Pub Date : 2025-02-24DOI: 10.1016/j.compbiomed.2025.109857
Guillermo Iglesias , Hector Menendez , Edgar Talavera
Explainability in Medical Computer Vision is one of the most sensible implementations of Artificial Intelligence nowadays in healthcare. In this work, we propose a novel Deep Learning architecture for eXplainable Artificial Intelligence, specially designed for medical diagnostic. The proposed approach leverages Variational Autoencoders properties to produce linear modifications of images in a lower-dimensional embedded space, and then reconstructs these modifications into non-linear explanations in the original image space. The proposed approach is based on global and local regularisation of the latent space, which stores visual and semantic information about images. Specifically, a multi-objective genetic algorithm is designed for searching explanations, finding individuals that can misclassify the classification output of the network while producing the minimum number of changes in the image descriptor. The genetic algorithm is able to search for explanations without defining any hyperparameters, and uses only one individual to provide a complete explanation of the whole image. Furthermore, the explanations found by the proposed approach are compared with state-of-the-art eXplainable Artificial Intelligence systems and the results show an improvement in the precision of the explanation between 56.39 and 7.23 percentage points.
{"title":"Improving explanations for medical X-ray diagnosis combining variational autoencoders and adversarial machine learning","authors":"Guillermo Iglesias , Hector Menendez , Edgar Talavera","doi":"10.1016/j.compbiomed.2025.109857","DOIUrl":"10.1016/j.compbiomed.2025.109857","url":null,"abstract":"<div><div>Explainability in Medical Computer Vision is one of the most sensible implementations of Artificial Intelligence nowadays in healthcare. In this work, we propose a novel Deep Learning architecture for eXplainable Artificial Intelligence, specially designed for medical diagnostic. The proposed approach leverages Variational Autoencoders properties to produce linear modifications of images in a lower-dimensional embedded space, and then reconstructs these modifications into non-linear explanations in the original image space. The proposed approach is based on global and local regularisation of the latent space, which stores visual and semantic information about images. Specifically, a multi-objective genetic algorithm is designed for searching explanations, finding individuals that can misclassify the classification output of the network while producing the minimum number of changes in the image descriptor. The genetic algorithm is able to search for explanations without defining any hyperparameters, and uses only one individual to provide a complete explanation of the whole image. Furthermore, the explanations found by the proposed approach are compared with state-of-the-art eXplainable Artificial Intelligence systems and the results show an improvement in the precision of the explanation between 56.39 and 7.23 percentage points.</div></div>","PeriodicalId":10578,"journal":{"name":"Computers in biology and medicine","volume":"188 ","pages":"Article 109857"},"PeriodicalIF":7.0,"publicationDate":"2025-02-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143479961","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The emergence of single-cell technologies, including flow and mass cytometry, as well as single-cell RNA sequencing, has revolutionized the study of cellular heterogeneity, generating vast datasets rich in biological insights. Despite the effectiveness of graph-based analyses in deciphering the complexities of these datasets, managing large-scale graph representations of single-cell data remains computationally challenging. Coarsening has been employed to tackle this difficulty. However, current coarsening techniques such as Cytocoarsening, Heavy Edge Matching (HEM), and Locally Variable Edges (LVE) often suffer from slow processing speeds and limited adaptability. To address these challenges, we propose a novel approach utilizing Feature-Aware Graph Coarsening via Hashing (FACH), which integrates locality-sensitive hashing for scalable and efficient single-cell data analysis. This method directly extracts informative, low-dimensional cell representations from raw single-cell RNA sequencing and mass cytometry data, significantly improving processing speed while preserving essential data features. We demonstrate its effectiveness in downstream tasks, such as scalable graph neural network training on coarsened single-cell data, highlighting its ability to retain crucial biological features and enable efficient, accurate analyses. Our method directly extracts informative, low-dimensional cell representations from raw single-cell RNA sequencing and mass cytometry data, significantly improving processing speed and preserving critical biological features, such as transcriptional signatures and network topology. It reduces computational time by at least 50% compared to existing methods and achieves superior classification accuracy, such as 88.1% on the Baron Human dataset, underscoring its efficiency and precision in large-scale single-cell analysis.
