11 % and cross-validation accuracy (88.2 %) by 5 %. High accuracy and low prediction time ensured the efficacy of the proposed method.</div></div>","PeriodicalId":73399,"journal":{"name":"Intelligence-based medicine","volume":"11 ","pages":"Article 100231"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143601341","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Using big data to predict young adult ischemic vs. non-ischemic heart disease risk factors: An artificial intelligence based model 利用大数据预测年轻人缺血性与非缺血性心脏病危险因素：基于人工智能的模型

Intelligence-based medicine

Pub Date : 2025-01-01 DOI: 10.1016/j.ibmed.2025.100207

Salam Bani Hani , Muayyad Ahmad

引用次数: 0

Open-source small language models for personal medical assistant chatbots 个人医疗助理聊天机器人的开源小语言模型

Intelligence-based medicine

Pub Date : 2025-01-01 DOI: 10.1016/j.ibmed.2024.100197

Matteo Magnini , Gianluca Aguzzi , Sara Montagna

Medical chatbots are becoming essential components of telemedicine applications as tools to assist patients in the self-management of their conditions. This trend is particularly driven by advancements in natural language processing techniques with pre-trained language models (LMs). However, the integration of LMs into clinical environments faces challenges related to reliability and privacy concerns.

This study seeks to address these issues by exploiting a privacy by design architectural solution that utilises the fully local deployment of open-source LMs. Specifically, to mitigate any risk of information leakage, we focus on evaluating the performance of open-source language models (SLMs) that can be deployed on personal devices, such as smartphones or laptops, without stringent hardware requirements.

We assess the effectiveness of this solution adopting hypertension management as a case study. Models are evaluated across various tasks, including intent recognition and empathetic conversation, using Gemini Pro 1.5 as a benchmark. The results indicate that, for certain tasks such as intent recognition, Gemini outperforms other models. However, by employing the “large language model (LLM) as a judge” approach for semantic evaluation of response correctness, we found several models that demonstrate a close alignment with the ground truth. In conclusion, this study highlights the potential of locally deployed SLMs as components of medical chatbots, while addressing critical concerns related to privacy and reliability.

医疗聊天机器人正在成为远程医疗应用的重要组成部分，作为帮助患者自我管理病情的工具。这一趋势尤其受到自然语言处理技术与预训练语言模型（LMs）的进步的推动。然而，将LMs集成到临床环境中面临着与可靠性和隐私问题相关的挑战。本研究试图通过利用开源LMs的完全本地部署来利用隐私设计架构解决方案来解决这些问题。具体来说，为了减少信息泄露的风险，我们着重于评估可以部署在个人设备（如智能手机或笔记本电脑）上的开源语言模型（slm）的性能，而不需要严格的硬件要求。我们以高血压管理为例来评估这种解决方案的有效性。模型在各种任务中进行评估，包括意图识别和移情对话，使用Gemini Pro 1.5作为基准。结果表明，对于某些任务，如意图识别，Gemini优于其他模型。然而，通过采用“大型语言模型（LLM）作为判断”的方法来对响应正确性进行语义评估，我们发现了几个与基本事实密切一致的模型。总之，本研究强调了本地部署的slm作为医疗聊天机器人组件的潜力，同时解决了与隐私和可靠性相关的关键问题。

{"title":"Open-source small language models for personal medical assistant chatbots","authors":"Matteo Magnini , Gianluca Aguzzi , Sara Montagna","doi":"10.1016/j.ibmed.2024.100197","DOIUrl":"10.1016/j.ibmed.2024.100197","url":null,"abstract":"<div><div>Medical chatbots are becoming essential components of telemedicine applications as tools to assist patients in the self-management of their conditions. This trend is particularly driven by advancements in natural language processing techniques with pre-trained language models (LMs). However, the integration of LMs into clinical environments faces challenges related to reliability and privacy concerns.</div><div>This study seeks to address these issues by exploiting a <em>privacy by design</em> architectural solution that utilises the fully local deployment of open-source LMs. Specifically, to mitigate any risk of information leakage, we focus on evaluating the performance of open-source language models (SLMs) that can be deployed on personal devices, such as smartphones or laptops, without stringent hardware requirements.</div><div>We assess the effectiveness of this solution adopting hypertension management as a case study. Models are evaluated across various tasks, including intent recognition and empathetic conversation, using Gemini Pro 1.5 as a benchmark. The results indicate that, for certain tasks such as intent recognition, Gemini outperforms other models. However, by employing the “large language model (LLM) as a judge” approach for semantic evaluation of response correctness, we found several models that demonstrate a close alignment with the ground truth. In conclusion, this study highlights the potential of locally deployed SLMs as components of medical chatbots, while addressing critical concerns related to privacy and reliability.</div></div>","PeriodicalId":73399,"journal":{"name":"Intelligence-based medicine","volume":"11 ","pages":"Article 100197"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143173635","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Development and validation of a moderate aortic stenosis disease progression model 中度主动脉瓣狭窄疾病进展模型的建立和验证

