Smart Health最新文献_第5页

A low-channel EEG-to-speech conversion approach for assisting people with communication disorders 一种帮助有沟通障碍人士的低通道脑电图-语言转换方法

Q2 Health Professions

Smart Health

Pub Date : 2025-03-26 DOI: 10.1016/j.smhl.2025.100568

Kunning Shen , Huining Li

Brain–Computer Interface (BCI) technology has emerged as a promising solution for individuals with communication disorders. However, current electroencephalography (EEG) to speech systems typically require high-channel EEG equipment (64＋ channels), limiting their accessibility in resource-constrained environments. This paper implements a novel low-channel EEG-to-speech framework that effectively operates with only 6 EEG channels. By leveraging a generator-discriminator architecture for speech reconstruction, our system achieves a Character Error Rate (CER) of 64.24%, outperforming baseline systems that utilize 64 channels (68.26% CER). We further integrate Undercomplete Independent Component Analysis (UICA) for channel reduction, maintaining comparable accuracy (64.99% CER) while reducing computational complexity from 6 channels to 4 channels. This breakthrough demonstrates the feasibility of efficient speech reconstruction from minimal EEG inputs, potentially enabling more widespread deployment of BCI technology in resource-limited healthcare settings.

脑机接口（BCI）技术已经成为一种有希望的解决个人沟通障碍的解决方案。然而，目前的脑电图（EEG）到语音系统通常需要高通道EEG设备（64+通道），这限制了它们在资源受限环境中的可访问性。本文实现了一种新颖的低通道脑电转语音框架，该框架仅在6个脑电通道下有效地工作。通过利用生成-鉴别器架构进行语音重建，我们的系统实现了64.24%的字符错误率（CER），优于使用64通道的基准系统（68.26%的CER）。我们进一步集成了欠完全独立分量分析（UICA）来减少信道，保持了相当的精度（64.99% CER），同时将计算复杂度从6个信道降低到4个信道。这一突破证明了从最小的脑电图输入进行高效语音重建的可行性，有可能使BCI技术在资源有限的医疗保健环境中得到更广泛的应用。

引用次数: 0

Background parenchymal uptake classification using deep transfer learning on digital mammograms 背景：基于深度迁移学习的数字化乳房x光片实质摄取分类

Q2 Health Professions

Smart Health

Pub Date : 2025-03-26 DOI: 10.1016/j.smhl.2025.100573

Xudong Liu , Christopher Scott , Imon Banerjee , Celine Vachon , Carrie Hruska

Background parenchymal uptake (BPU) in fibroglandular tissue on a molecular breast image (MBI) has been shown to be a strong risk factor for breast cancer and complementary to mammographic density. However, MBI is generally performed on women with dense breasts and only available at institutions with nuclear medicine capabilities, limiting the utility of this measure in routine breast screening and risk assessment. Digital mammography is used for routine breast screening. Our goal was to evaluate whether BPU features could be identified from digital mammograms (DMs) using deep transfer learning. Specifically, we identified a cohort of about 2000 women from a breast screening center who had DM and MBI performed at the same time period and trained models on DMs to classify BPU categories. We consider two types of classification problems in this work: a five-category classification of BPU and two combined classes. We designed and implemented machine learning algorithms leveraging state-of-the-art pre-trained deep neural networks, evaluated these algorithms on the collected data based using metrics such as accuracy, F1-score, and AUROC, and provided visual explanations using saliency mapping and gradient-weighted class activation mapping (GradCAM). Our results show that, among the experimented models, WideResNet-50 demonstrates the best performance on a hold-out test set with 58% accuracy, 0.82 micro-average AUROC and 0.72 macro-average AUROC on the five-category classification, while ResNet-18 comes out on top with 77% accuracy, 0.86 AUROC and 0.77 F1-score on the binary categorization. We also found that incorporating age, body mass index (BMI) and menopausal status improved classification of BPU compared to DM alone.