{"title":"A novel coarsened graph learning method for scalable single-cell data analysis","authors":"Mohit Kataria , Ekta Srivastava , Kumar Arjun , Sandeep Kumar , Ishaan Gupta , Jayadeva","doi":"10.1016/j.compbiomed.2025.109873","DOIUrl":"10.1016/j.compbiomed.2025.109873","url":null,"abstract":"<div><div>The emergence of single-cell technologies, including flow and mass cytometry, as well as single-cell RNA sequencing, has revolutionized the study of cellular heterogeneity, generating vast datasets rich in biological insights. Despite the effectiveness of graph-based analyses in deciphering the complexities of these datasets, managing large-scale graph representations of single-cell data remains computationally challenging. Coarsening has been employed to tackle this difficulty. However, current coarsening techniques such as Cytocoarsening, Heavy Edge Matching (HEM), and Locally Variable Edges (LVE) often suffer from slow processing speeds and limited adaptability. To address these challenges, we propose a novel approach utilizing Feature-Aware Graph Coarsening via Hashing (FACH), which integrates locality-sensitive hashing for scalable and efficient single-cell data analysis. This method directly extracts informative, low-dimensional cell representations from raw single-cell RNA sequencing and mass cytometry data, significantly improving processing speed while preserving essential data features. We demonstrate its effectiveness in downstream tasks, such as scalable graph neural network training on coarsened single-cell data, highlighting its ability to retain crucial biological features and enable efficient, accurate analyses. Our method directly extracts informative, low-dimensional cell representations from raw single-cell RNA sequencing and mass cytometry data, significantly improving processing speed and preserving critical biological features, such as transcriptional signatures and network topology. It reduces computational time by at least 50% compared to existing methods and achieves superior classification accuracy, such as 88.1% on the Baron Human dataset, underscoring its efficiency and precision in large-scale single-cell analysis.</div></div>","PeriodicalId":10578,"journal":{"name":"Computers in biology and medicine","volume":"188 ","pages":"Article 109873"},"PeriodicalIF":7.0,"publicationDate":"2025-02-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143474789","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-02-24DOI: 10.1016/j.compbiomed.2025.109835
Akeem Temitope Otapo , Alice Othmani , Ghazaleh Khodabandelou , Zuheng Ming
The integration of Artificial Intelligence (AI) with the Internet of Medical Things (IoMT) has revolutionized disease prediction and detection, but challenges such as data heterogeneity, privacy concerns, and model generalizability hinder its full potential in healthcare. This review examines these challenges and evaluates the effectiveness of AI-IoMT techniques in predicting chronic and terminal diseases, including cardiovascular conditions, Alzheimer’s disease, and cancers. We analyze a range of Machine Learning (ML) and Deep Learning (DL) approaches (e.g., XGBoost, Random Forest, CNN, LSTM), alongside advanced strategies like federated learning, transfer learning, and blockchain, to improve model robustness, data security, and interoperability. Findings highlight that transfer learning and ensemble methods enhance model adaptability across clinical settings, while blockchain and federated learning effectively address privacy and data standardization. Ultimately, the review emphasizes the importance of data harmonization, secure frameworks, and multi-disease models as critical research directions for scalable, comprehensive AI-IoMT solutions in healthcare.
