Smart Health最新文献

{"title":"Privacy-preserving LLM-based chatbots for hypertensive patient self-management","authors":"Sara Montagna ,&nbsp;Stefano Ferretti ,&nbsp;Lorenz Cuno Klopfenstein ,&nbsp;Michelangelo Ungolo ,&nbsp;Martino Francesco Pengo ,&nbsp;Gianluca Aguzzi ,&nbsp;Matteo Magnini","doi":"10.1016/j.smhl.2025.100552","DOIUrl":"10.1016/j.smhl.2025.100552","url":null,"abstract":"<div><div>Medical chatbots are becoming a basic component in telemedicine, propelled by advancements in Large Language Models (LLMs). However, LLMs’ integration into clinical settings comes with several issues, with privacy concerns being particularly significant.</div><div>The paper proposes a tailored architectural solution and an information workflow that address privacy issues, while preserving the benefits of LLMs. We examine two solutions to prevent the disclosure of sensitive information: <em>(i)</em> a filtering mechanism that processes sensitive data locally but leverage a robust OpenAI’s online LLM for engaging with the user effectively, and <em>(ii)</em> a fully local deployment of open-source LLMs. The effectiveness of these solutions is assessed in the context of hypertension management across various tasks, ranging from intent recognition to reliable and emphatic conversation. Interestingly, while the first solution proved to be more robust in intent recognition, an evaluation by domain experts of the models’ responses, based on reliability and empathetic principles, revealed that two out of six open LLMs received the highest scores.</div><div>The study underscores the viability of incorporating LLMs into medical chatbots. In particular, our findings suggest that open LLMs can offer a privacy-preserving, yet promising, alternative to external LLM services, ensuring safer and more reliable telemedicine practices. Future efforts will focus on fine-tuning local models to enhance their performance across all tasks.</div></div>","PeriodicalId":37151,"journal":{"name":"Smart Health","volume":"36 ","pages":"Article 100552"},"PeriodicalIF":0.0,"publicationDate":"2025-03-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143601759","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}

{"title":"Automated supplement correction in passive urine color measurement device for real-time hydration testing","authors":"Kelvin M. Frazier,&nbsp;Brian F. Bender","doi":"10.1016/j.smhl.2025.100554","DOIUrl":"10.1016/j.smhl.2025.100554","url":null,"abstract":"<div><div>Monitoring urine color as a means of assessing hydration status has long been a recommended technique for active populations like athletes, military personnel, and outdoor laborers. Urine color correlates well to urine concentration and is a simple, non-invasive practice. However, currently this approach is subjective, and errors arise in variation in ambient lighting conditions, comparator colors used, and individual perception. In addition, certain supplements such as riboflavin (vitamin B2) and beet juice are known to distort urine color and can confound hydration assessment. An automated urinalysis device (InFlow) was developed to measure urine color, an index of hydration status, in real-time during urination in the presence of these supplements. Machine learning techniques were used to reduce mean absolute hydration assessment error from riboflavin-derived color skew from 2.50 ± 0.37 to 0.85 (±0.06) color units on a 7-point color chart scale compared to a commercial colorimeter. In the absence of supplements and in the samples spiked with beet juice the InFlow device produced a mean absolute error of 0.48 (±0.06) color units. Finally, we demonstrate the feasibility of detecting myoglobinuria for potential future use in rhabdomyolysis screening. Our results show the InFlow device provides a novel approach with appropriate accuracy for standardizing hydration assessment via urinalysis in environments with high testing frequency demands in the presence of common urine color interferents including riboflavin and beet juice.</div></div>","PeriodicalId":37151,"journal":{"name":"Smart Health","volume":"36 ","pages":"Article 100554"},"PeriodicalIF":0.0,"publicationDate":"2025-03-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143579868","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}