背景：在分子乳腺图像（MBI）上纤维腺组织的实质摄取（BPU）已被证明是乳腺癌的一个强大的危险因素，并与乳房x线摄影密度互补。然而，MBI通常是对乳房致密的妇女进行的，而且只有在具有核医学能力的机构才能进行，这限制了这项措施在常规乳房筛查和风险评估中的效用。数字乳房x线照相术用于常规乳房筛查。我们的目标是评估是否可以使用深度迁移学习从数字乳房x光片（dm）中识别BPU特征。具体来说，我们从一个乳房筛查中心确定了一个大约2000名女性的队列，这些女性在同一时期接受了糖尿病和MBI，并训练了DM模型来分类BPU类别。在这项工作中，我们考虑了两种类型的分类问题：BPU的五类分类和两个组合类。我们设计并实现了利用最先进的预训练深度神经网络的机器学习算法，使用精确度、f1分数和AUROC等指标在收集的数据上评估这些算法，并使用显著性映射和梯度加权类激活映射（GradCAM）提供可视化解释。结果表明，在实验模型中，WideResNet-50在hold-out测试集上表现最佳，在五类分类上的准确率为58%，微观平均AUROC为0.82，宏观平均AUROC为0.72，而ResNet-18在二元分类上的准确率为77%，AUROC为0.86，f1得分为0.77。我们还发现，与单独治疗糖尿病相比，结合年龄、体重指数（BMI）和绝经状态可改善BPU的分类。

{"title":"Background parenchymal uptake classification using deep transfer learning on digital mammograms","authors":"Xudong Liu , Christopher Scott , Imon Banerjee , Celine Vachon , Carrie Hruska","doi":"10.1016/j.smhl.2025.100573","DOIUrl":"10.1016/j.smhl.2025.100573","url":null,"abstract":"<div><div>Background parenchymal uptake (BPU) in fibroglandular tissue on a molecular breast image (MBI) has been shown to be a strong risk factor for breast cancer and complementary to mammographic density. However, MBI is generally performed on women with dense breasts and only available at institutions with nuclear medicine capabilities, limiting the utility of this measure in routine breast screening and risk assessment. Digital mammography is used for routine breast screening. Our goal was to evaluate whether BPU features could be identified from digital mammograms (DMs) using deep transfer learning. Specifically, we identified a cohort of about 2000 women from a breast screening center who had DM and MBI performed at the same time period and trained models on DMs to classify BPU categories. We consider two types of classification problems in this work: a five-category classification of BPU and two combined classes. We designed and implemented machine learning algorithms leveraging state-of-the-art pre-trained deep neural networks, evaluated these algorithms on the collected data based using metrics such as accuracy, F1-score, and AUROC, and provided visual explanations using saliency mapping and gradient-weighted class activation mapping (GradCAM). Our results show that, among the experimented models, WideResNet-50 demonstrates the best performance on a hold-out test set with 58% accuracy, 0.82 micro-average AUROC and 0.72 macro-average AUROC on the five-category classification, while ResNet-18 comes out on top with 77% accuracy, 0.86 AUROC and 0.77 F1-score on the binary categorization. We also found that incorporating age, body mass index (BMI) and menopausal status improved classification of BPU compared to DM alone.</div></div>","PeriodicalId":37151,"journal":{"name":"Smart Health","volume":"36 ","pages":"Article 100573"},"PeriodicalIF":0.0,"publicationDate":"2025-03-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143747323","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

Mobile app-based study of driving behaviors under the influence of cannabis 基于手机app的大麻影响下驾驶行为研究

Q2 Health Professions

Smart Health

Pub Date : 2025-03-26 DOI: 10.1016/j.smhl.2025.100558

Honglu Li , Bin Han , Cong Shi , Yan Wang , Tammy Chung , Yingying Chen

Cannabis use has become increasingly prevalent due to evolving legal and societal attitudes, raising concerns about its influence on public safety, particularly in driving. Existing studies mostly rely on simulators or specialized equipment, which do not capture the complexities of real-world driving and pose cost and scalability issues. In this paper, we investigate the effects of cannabis on driving behavior using participants’ smartphones to gather data in natural settings. Our method focuses on three critical behaviors: weaving & swerving, wide turning, and hard braking. We propose a two-step segmentation algorithm for processing continuous motion sensor data and use threshold-based methods for efficient detection. A custom application autonomously records driving events during actual road scenarios. On-road experiments with 9 participants who consumed cannabis under controlled conditions reveal a correlation between cannabis use and altered driving behaviors, with significant effects emerging approximately 2

\sim

3 h after consumption.

由于法律和社会态度的变化，大麻的使用越来越普遍，这引起了人们对其对公共安全，特别是驾驶的影响的关注。现有的研究大多依赖于模拟器或专用设备，这些设备无法捕捉到真实驾驶的复杂性，并且存在成本和可扩展性问题。在本文中，我们研究了大麻对驾驶行为的影响，使用参与者的智能手机在自然环境中收集数据。我们的方法侧重于三个关键行为：编织；急转弯、大转弯和急刹车。我们提出了一种两步分割算法来处理连续运动传感器数据，并使用基于阈值的方法进行有效的检测。自定义应用程序在实际道路场景中自动记录驾驶事件。对9名在受控条件下吸食大麻的参与者进行的道路实验显示，吸食大麻与改变驾驶行为之间存在相关性，在吸食后约2 ~ 3小时出现显著影响。