{"title":"Prediction and detection of terminal diseases using Internet of Medical Things: A review","authors":"Akeem Temitope Otapo , Alice Othmani , Ghazaleh Khodabandelou , Zuheng Ming","doi":"10.1016/j.compbiomed.2025.109835","DOIUrl":"10.1016/j.compbiomed.2025.109835","url":null,"abstract":"<div><div>The integration of Artificial Intelligence (AI) with the Internet of Medical Things (IoMT) has revolutionized disease prediction and detection, but challenges such as data heterogeneity, privacy concerns, and model generalizability hinder its full potential in healthcare. This review examines these challenges and evaluates the effectiveness of AI-IoMT techniques in predicting chronic and terminal diseases, including cardiovascular conditions, Alzheimer’s disease, and cancers. We analyze a range of Machine Learning (ML) and Deep Learning (DL) approaches (e.g., XGBoost, Random Forest, CNN, LSTM), alongside advanced strategies like federated learning, transfer learning, and blockchain, to improve model robustness, data security, and interoperability. Findings highlight that transfer learning and ensemble methods enhance model adaptability across clinical settings, while blockchain and federated learning effectively address privacy and data standardization. Ultimately, the review emphasizes the importance of data harmonization, secure frameworks, and multi-disease models as critical research directions for scalable, comprehensive AI-IoMT solutions in healthcare.</div></div>","PeriodicalId":10578,"journal":{"name":"Computers in biology and medicine","volume":"188 ","pages":"Article 109835"},"PeriodicalIF":7.0,"publicationDate":"2025-02-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143474790","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-02-22DOI: 10.1016/j.compbiomed.2025.109815
Abraham Bautista-Castillo , Angela Chun , Tiphanie P. Vogel , Ioannis A. Kakadiaris
The COVID-19 pandemic brought several diagnostic challenges, including the post-infectious sequelae multisystem inflammatory syndrome in children (MIS-C). Some of the clinical features of this syndrome can be found in other pathologies such as Kawasaki disease, toxic shock syndrome, and endemic typhus. Endemic typhus, or murine typhus, is an acute infection treated much differently than MIS-C, so early detection is crucial to a favorable prognosis for patients with these disorders. Clinical Decision Support Systems (CDSS) are computer systems designed to support the decision-making of medical teams about their patients and intended to improve uprising clinical challenges in healthcare. In this article, we present a CDSS to distinguish between MIS-C and typhus, which includes a scoring system that allows the timely distinction of both pathologies using only clinical and laboratory features typically available within the first six hours of presentation to the Emergency Department. The proposed approach was trained and tested on datasets of 87 typhus patients and 133 MIS-C patients. A comparison was made against five well-known statistical and machine-learning models. A second dataset with 111 MIS-C patients was used to verify the effectiveness and robustness of the AI-MET system. The performance assessment for AI-MET and the five statistical and machine learning models was performed by computing sensitivity, specificity, accuracy, and precision. The AI-MET system scores 100 percent in the five metrics used on the training and testing dataset and 99 percent on the validation dataset. Statistical analysis tests were also performed to evaluate the robustness and ensure a thorough and balanced evaluation, in addition to demonstrating the statistical significance of MET-30 performance compared to the baseline models.
{"title":"AI-MET: A deep learning-based clinical decision support system for distinguishing multisystem inflammatory syndrome in children from endemic typhus","authors":"Abraham Bautista-Castillo , Angela Chun , Tiphanie P. Vogel , Ioannis A. Kakadiaris","doi":"10.1016/j.compbiomed.2025.109815","DOIUrl":"10.1016/j.compbiomed.2025.109815","url":null,"abstract":"<div><div>The COVID-19 pandemic brought several diagnostic challenges, including the post-infectious sequelae multisystem inflammatory syndrome in children (MIS-C). Some of the clinical features of this syndrome can be found in other pathologies such as Kawasaki disease, toxic shock syndrome, and endemic typhus. Endemic typhus, or murine typhus, is an acute infection treated much differently than MIS-C, so early detection is crucial to a favorable prognosis for patients with these disorders. Clinical Decision Support Systems (CDSS) are computer systems designed to support the decision-making of medical teams about their patients and intended to improve uprising clinical challenges in healthcare. In this article, we present a CDSS to distinguish between MIS-C and typhus, which includes a scoring system that allows the timely distinction of both pathologies using only clinical and laboratory features typically available within the first six hours of presentation to the Emergency Department. The proposed approach was trained and tested on datasets of 87 typhus patients and 133 MIS-C patients. A comparison was made against five well-known statistical and machine-learning models. A second dataset with 111 MIS-C patients was used to verify the effectiveness and robustness of the AI-MET system. The performance assessment for AI-MET and the five statistical and machine learning models was performed by computing sensitivity, specificity, accuracy, and precision. The AI-MET system scores 100 percent in the five metrics used on the training and testing dataset and 99 percent on the validation dataset. Statistical analysis tests were also performed to evaluate the robustness and ensure a thorough and balanced evaluation, in addition to demonstrating the statistical significance of MET-30 performance compared to the baseline models.