{"title":"Privacy-preserving polyp segmentation using federated learning with differential privacy","authors":"Md. Mahmodul Hasan ,&nbsp;Mohammad Motiur Rahman","doi":"10.1016/j.smhl.2025.100551","DOIUrl":"10.1016/j.smhl.2025.100551","url":null,"abstract":"<div><h3>Background and Objective:</h3><div>Patient privacy is of paramount importance in the medical field, especially as data-driven medical applications gain popularity. The privacy of medical records is increasingly crucial. In this context, data-oriented polyp (a precancerous stage of colon cancer) segmentation is a critical area of ongoing research, aiming to improve automated segmentation. Accurate segmentation is essential for the complete removal of these overgrown cells from the gastrointestinal system. Although large data sets using data-driven algorithms have shown excellent performance in image segmentation, privacy concerns have limited the availability of such datasets for medical image segmentation tasks, including polyp segmentation. This research aims to develop an approach for polyp segmentation that combines data from multiple sources without compromising patient privacy.</div></div><div><h3>Methods:</h3><div>We design a differentially private federated learning system to segment polyps without compromising privacy. Our approach employs the encoder–decoder architecture UNet 3+ with a deep supervision technique to achieve effective segmentation of polyps in a federated setup. The federated training process aims to find generalized global models for the entities participating in the federation. The study uses four public databases to train and evaluate the proposed method.</div></div><div><h3>Results:</h3><div>The proposed privacy-protected technique demonstrates promising outcomes in polyp segmentation, achieving an average Intersection over Union (IoU) score of 0.90881 ± 0.00355 over four publicly available datasets. Evaluation metrics include precision, sensitivity, and specificity values, indicating the effectiveness of our approach in accurately segmenting polyps.</div></div><div><h3>Conclusions:</h3><div>Our differentially private federated learning system successfully segments polyps without compromising patient privacy. The promising results suggest that this approach can significantly contribute to the field of polyp segmentation, facilitating the use of large datasets while maintaining strict privacy standards.</div></div>","PeriodicalId":37151,"journal":{"name":"Smart Health","volume":"36 ","pages":"Article 100551"},"PeriodicalIF":0.0,"publicationDate":"2025-02-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143551865","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}

{"title":"PulseSight : A novel method for contactless oxygen saturation (SpO2) monitoring using smartphone cameras, remote photoplethysmography and machine learning","authors":"Kazi Zawad Arefin ,&nbsp;Kazi Shafiul Alam ,&nbsp;Sayed Mashroor Mamun ,&nbsp;Nafi Us Sabbir Sabith ,&nbsp;Masud Rabbani ,&nbsp;Parama Sridevi ,&nbsp;Sheikh Iqbal Ahamed","doi":"10.1016/j.smhl.2025.100542","DOIUrl":"10.1016/j.smhl.2025.100542","url":null,"abstract":"<div><div>Monitoring oxygen saturation (SpO<span><math><msub><mrow></mrow><mrow><mn>2</mn></mrow></msub></math></span>) level is crucial for evaluating the current cardiac and respiratory condition of a person, particularly in medical settings. Conventional pulse oximetry, while efficient, has drawbacks such as the requirement for physical touch and vulnerability to certain environmental influences. In this paper, we propose an innovative approach for estimating SpO<span><math><msub><mrow></mrow><mrow><mn>2</mn></mrow></msub></math></span> levels utilizing smartphone cameras and video-based photoplethysmography (PPG) without physical touch. Our framework consists of an Android mobile application that records 20-second face videos, which a cloud-based backend server then analyzes. The server utilizes deep learning-based facial recognition and signal processing techniques to extract remote photoplethysmography (rPPG) signals from specific facial regions and predict oxygen saturation (SpO<span><math><msub><mrow></mrow><mrow><mn>2</mn></mrow></msub></math></span>) levels using a Support Vector Regression (SVR) Machine learning model. Signal noise and motion artifacts are mitigated by extracting relevant characteristics from the rPPG. The system was validated by experimental studies, which contained 40 sets of videos collected from 10 participants. The study was conducted under different illumination conditions, which showed low RMSE score (1.45 ±0.1) and MAE score (0.92 ±0.01). Also, our system shows high usability, as indicated by the System Usability Scale (SUS) score of 80.5. The results demonstrate that our method offers a dependable and contactless substitute for continuous SpO2 monitoring, with potential uses in telemedicine and remote patient monitoring.</div></div>","PeriodicalId":37151,"journal":{"name":"Smart Health","volume":"36 ","pages":"Article 100542"},"PeriodicalIF":0.0,"publicationDate":"2025-02-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143395171","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}

{"title":"Smart health practices: Strategies to improve healthcare efficiency through digital twin technology","authors":"Md. Armanul Hasan ,&nbsp;Ridwan Mustofa ,&nbsp;Niamat Ullah Ibne Hossain ,&nbsp;Md. Saiful Islam","doi":"10.1016/j.smhl.2025.100541","DOIUrl":"10.1016/j.smhl.2025.100541","url":null,"abstract":"<div><div>A digital twin (DT) is a virtual representation of a real-world object that has dynamic, bidirectional connections between the real-world object and its digital domain. With the advent of Industry 4.0, DT technology was initially applied in the engineering and manufacturing sectors, but recent research indicates DT may also be useful within the healthcare sector. The purpose of this study was to determine the potential applications of DT technology in the healthcare sector and offer suggestions for its effective implementation by healthcare institutions to increase service efficiency. Based on a review of the literature, we developed a model to demonstrate the applications of DTs on public and personal health. A questionnaire with five points Likert scale was then designed based on this model. Data were collected through an online survey conducted with 306 participants. To verify our hypothesized correlations among the constructs, structural equation modeling was used. The findings suggested that explainable artificial intelligence-based early diagnosis, simulation model-based vaccination, artificial intelligence location technology, sensor-based real-time health monitoring, and in silico personalized medicine are potential applications of DT that can increase healthcare efficiency. We also considered the moderating influence of (a) security and privacy and (b) certification and regulatory issues, acknowledging their pivotal roles in ensuring the successful implementation and widespread acceptance of DT technology in the field of healthcare. This study contributes to the body of knowledge in academia and offers useful insights for technologists, policymakers, and healthcare professionals who want to fully utilize DT technology to build an effective healthcare system that can adapt to the changing needs of communities and individuals.</div></div>","PeriodicalId":37151,"journal":{"name":"Smart Health","volume":"36 ","pages":"Article 100541"},"PeriodicalIF":0.0,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143372960","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}