引用次数: 0

SNOMED CT ontology multi-relation classification by using knowledge embedding in neural network 基于神经网络知识嵌入的SNOMED CT本体多关系分类

Q2 Health Professions

Smart Health

Pub Date : 2025-03-26 DOI: 10.1016/j.smhl.2025.100560

Bofan He, Jerry Q. Cheng, Huanying Gu

SNOMED CT is a widely recognized healthcare terminology designed to comprehensively represent clinical knowledge. Identifying missing or incorrect relationships between medical concepts is crucial for enhancing the scope and quality of this ontology, thereby improving healthcare analytics and decision support. In this study, we propose a novel multi-link prediction approach that utilizes knowledge graph embeddings and neural networks to infer missing relationships within the SNOMED CT knowledge graph. By utilizing TransE, we train embeddings for triples (concept, relation, concept) and develop a multi-head classifier to predict relationship types based solely on concept pairs. With an embedding dimension of 200, a batch size of 128, and 10 epochs, we achieved the highest test accuracy of 91.96% in relationships prediction tasks. This study demonstrates an optimal balance between efficiency, generalization, and representational capacity. By expanding on existing methodologies, this work offers insights into practical applications for ontology enrichment and contributes to the ongoing advancement of predictive models in healthcare informatics. Furthermore, it highlights the potential scalability of the approach, providing a framework that can be extended to other knowledge graphs and domains.

SNOMED CT是一个广泛认可的医疗保健术语，旨在全面代表临床知识。识别医学概念之间缺失或不正确的关系对于增强本体论的范围和质量至关重要，从而改进医疗保健分析和决策支持。在这项研究中，我们提出了一种新的多链路预测方法，该方法利用知识图嵌入和神经网络来推断SNOMED CT知识图中的缺失关系。通过使用TransE，我们训练了三元组（概念、关系、概念）的嵌入，并开发了一个多头分类器，仅基于概念对来预测关系类型。在嵌入维数为200、批大小为128、epoch为10的情况下，我们在关系预测任务中获得了最高的测试准确率91.96%。本研究展示了效率、泛化和表征能力之间的最佳平衡。通过扩展现有的方法，这项工作为丰富本体的实际应用提供了见解，并有助于医疗信息学预测模型的持续发展。此外，它强调了该方法的潜在可扩展性，提供了一个可以扩展到其他知识图和领域的框架。

{"title":"SNOMED CT ontology multi-relation classification by using knowledge embedding in neural network","authors":"Bofan He, Jerry Q. Cheng, Huanying Gu","doi":"10.1016/j.smhl.2025.100560","DOIUrl":"10.1016/j.smhl.2025.100560","url":null,"abstract":"<div><div>SNOMED CT is a widely recognized healthcare terminology designed to comprehensively represent clinical knowledge. Identifying missing or incorrect relationships between medical concepts is crucial for enhancing the scope and quality of this ontology, thereby improving healthcare analytics and decision support. In this study, we propose a novel multi-link prediction approach that utilizes knowledge graph embeddings and neural networks to infer missing relationships within the SNOMED CT knowledge graph. By utilizing TransE, we train embeddings for triples (concept, relation, concept) and develop a multi-head classifier to predict relationship types based solely on concept pairs. With an embedding dimension of 200, a batch size of 128, and 10 epochs, we achieved the highest test accuracy of 91.96% in relationships prediction tasks. This study demonstrates an optimal balance between efficiency, generalization, and representational capacity. By expanding on existing methodologies, this work offers insights into practical applications for ontology enrichment and contributes to the ongoing advancement of predictive models in healthcare informatics. Furthermore, it highlights the potential scalability of the approach, providing a framework that can be extended to other knowledge graphs and domains.</div></div>","PeriodicalId":37151,"journal":{"name":"Smart Health","volume":"36 ","pages":"Article 100560"},"PeriodicalIF":0.0,"publicationDate":"2025-03-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143760190","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Exploring finetuned audio-LLM on heart murmur features 探索微调音频- llm心脏杂音的特点