</div></div>","PeriodicalId":10578,"journal":{"name":"Computers in biology and medicine","volume":"188 ","pages":"Article 109815"},"PeriodicalIF":7.0,"publicationDate":"2025-02-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143464358","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Brain tumors pose a significant health challenge due to their aggressive nature, complex structure, and often poor prognosis. They can be categorized as benign or malignant, with gliomas being the most prevalent and deadly form. Conventional treatments like surgery, radiation, and chemotherapy often fall short in effectiveness, prompting the need for innovative therapeutic approaches. Quantitative Structure-Property Relationship (QSPR) analysis has emerged as a cutting-edge computational tool for predicting molecular properties and aiding in the discovery of potential anti-tumor agents. This study leverages QSPR analysis to evaluate and forecast the bioactivity and pharmacokinetics of compounds designed to target brain tumors. The Banhatti indices consistently demonstrate high correlation values ranging from 0.8 to 0.9 with the specified properties. To enhance the decision-making process, the CRITIC method assigns weights to each criterion (totaling 1) and employs two Multi-Criteria Decision-Making (MCDM) techniques, Combined Compromise Solution (CoCoSo) and Multi-Attributive Border Approximation area Comparison (MABAC). CoCoSo integrates various criteria in a compromise-based approach, while MABAC offers a precise comparative framework for ranking therapeutic options. Notably, the Afinitor anti-brain tumor medications analyzed in this study were ranked No. 1 in both the CoCoSo and MABAC methods, underscoring the reliability of these approaches for decision-making purposes. In contrast to earlier research that mostly relies on single-criterion evaluation or various degree-based topological indices for drug discovery, this study fills the gap by integrating topological indices such as Banhatti indices with drug physical characteristics to offer an extensive perspective. The findings demonstrate the effectiveness of the methodology, with consistent rankings aligned with known therapeutic outcomes. This work establishes a foundation for integrating QSPR and MCDM techniques, contributing to advancements in drug discovery for complex diseases such as brain tumors.
{"title":"Multi-criteria decision making: Revealing Afinitor as the leading brain tumor drug Using CRITIC, CoCoSo, and MABAC methods combined with QSPR analysis via Banhatti indices","authors":"Abid Mahboob , Laiba Amin , Muhammad Waheed Rasheed , Jahangeer Karamat","doi":"10.1016/j.compbiomed.2025.109820","DOIUrl":"10.1016/j.compbiomed.2025.109820","url":null,"abstract":"<div><div>Brain tumors pose a significant health challenge due to their aggressive nature, complex structure, and often poor prognosis. They can be categorized as benign or malignant, with gliomas being the most prevalent and deadly form. Conventional treatments like surgery, radiation, and chemotherapy often fall short in effectiveness, prompting the need for innovative therapeutic approaches. Quantitative Structure-Property Relationship (QSPR) analysis has emerged as a cutting-edge computational tool for predicting molecular properties and aiding in the discovery of potential anti-tumor agents. This study leverages QSPR analysis to evaluate and forecast the bioactivity and pharmacokinetics of compounds designed to target brain tumors. The Banhatti indices consistently demonstrate high correlation values ranging from 0.8 to 0.9 with the specified properties. To enhance the decision-making process, the CRITIC method assigns weights to each criterion (totaling 1) and employs two Multi-Criteria Decision-Making (MCDM) techniques, Combined Compromise Solution (CoCoSo) and Multi-Attributive Border Approximation area Comparison (MABAC). CoCoSo integrates various criteria in a compromise-based approach, while MABAC offers a precise comparative framework for ranking therapeutic options. Notably, the Afinitor anti-brain tumor medications analyzed in this study were ranked No. 1 in both the CoCoSo and MABAC methods, underscoring the reliability of these approaches for decision-making purposes. In contrast to earlier research that mostly relies on single-criterion evaluation or various degree-based topological indices for drug discovery, this study fills the gap by integrating topological indices such as Banhatti indices with drug physical characteristics to offer an extensive perspective. The findings demonstrate the effectiveness of the methodology, with consistent rankings aligned with known therapeutic outcomes. This work establishes a foundation for integrating QSPR and MCDM techniques, contributing to advancements in drug discovery for complex diseases such as brain tumors.</div></div>","PeriodicalId":10578,"journal":{"name":"Computers in biology and medicine","volume":"188 ","pages":"Article 109820"},"PeriodicalIF":7.0,"publicationDate":"2025-02-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143464397","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-02-22DOI: 10.1016/j.compbiomed.2025.109848
Belmiro P.M. Duarte , Anthony C. Atkinson
Phase I clinical trials are the first-in-human studies that primarily focus on the safety profile of drugs. Traditionally, the primary aim of a phase I clinical trial is to establish the maximum tolerated dose and characterize the toxicity profile of the tested agents. As a secondary aim, some phase I studies also include studies to obtain preliminary efficacy information about the experimental agents. In our research, we consider the optimal design of experiments in extended phase I clinical trials where both efficacy and toxicity are measured and the maximum tolerated dose has been established. We represent the response of both outcomes using a bivariate probit model for correlated responses and propose systematic numerical approaches based on Semidefinite Programming to address the problem. We construct locally optimal experimental designs for the following situations: (i) responses with efficacy and toxicity strongly correlated versus non-correlated, by varying the correlation parameter; (ii) a priori known correlation versus unknown correlation; (iii) unconstrained versus constrained designs, where the constraints represent safety limits, budget constraints and probability bounds; (iv) single versus combined drugs. Additionally, we consider four distinct optimality criteria: D–, A–, E–, and K–optimality. Our methodologies are extensively tested, and we demonstrate the optimality of the designs using equivalence theorems. To enrich our analysis, an equivalence theorem for the K–optimality criterion is derived.