{"title":"Human knowledge-based artificial intelligence methods for skin cancer management: Accuracy and interpretability study","authors":"Eman Rezk ,&nbsp;Mohamed Eltorki ,&nbsp;Wael El-Dakhakhni","doi":"10.1016/j.smhl.2025.100540","DOIUrl":"10.1016/j.smhl.2025.100540","url":null,"abstract":"<div><div>Skin cancer management, including monitoring and excision, involves sophisticated decisions reliant on several interdependent factors. This complexity leads to a scarcity of data useful for skin cancer management. Deep learning achieved massive success in computer vision due to its ability to extract representative features from images. However, deep learning methods require large amounts of data to develop accurate models, whereas machine learning methods perform well with small datasets. In this work, we aim to compare the accuracy and interpretability of skin cancer management prediction 1) using deep learning and machine learning methods and 2) utilizing various inputs including clinical images, dermoscopic images, and lesion clinical tabular features created by experts to represent lesion characteristics. We implemented two approaches, a deep learning pipeline for feature extraction and classification trained on different input modalities including images and lesion clinical features. The second approach uses lesion clinical features to train machine learning classifiers. The results show that the machine learning approach trained on clinical features achieves higher accuracy (0.80) and higher area under the curve (0.92) compared to the deep learning pipeline trained on skin images and lesion clinical features which achieves an accuracy of 0.66 and area under the curve of 0.74. Additionally, the machine learning approach provides more informative and understandable interpretations of the results. This work emphasizes the significance of utilizing human knowledge in developing precise and transparent predictive models. In addition, our findings highlight the potential of machine learning methods in predicting lesion management in situation where the data size is insufficient to leverage deep learning capabilities.</div></div>","PeriodicalId":37151,"journal":{"name":"Smart Health","volume":"36 ","pages":"Article 100540"},"PeriodicalIF":0.0,"publicationDate":"2025-01-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143100542","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}

{"title":"SAFE: Sound Analysis for Fall Event detection using machine learning","authors":"Antony Garcia ,&nbsp;Xinming Huang","doi":"10.1016/j.smhl.2024.100539","DOIUrl":"10.1016/j.smhl.2024.100539","url":null,"abstract":"<div><div>This study evaluates the application of machine learning (ML) and deep learning (DL) algorithms for fall detection using sound signals. The work is supported by the Sound Analysis for Fall Events (SAFE) dataset, comprising 950 audio samples, including 475 fall events recorded with a grappling dummy to simulate realistic scenarios. Decision tree-based ML algorithms achieved a classification accuracy of 93% at lower sampling rates, indicating that critical features are preserved despite reduced resolution. DL models, using spectrogram-based feature extraction, reached accuracies up to 99%, surpassing traditional ML methods in performance. Linear models also achieved high accuracy (up to 97%) in various spectrogram techniques, emphasizing the separability of audio features. These results establish the viability of sound-based fall detection systems as efficient and accurate solutions.</div></div>","PeriodicalId":37151,"journal":{"name":"Smart Health","volume":"35 ","pages":"Article 100539"},"PeriodicalIF":0.0,"publicationDate":"2025-01-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143129214","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}