Q2 Health Professions

Smart Health

Pub Date : 2025-03-26 DOI: 10.1016/j.smhl.2025.100557

Adrian Florea, Xilin Jiang, Nima Mesgarani, Xiaofan Jiang

Large language models (LLMs) for audio have excelled in recognizing and analyzing human speech, music, and environmental sounds. However, their potential for understanding other types of sounds, particularly biomedical sounds, remains largely underexplored despite significant scientific interest. In this study, we focus on diagnosing cardiovascular diseases using phonocardiograms, i.e., heart sounds. Most existing deep neural network (DNN) paradigms are restricted to heart murmur classification (healthy vs unhealthy) and do not predict other acoustic features of the murmur such as grading, harshness, pitch, and quality, which are important in helping physicians diagnose the underlying heart conditions. We propose to finetune an audio LLM, Qwen2-Audio, on the PhysioNet CirCor DigiScope phonocardiogram (PCG) dataset and evaluate its performance in classifying 11 expert-labeled features. Additionally, we aim to achieve more noise-robust and generalizable system by exploring a preprocessing segmentation algorithm using an audio representation model, SSAMBA. Our results indicate that the LLM-based model outperforms state-of-the-art methods in 10 of the 11 tasks. Moreover, the LLM successfully classifies long-tail features with limited training data, a task that all previous methods have failed to classify. These findings underscore the potential of audio LLMs as assistants to human cardiologists in enhancing heart disease diagnosis.

用于音频的大型语言模型（llm）在识别和分析人类语音、音乐和环境声音方面表现出色。然而，它们理解其他类型声音的潜力，特别是生物医学声音，在很大程度上仍未得到充分开发，尽管有重大的科学兴趣。在这项研究中，我们的重点是使用心音图，即心音来诊断心血管疾病。大多数现有的深度神经网络（DNN）范例仅限于心脏杂音的分类（健康与不健康），而不能预测杂音的其他声学特征，如分级、刺耳程度、音高和质量，这些对帮助医生诊断潜在的心脏疾病很重要。我们建议在PhysioNet CirCor DigiScope心音图（PCG）数据集上对音频LLM Qwen2-Audio进行微调，并评估其在分类11个专家标记特征方面的性能。此外，我们的目标是通过探索使用音频表示模型SSAMBA的预处理分割算法来实现更强的噪声鲁棒性和可泛化的系统。我们的结果表明，基于法学硕士的模型在11个任务中的10个任务中优于最先进的方法。此外，LLM在有限的训练数据下成功地对长尾特征进行了分类，这是以前所有方法都无法分类的任务。这些发现强调了音频llm作为人类心脏病专家在加强心脏病诊断方面的助手的潜力。

{"title":"Exploring finetuned audio-LLM on heart murmur features","authors":"Adrian Florea, Xilin Jiang, Nima Mesgarani, Xiaofan Jiang","doi":"10.1016/j.smhl.2025.100557","DOIUrl":"10.1016/j.smhl.2025.100557","url":null,"abstract":"<div><div>Large language models (LLMs) for audio have excelled in recognizing and analyzing human speech, music, and environmental sounds. However, their potential for understanding other types of sounds, particularly biomedical sounds, remains largely underexplored despite significant scientific interest. In this study, we focus on diagnosing cardiovascular diseases using phonocardiograms, i.e., heart sounds. Most existing deep neural network (DNN) paradigms are restricted to heart murmur classification (healthy vs unhealthy) and do not predict other acoustic features of the murmur such as grading, harshness, pitch, and quality, which are important in helping physicians diagnose the underlying heart conditions. We propose to finetune an audio LLM, Qwen2-Audio, on the PhysioNet CirCor DigiScope phonocardiogram (PCG) dataset and evaluate its performance in classifying 11 expert-labeled features. Additionally, we aim to achieve more noise-robust and generalizable system by exploring a preprocessing segmentation algorithm using an audio representation model, SSAMBA. Our results indicate that the LLM-based model outperforms state-of-the-art methods in 10 of the 11 tasks. Moreover, the LLM successfully classifies long-tail features with limited training data, a task that all previous methods have failed to classify. These findings underscore the potential of audio LLMs as assistants to human cardiologists in enhancing heart disease diagnosis.</div></div>","PeriodicalId":37151,"journal":{"name":"Smart Health","volume":"36 ","pages":"Article 100557"},"PeriodicalIF":0.0,"publicationDate":"2025-03-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143714430","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

An adaptive multimodal fusion framework for smartphone-based medication adherence monitoring of Parkinson’s disease 基于智能手机的帕金森病药物依从性监测的自适应多模式融合框架