Intelligence-based medicine

Pub Date : 2025-01-01 DOI: 10.1016/j.ibmed.2025.100201

Miguel R. Sotelo , Paul Nona , Loren Wagner , Chris Rogers , Julian Booker , Efstathia Andrikopoulou

Background

Understanding the multifactorial determinants of rapid progression in patients with aortic stenosis (AS) remains limited. We aimed to develop and validate a machine learning model (ML) for predicting rapid progression from moderate to severe AS within one year.

Methods

8746 patients were identified with moderate AS across seven healthcare organizations. Three ML models were trained using demographic, and echocardiographic variables, namely Random Forest, XGBoost and causal discovery-logistic regression. An ensemble model was developed integrating the aforementioned three. A total of 3355 patients formed the training and internal validation cohort. External validation was performed on 171 patients from one institution.

Results

An ensemble model was selected due to its superior F1 score and precision in internal validation (0.382 and 0.301, respectively). Its performance on the external validation cohort was modest (F1 score = 0.626, precision = 0.532).

Conclusion

An ensemble model comprising only demographic and echocardiographic variables was shown to have modest performance in predicting one-year progression from moderate to severe AS. Further validation in larger populations, along with integration of comprehensive clinical data, is crucial for broader applicability.

背景：对主动脉瓣狭窄（AS）患者快速进展的多因素决定因素的了解仍然有限。我们的目标是开发和验证一种机器学习模型（ML），用于预测一年内从中度到重度AS的快速进展。方法7家医疗机构共8746例中度AS患者。使用人口统计学和超声心动图变量，即随机森林、XGBoost和因果发现-逻辑回归，训练了三个ML模型。开发了一个集成模型，将上述三者集成在一起。共有3355名患者组成了培训和内部验证队列。对来自同一机构的171例患者进行了外部验证。结果集成模型在内部验证中F1得分和精密度均较优（分别为0.382和0.301）。其在外部验证队列中的表现一般（F1评分= 0.626，精密度= 0.532）。结论仅包含人口统计学和超声心动图变量的集成模型在预测中度到重度AS的一年进展方面表现不佳。在更大的人群中进一步验证，以及综合临床数据的整合，对于更广泛的适用性至关重要。

{"title":"Development and validation of a moderate aortic stenosis disease progression model","authors":"Miguel R. Sotelo , Paul Nona , Loren Wagner , Chris Rogers , Julian Booker , Efstathia Andrikopoulou","doi":"10.1016/j.ibmed.2025.100201","DOIUrl":"10.1016/j.ibmed.2025.100201","url":null,"abstract":"<div><h3>Background</h3><div>Understanding the multifactorial determinants of rapid progression in patients with aortic stenosis (AS) remains limited. We aimed to develop and validate a machine learning model (ML) for predicting rapid progression from moderate to severe AS within one year.</div></div><div><h3>Methods</h3><div>8746 patients were identified with moderate AS across seven healthcare organizations. Three ML models were trained using demographic, and echocardiographic variables, namely Random Forest, XGBoost and causal discovery-logistic regression. An ensemble model was developed integrating the aforementioned three. A total of 3355 patients formed the training and internal validation cohort. External validation was performed on 171 patients from one institution.</div></div><div><h3>Results</h3><div>An ensemble model was selected due to its superior F1 score and precision in internal validation (0.382 and 0.301, respectively). Its performance on the external validation cohort was modest (F1 score = 0.626, precision = 0.532).</div></div><div><h3>Conclusion</h3><div>An ensemble model comprising only demographic and echocardiographic variables was shown to have modest performance in predicting one-year progression from moderate to severe AS. Further validation in larger populations, along with integration of comprehensive clinical data, is crucial for broader applicability.</div></div>","PeriodicalId":73399,"journal":{"name":"Intelligence-based medicine","volume":"11 ","pages":"Article 100201"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143174355","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

ESC-UNET: A hybrid CNN and Swin Transformers for skin lesion segmentation ESC-UNET：一种混合CNN和Swin变压器的皮肤损伤分割方法