{"title":"Latent Space Representation of Adversarial AutoEncoder for Human Activity Recognition: Application to a low-cost commercial force plate and inertial measurement units","authors":"Kenta Kamikokuryo ,&nbsp;Gentiane Venture ,&nbsp;Vincent Hernandez","doi":"10.1016/j.smhl.2024.100537","DOIUrl":"10.1016/j.smhl.2024.100537","url":null,"abstract":"<div><div>Human Activity Recognition (HAR) is a key component of a home rehabilitation system that provides real-time monitoring and personalized feedback. This research explores the application of Adversarial AutoEncoder (AAE) models for data dimensionality reduction in the context of HAR. Visualizing data in a lower-dimensional space is important to understand changes in motor control due to medical conditions or aging, to aid personalized interventions, and to ensure continuous benefits in remote rehabilitation settings. This makes patient assessment effective, easier, and faster.</div><div>In this study, the classification performance of the latent space created by the AAE is evaluated using the Wii Balance Board (WiiBB) and/or three Inertial Measurement Units (IMUs) placed on the forearms and hip. Various sensor configurations are considered, including only WiiBB, only IMUs, combinations of WiiBB with the IMU at the hip, and combinations of WiiBB with the 3 IMUs.</div><div>The accuracy of the latent space representation is compared with two common supervised classification models, which are the Convolutional Neural Network (CNN) and the neural network called CNNLSTM, which is composed of convolution layers followed by recurrent layers. The approach was demonstrated for two different sets of exercises consisting of upper and lower body exercises collected with 19 participants.</div><div>The results show that the latent space representation of the AAE achieves a strong classification accuracy performance while also serving as a visualization tool. This study is an initial demonstration of the potential of integrating WiiBB and IMU sensors for comprehensive activity recognition for upper and lower body movement analysis.</div></div>","PeriodicalId":37151,"journal":{"name":"Smart Health","volume":"35 ","pages":"Article 100537"},"PeriodicalIF":0.0,"publicationDate":"2025-01-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143129614","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}

{"title":"Novel EEG feature selection based on hellinger distance for epileptic seizure detection","authors":"Muhammed Sadiq ,&nbsp;Mustafa Noaman Kadhim ,&nbsp;Dhiah Al-Shammary ,&nbsp;Mariofanna Milanova","doi":"10.1016/j.smhl.2024.100536","DOIUrl":"10.1016/j.smhl.2024.100536","url":null,"abstract":"<div><div>This study introduces a novel feature selection method based on Hellinger distance and particle swarm optimization (PSO) for reducing the dimensionality of features in electroencephalogram (EEG) signals and improving epileptic seizure detection accuracy. In the first phase, the Hellinger distance is used as a filter to remove redundant and irrelevant features by calculating the similarity between blocks within the feature, thus reducing the search space for the subsequent second phase. In the second phase, PSO searches the reduced feature space to select the best subset. Recognizing that both classification accuracy and dimensionality play crucial roles in the performance of feature subsets, PSO searches various sets of features (ranging from 410 to 2867 in EEG signals) derived from the first stage using Hellinger distance, rather than searching through the full set of 4047 features, to select the optimal subset. The proposed Hellinger-PSO approach demonstrates significant improvements in classification accuracy across multiple models. Specifically, Logistic Regression (LR) improved from 91% to 95% (4% improvement), Decision Tree (DT) from 95% to 97% (2% improvement), Naive Bayes (NB) from 94% to 99% (5% improvement), and Random Forest (RF) from 96% to 98% (2% improvement) on the Bonn dataset. Additionally, the method reduces dimensionality while maintaining high classification performance. The results validate the efficacy of the Hellinger-PSO technique, which enhances both the accuracy and efficiency of epileptic seizure detection. This approach has the potential to improve diagnostic accuracy in medical settings, aiding in better patient care and more effective clinical decision-making.</div></div>","PeriodicalId":37151,"journal":{"name":"Smart Health","volume":"35 ","pages":"Article 100536"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143129218","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}

{"title":"Explainable screening of oral cancer via deep learning and case-based reasoning","authors":"Mario G.C.A. Cimino ,&nbsp;Giuseppina Campisi ,&nbsp;Federico A. Galatolo ,&nbsp;Paolo Neri ,&nbsp;Pietro Tozzo ,&nbsp;Marco Parola ,&nbsp;Gaetano La Mantia ,&nbsp;Olga Di Fede","doi":"10.1016/j.smhl.2024.100538","DOIUrl":"10.1016/j.smhl.2024.100538","url":null,"abstract":"<div><div>Oral Squamous Cell Carcinoma is characterized by significant mortality and morbidity. Dental professionals can play an important role in its early detection, thanks to the availability of embedded smart cameras for oral photos and remote screening supported by Deep Learning (DL). Despite the promising results of DL for automated detection and classification of oral lesions, its effectiveness is based on a clearly defined protocol, on the explainability of results, and on periodic cases collection. This paper proposes a novel method, combining DL and Case-Based Reasoning (CBR), to allow the post-hoc explanation of the system answer. The method uses explainability tools organized in a protocol defined in the Business Process Model and Notation (BPMN) to allow its experimental validation. A redesign of the Faster-R-CNN Feature Pyramid Networks (FPN) + DL architecture is also proposed for lesions detection and classification, fine-tuned on 160 cases belonging to three classes of oral ulcers. The DL system achieves state-of-the-art performance, i.e., 83% detection and 92% classification rate (98% for neoplastic vs. non-neoplastic binary classification). A preliminary experimentation of the protocol involved both resident and specialized doctors over selected difficult cases. The system and cases have been publicly released to foster collaboration between research centers.</div></div>","PeriodicalId":37151,"journal":{"name":"Smart Health","volume":"35 ","pages":"Article 100538"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143129615","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}