Q2 Health Professions

Smart Health

Pub Date : 2025-03-25 DOI: 10.1016/j.smhl.2025.100561

Chongxin Zhong , Jinyuan Jia , Huining Li

Ensuring medication adherence for Parkinson’s disease (PD) patients is crucial to relieve patients’ symptoms and better customizing regimens according to patient’s clinical responses. However, traditional self-management approaches are often error-prone and have limited effectiveness in improving adherence. While smartphone-based solutions have been introduced to monitor various PD metrics, including medication adherence, these methods often rely on single-modality data or fail to fully leverage the advantages of multimodal integration. To address the issues, we present an adaptive multimodal fusion framework for monitoring medication adherence of PD based on a smartphone. Specifically, we segment and transform raw data from sensors to spectrograms. Then, we integrate multimodal data with quantification of their qualities and perform gradient modulation based on the contribution of each modality. Afterward, we monitor medication adherence in PD patients by detecting their medicine intake status. We evaluate the performance with the dataset from daily-life scenarios involving 455 patients. The results show that our work can achieve around 94% accuracy in medication adherence monitoring, indicating that our proposed framework is a promising tool to facilitate medication adherence monitoring in PD patients’ daily lives.

确保帕金森病（PD）患者的药物依从性对于缓解患者症状和根据患者的临床反应更好地定制治疗方案至关重要。然而，传统的自我管理方法往往容易出错，并且在提高依从性方面效果有限。虽然已经引入了基于智能手机的解决方案来监测各种PD指标，包括药物依从性，但这些方法通常依赖于单模态数据，或者不能充分利用多模态集成的优势。为了解决这些问题，我们提出了一种基于智能手机监测PD药物依从性的自适应多模式融合框架。具体来说，我们将原始数据从传感器分割并转换为频谱图。然后，我们对多模态数据进行整合，量化其质量，并根据每个模态的贡献进行梯度调制。之后，我们通过检测PD患者的药物摄入状况来监测他们的药物依从性。我们使用涉及455名患者的日常生活场景数据集来评估性能。结果表明，我们的工作在药物依从性监测方面的准确率可以达到94%左右，表明我们提出的框架是一个很有前途的工具，可以促进PD患者日常生活中的药物依从性监测。

{"title":"An adaptive multimodal fusion framework for smartphone-based medication adherence monitoring of Parkinson’s disease","authors":"Chongxin Zhong , Jinyuan Jia , Huining Li","doi":"10.1016/j.smhl.2025.100561","DOIUrl":"10.1016/j.smhl.2025.100561","url":null,"abstract":"<div><div>Ensuring medication adherence for Parkinson’s disease (PD) patients is crucial to relieve patients’ symptoms and better customizing regimens according to patient’s clinical responses. However, traditional self-management approaches are often error-prone and have limited effectiveness in improving adherence. While smartphone-based solutions have been introduced to monitor various PD metrics, including medication adherence, these methods often rely on single-modality data or fail to fully leverage the advantages of multimodal integration. To address the issues, we present an adaptive multimodal fusion framework for monitoring medication adherence of PD based on a smartphone. Specifically, we segment and transform raw data from sensors to spectrograms. Then, we integrate multimodal data with quantification of their qualities and perform gradient modulation based on the contribution of each modality. Afterward, we monitor medication adherence in PD patients by detecting their medicine intake status. We evaluate the performance with the dataset from daily-life scenarios involving 455 patients. The results show that our work can achieve around 94% accuracy in medication adherence monitoring, indicating that our proposed framework is a promising tool to facilitate medication adherence monitoring in PD patients’ daily lives.</div></div>","PeriodicalId":37151,"journal":{"name":"Smart Health","volume":"36 ","pages":"Article 100561"},"PeriodicalIF":0.0,"publicationDate":"2025-03-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143704184","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

Transforming stroop task cognitive assessments with multimodal inverse reinforcement learning 用多模态逆强化学习转化部队任务认知评估

Q2 Health Professions

Smart Health

Pub Date : 2025-03-25 DOI: 10.1016/j.smhl.2025.100567

Ali Abbasi , Jiaqi Gong , Soroush Korivand

Stroop tasks, recognized for their cognitively demanding nature, hold promise for diagnosing and monitoring neurodegenerative diseases. Understanding how humans allocate attention and resolve interference in the Stroop test remains a challenge; yet addressing this gap could reveal key opportunities for early-stage detection. Traditional approaches overlook the interplay between overt behavior and underlying neural processes, limiting insights into the complex color-word associations at play. To tackle this, we propose a framework that applies Inverse Reinforcement Learning (IRL) to fuse electroencephalography (EEG) signals with eye-tracking data, bridging the gap between neural and behavioral markers of cognition. We designed a Stroop experiment featuring congruent and incongruent conditions to evaluate attention allocation under varying levels of interference. By framing gaze as actions guided by an internally derived reward, IRL uncovers hidden motivations behind scanning patterns, while EEG data — processed with advanced feature extraction — reveals task-specific neural dynamics under high conflict. We validate our approach by measuring Probability Mismatch, Target Fixation Probability-Area Under the Curve, Sequence Score, and MultiMatch metrics. Results show that the IRL-EEG model outperforms an IRL-Image baseline, demonstrating improved alignment with human scanpaths and heightened sensitivity to attentional shifts in incongruent trials. These findings highlight the value of integrating neural data into computational models of cognition and illuminate possibilities for early detection of neurodegenerative disorders, where subclinical deficits may first emerge. Our IRL-based integration of EEG and eye-tracking further supports personalized cognitive assessments and adaptive user interfaces.