Intelligence-based medicine

Pub Date : 2025-01-01 DOI: 10.1016/j.ibmed.2025.100257

Anwar Jimi , Nabila Zrira , Oumaima Guendoul , Ibtissam Benmiloud , Haris Ahmad Khan , Shah Nawaz

One of the most important tasks in computer-aided diagnostics is the automatic segmentation of skin lesions, which plays an essential role in the early diagnosis and treatment of skin cancer. In recent years, the Convolutional Neural Network (CNN) has largely replaced other traditional methods for segmenting skin lesions. However, due to insufficient information and unclear lesion region segmentation, skin lesion image segmentation still has challenges. In this paper, we propose a novel deep medical image segmentation approach named “ESC-UNET” which combines the advantages of CNN and Transformer to effectively leverage local information and long-range dependencies to enhance medical image segmentation. In terms of the local information, we use a CNN-based encoder and decoder framework. The CNN branch mines local information from medical images using the locality of convolution processes and the pre-trained EfficientNetB5 network. As for the long-range dependencies, we build a Transformer branch that emphasizes the global context. In addition, we employ Atrous Spatial Pyramid Pooling (ASPP) to gather network-wide relevant information. The Convolution Block Attention Module (CBAM) is added to the model to promote effective features and suppress ineffective features in segmentation. We have evaluated our network using the ISIC 2016, ISIC 2017, and ISIC 2018 datasets. The results demonstrate the efficiency of the proposed model in segmenting skin lesions.

计算机辅助诊断中最重要的任务之一是皮肤病变的自动分割，这对皮肤癌的早期诊断和治疗起着至关重要的作用。近年来，卷积神经网络（CNN）在很大程度上取代了其他传统的皮肤损伤分割方法。然而，由于信息不足和病灶区域分割不清，皮肤病灶图像分割仍然存在挑战。本文提出了一种新的医学图像深度分割方法“ESC-UNET”，该方法结合了CNN和Transformer的优点，有效地利用了局部信息和远程依赖关系来增强医学图像分割。在局部信息方面，我们使用了基于cnn的编码器和解码器框架。CNN分支使用卷积过程的局部性和预训练的effentnetb5网络从医学图像中挖掘局部信息。至于远程依赖，我们构建一个强调全局上下文的Transformer分支。此外，我们采用亚特劳斯空间金字塔池（ASPP）来收集全网络的相关信息。在该模型中加入了卷积块注意模块（CBAM），在分割中提升有效特征，抑制无效特征。我们使用ISIC 2016、ISIC 2017和ISIC 2018数据集评估了我们的网络。实验结果证明了该模型在皮肤损伤分割方面的有效性。

{"title":"ESC-UNET: A hybrid CNN and Swin Transformers for skin lesion segmentation","authors":"Anwar Jimi , Nabila Zrira , Oumaima Guendoul , Ibtissam Benmiloud , Haris Ahmad Khan , Shah Nawaz","doi":"10.1016/j.ibmed.2025.100257","DOIUrl":"10.1016/j.ibmed.2025.100257","url":null,"abstract":"<div><div>One of the most important tasks in computer-aided diagnostics is the automatic segmentation of skin lesions, which plays an essential role in the early diagnosis and treatment of skin cancer. In recent years, the Convolutional Neural Network (CNN) has largely replaced other traditional methods for segmenting skin lesions. However, due to insufficient information and unclear lesion region segmentation, skin lesion image segmentation still has challenges. In this paper, we propose a novel deep medical image segmentation approach named “ESC-UNET” which combines the advantages of CNN and Transformer to effectively leverage local information and long-range dependencies to enhance medical image segmentation. In terms of the local information, we use a CNN-based encoder and decoder framework. The CNN branch mines local information from medical images using the locality of convolution processes and the pre-trained EfficientNetB5 network. As for the long-range dependencies, we build a Transformer branch that emphasizes the global context. In addition, we employ Atrous Spatial Pyramid Pooling (ASPP) to gather network-wide relevant information. The Convolution Block Attention Module (CBAM) is added to the model to promote effective features and suppress ineffective features in segmentation. We have evaluated our network using the ISIC 2016, ISIC 2017, and ISIC 2018 datasets. The results demonstrate the efficiency of the proposed model in segmenting skin lesions.</div></div>","PeriodicalId":73399,"journal":{"name":"Intelligence-based medicine","volume":"12 ","pages":"Article 100257"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144168078","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