Stroop任务因其对认知的要求而被认可，有望用于诊断和监测神经退行性疾病。理解人类如何在Stroop测试中分配注意力和解决干扰仍然是一个挑战；然而，解决这一差距可能会为早期检测提供关键机会。传统的方法忽略了显性行为和潜在神经过程之间的相互作用，限制了对复杂的颜色-单词关联的了解。为了解决这个问题，我们提出了一个应用逆强化学习（IRL）的框架，将脑电图（EEG）信号与眼动追踪数据融合在一起，弥合认知的神经和行为标记之间的差距。我们设计了一个具有一致和不一致条件的Stroop实验来评估不同干扰水平下的注意分配。通过将凝视视为由内部衍生奖励引导的行为，IRL揭示了扫描模式背后隐藏的动机，而经过高级特征提取处理的EEG数据揭示了高冲突下特定任务的神经动力学。我们通过测量概率失配、目标固定概率-曲线下面积、序列得分和多匹配指标来验证我们的方法。结果表明，IRL-EEG模型优于IRL-Image基线，显示出与人类扫描路径的一致性改善，并且在不一致试验中对注意力转移的敏感性提高。这些发现突出了将神经数据整合到认知计算模型中的价值，并阐明了早期检测神经退行性疾病的可能性，其中亚临床缺陷可能首先出现。我们基于irl的EEG和眼动追踪集成进一步支持个性化认知评估和自适应用户界面。

{"title":"Transforming stroop task cognitive assessments with multimodal inverse reinforcement learning","authors":"Ali Abbasi , Jiaqi Gong , Soroush Korivand","doi":"10.1016/j.smhl.2025.100567","DOIUrl":"10.1016/j.smhl.2025.100567","url":null,"abstract":"<div><div>Stroop tasks, recognized for their cognitively demanding nature, hold promise for diagnosing and monitoring neurodegenerative diseases. Understanding how humans allocate attention and resolve interference in the Stroop test remains a challenge; yet addressing this gap could reveal key opportunities for early-stage detection. Traditional approaches overlook the interplay between overt behavior and underlying neural processes, limiting insights into the complex color-word associations at play. To tackle this, we propose a framework that applies Inverse Reinforcement Learning (IRL) to fuse electroencephalography (EEG) signals with eye-tracking data, bridging the gap between neural and behavioral markers of cognition. We designed a Stroop experiment featuring congruent and incongruent conditions to evaluate attention allocation under varying levels of interference. By framing gaze as actions guided by an internally derived reward, IRL uncovers hidden motivations behind scanning patterns, while EEG data — processed with advanced feature extraction — reveals task-specific neural dynamics under high conflict. We validate our approach by measuring Probability Mismatch, Target Fixation Probability-Area Under the Curve, Sequence Score, and MultiMatch metrics. Results show that the IRL-EEG model outperforms an IRL-Image baseline, demonstrating improved alignment with human scanpaths and heightened sensitivity to attentional shifts in incongruent trials. These findings highlight the value of integrating neural data into computational models of cognition and illuminate possibilities for early detection of neurodegenerative disorders, where subclinical deficits may first emerge. Our IRL-based integration of EEG and eye-tracking further supports personalized cognitive assessments and adaptive user interfaces.</div></div>","PeriodicalId":37151,"journal":{"name":"Smart Health","volume":"36 ","pages":"Article 100567"},"PeriodicalIF":0.0,"publicationDate":"2025-03-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143725638","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

Improving gastric lesion detection with synthetic images from diffusion models 利用扩散模型合成图像改进胃病变检测

Q2 Health Professions

Smart Health

Pub Date : 2025-03-25 DOI: 10.1016/j.smhl.2025.100569

Yanhua Si , Yingyun Yang , Qilei Chen , Zinan Xiong , Yu Cao , Xinwen Fu , Benyuan Liu , Aiming Yang

In the application of deep learning for gastric cancer detection, the quality of the data set is as important as, if not more, the design of the network architecture. However, obtaining labeled data, especially in fields such as medical imaging to detect gastric cancer, can be expensive and challenging. This scarcity is exacerbated by stringent privacy regulations and the need for annotations by specialists. Conventional methods of data augmentation fall short due to the complexities of medical imagery. In this paper, we explore the use of diffusion models to generate synthetic medical images for the detection of gastric cancer. We evaluate their capability to produce realistic images that can augment small datasets, potentially enhancing the accuracy and robustness of detection algorithms. By training diffusion models on existing gastric cancer data and producing new images, our aim is to expand these datasets, thereby enhancing the efficiency of deep learning model training to achieve better precision and generalization in lesion detection. Our findings indicate that images generated by diffusion models significantly mitigate the issue of data scarcity, advancing the field of deep learning in medical imaging.

在将深度学习应用于胃癌检测中，数据集的质量与网络架构的设计同样重要，甚至更重要。然而，获得标记数据，特别是在医学成像检测胃癌等领域，可能是昂贵和具有挑战性的。严格的隐私法规和对专家注释的需求加剧了这种稀缺性。由于医学图像的复杂性，传统的数据增强方法存在不足。在本文中，我们探索使用扩散模型来生成用于胃癌检测的合成医学图像。我们评估了它们产生真实图像的能力，这些图像可以增强小数据集，潜在地提高检测算法的准确性和鲁棒性。通过在现有胃癌数据上训练扩散模型并生成新的图像，我们的目标是扩展这些数据集，从而提高深度学习模型训练的效率，以达到更好的病变检测精度和泛化。我们的研究结果表明，扩散模型生成的图像显著缓解了数据稀缺的问题，推动了医学成像领域的深度学习。

{"title":"Improving gastric lesion detection with synthetic images from diffusion models","authors":"Yanhua Si , Yingyun Yang , Qilei Chen , Zinan Xiong , Yu Cao , Xinwen Fu , Benyuan Liu , Aiming Yang","doi":"10.1016/j.smhl.2025.100569","DOIUrl":"10.1016/j.smhl.2025.100569","url":null,"abstract":"<div><div>In the application of deep learning for gastric cancer detection, the quality of the data set is as important as, if not more, the design of the network architecture. However, obtaining labeled data, especially in fields such as medical imaging to detect gastric cancer, can be expensive and challenging. This scarcity is exacerbated by stringent privacy regulations and the need for annotations by specialists. Conventional methods of data augmentation fall short due to the complexities of medical imagery. In this paper, we explore the use of diffusion models to generate synthetic medical images for the detection of gastric cancer. We evaluate their capability to produce realistic images that can augment small datasets, potentially enhancing the accuracy and robustness of detection algorithms. By training diffusion models on existing gastric cancer data and producing new images, our aim is to expand these datasets, thereby enhancing the efficiency of deep learning model training to achieve better precision and generalization in lesion detection. Our findings indicate that images generated by diffusion models significantly mitigate the issue of data scarcity, advancing the field of deep learning in medical imaging.</div></div>","PeriodicalId":37151,"journal":{"name":"Smart Health","volume":"36 ","pages":"Article 100569"},"PeriodicalIF":0.0,"publicationDate":"2025-03-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143738102","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

Continuous prediction of user dropout in a mobile mental health intervention program: An exploratory machine learning approach 移动心理健康干预项目中用户退出的连续预测：一种探索性机器学习方法

Q2 Health Professions

Smart Health

Pub Date : 2025-03-25 DOI: 10.1016/j.smhl.2025.100565

Pinxiang Wang , Hanqi Chen , Zhouyu Li , Wenyao Xu , Yu-Ping Chang , Huining Li

Mental health intervention can help to release individuals’ mental symptoms like anxiety and depression. A typical mental health intervention program can last for several months, people may lose interests along with the time and cannot insist till the end. Accurately predicting user dropout is crucial for delivering timely measures to address user disengagement and reduce its adverse effects on treatment. We develop a temporal deep learning approach to accurately predict dropout, leveraging advanced data augmentation and feature engineering techniques. By integrating interaction metrics from user behavior logs and semantic features from user self-reflections over a nine-week intervention program, our approach effectively characterizes user’s mental health intervention behavior patterns. The results validate the efficacy of temporal models for continuous dropout prediction.

心理健康干预可以帮助缓解个人的心理症状，如焦虑和抑郁。一个典型的心理健康干预项目可以持续几个月，人们可能会随着时间的推移而失去兴趣，无法坚持到最后。准确预测用户退出对于及时采取措施解决用户脱离接触问题并减少其对治疗的不利影响至关重要。我们开发了一种时间深度学习方法来准确预测辍学，利用先进的数据增强和特征工程技术。通过整合来自用户行为日志的交互指标和来自用户自我反思的语义特征，我们的方法有效地描述了用户心理健康干预行为模式。结果验证了时间模型对连续辍学预测的有效性。

引用次数: 0

HealthQ: Unveiling questioning capabilities of LLM chains in healthcare conversations HealthQ：揭示法学硕士链在医疗保健对话中的提问能力

Q2 Health Professions

Smart Health

Pub Date : 2025-03-25 DOI: 10.1016/j.smhl.2025.100570

Ziyu Wang , Hao Li , Di Huang , Hye-Sung Kim , Chae-Won Shin , Amir M. Rahmani

Effective patient care in digital healthcare requires large language models (LLMs) that not only answer questions but also actively gather critical information through well-crafted inquiries. This paper introduces HealthQ, a novel framework for evaluating the questioning capabilities of LLM healthcare chains. By implementing advanced LLM chains, including Retrieval-Augmented Generation (RAG), Chain of Thought (CoT), and reflective chains, HealthQ assesses how effectively these chains elicit comprehensive and relevant patient information. To achieve this, we integrate an LLM judge to evaluate generated questions across metrics such as specificity, relevance, and usefulness, while aligning these evaluations with traditional Natural Language Processing (NLP) metrics like ROUGE and Named Entity Recognition (NER)-based set comparisons. We validate HealthQ using two custom datasets constructed from public medical datasets, ChatDoctor and MTS-Dialog, and demonstrate its robustness across multiple LLM judge models, including GPT-3.5, GPT-4, and Claude. Our contributions are threefold: we present the first systematic framework for assessing questioning capabilities in healthcare conversations, establish a model-agnostic evaluation methodology, and provide empirical evidence linking high-quality questions to improved patient information elicitation.

在数字医疗保健中，有效的患者护理需要大型语言模型（llm），不仅要回答问题，还要通过精心设计的查询积极收集关键信息。本文介绍了HealthQ，一个用于评估法学硕士医疗保健链的提问能力的新框架。通过实施先进的LLM链，包括检索增强生成（RAG）、思维链（CoT）和反射链，HealthQ评估这些链如何有效地获取全面和相关的患者信息。为了实现这一目标，我们集成了一个法学硕士判断器，通过特异性、相关性和有用性等指标来评估生成的问题，同时将这些评估与传统的自然语言处理（NLP）指标（如ROUGE和基于命名实体识别（NER）的集合比较）保持一致。我们使用从公共医疗数据集ChatDoctor和MTS-Dialog构建的两个自定义数据集验证HealthQ，并演示其在多个LLM判断模型（包括GPT-3.5、GPT-4和Claude）中的鲁棒性。我们的贡献有三个方面：我们提出了第一个用于评估医疗保健对话中的提问能力的系统框架，建立了一个模型不可知论的评估方法，并提供了将高质量问题与改进的患者信息获取联系起来的经验证据。

{"title":"HealthQ: Unveiling questioning capabilities of LLM chains in healthcare conversations","authors":"Ziyu Wang , Hao Li , Di Huang , Hye-Sung Kim , Chae-Won Shin , Amir M. Rahmani","doi":"10.1016/j.smhl.2025.100570","DOIUrl":"10.1016/j.smhl.2025.100570","url":null,"abstract":"<div><div>Effective patient care in digital healthcare requires large language models (LLMs) that not only answer questions but also actively gather critical information through well-crafted inquiries. This paper introduces HealthQ, a novel framework for evaluating the questioning capabilities of LLM healthcare chains. By implementing advanced LLM chains, including Retrieval-Augmented Generation (RAG), Chain of Thought (CoT), and reflective chains, HealthQ assesses how effectively these chains elicit comprehensive and relevant patient information. To achieve this, we integrate an LLM judge to evaluate generated questions across metrics such as specificity, relevance, and usefulness, while aligning these evaluations with traditional Natural Language Processing (NLP) metrics like ROUGE and Named Entity Recognition (NER)-based set comparisons. We validate HealthQ using two custom datasets constructed from public medical datasets, ChatDoctor and MTS-Dialog, and demonstrate its robustness across multiple LLM judge models, including GPT-3.5, GPT-4, and Claude. Our contributions are threefold: we present the first systematic framework for assessing questioning capabilities in healthcare conversations, establish a model-agnostic evaluation methodology, and provide empirical evidence linking high-quality questions to improved patient information elicitation.</div></div>","PeriodicalId":37151,"journal":{"name":"Smart Health","volume":"36 ","pages":"Article 100570"},"PeriodicalIF":0.0,"publicationDate":"2025-03-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143704085","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