{"title":"Optimal designs for efficacy-toxicity response in dose finding studies using the bivariate probit model","authors":"Belmiro P.M. Duarte , Anthony C. Atkinson","doi":"10.1016/j.compbiomed.2025.109848","DOIUrl":"10.1016/j.compbiomed.2025.109848","url":null,"abstract":"<div><div>Phase I clinical trials are the first-in-human studies that primarily focus on the safety profile of drugs. Traditionally, the primary aim of a phase I clinical trial is to establish the maximum tolerated dose and characterize the toxicity profile of the tested agents. As a secondary aim, some phase I studies also include studies to obtain preliminary efficacy information about the experimental agents. In our research, we consider the optimal design of experiments in extended phase I clinical trials where both efficacy and toxicity are measured and the maximum tolerated dose has been established. We represent the response of both outcomes using a bivariate probit model for correlated responses and propose systematic numerical approaches based on Semidefinite Programming to address the problem. We construct locally optimal experimental designs for the following situations: (i) responses with efficacy and toxicity strongly correlated versus non-correlated, by varying the correlation parameter; (ii) <em>a priori</em> known correlation versus unknown correlation; (iii) unconstrained versus constrained designs, where the constraints represent safety limits, budget constraints and probability bounds; (iv) single versus combined drugs. Additionally, we consider four distinct optimality criteria: D–, A–, E–, and K–optimality. Our methodologies are extensively tested, and we demonstrate the optimality of the designs using equivalence theorems. To enrich our analysis, an equivalence theorem for the K–optimality criterion is derived.</div></div>","PeriodicalId":10578,"journal":{"name":"Computers in biology and medicine","volume":"188 ","pages":"Article 109848"},"PeriodicalIF":7.0,"publicationDate":"2025-02-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143471615","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-02-22DOI: 10.1016/j.compbiomed.2025.109838
Felix Tempel , Espen Alexander F. Ihlen , Lars Adde , Inga Strümke
In Human Activity Recognition (HAR), understanding the intricacy of body movements within high-risk applications is essential. This study uses SHapley Additive exPlanations (SHAP) to explain the decision-making process of Graph Convolution Networks (GCNs) when classifying activities with skeleton data. We employ SHAP to explain two real-world datasets: one for cerebral palsy (CP) classification and the widely used NTU RGB+D 60 action recognition dataset. To test the explanation, we introduce a novel perturbation approach that modifies the model’s edge importance matrix, allowing us to evaluate the impact of specific body key points on prediction outcomes. To assess the fidelity of our explanations, we employ informed perturbation, targeting body key points identified as important by SHAP and comparing them against random perturbation as a control condition. This perturbation enables a judgment on whether the body key points are truly influential or non-influential based on the SHAP values. Results on both datasets show that body key points identified as important through SHAP have the largest influence on the accuracy, specificity, and sensitivity metrics. Our findings highlight that SHAP can provide granular insights into the input feature contribution to the prediction outcome of GCNs in HAR tasks. This demonstrates the potential for more interpretable and trustworthy models in high-stakes applications like healthcare or rehabilitation.
{"title":"Explaining Human Activity Recognition with SHAP: Validating insights with perturbation and quantitative measures","authors":"Felix Tempel , Espen Alexander F. Ihlen , Lars Adde , Inga Strümke","doi":"10.1016/j.compbiomed.2025.109838","DOIUrl":"10.1016/j.compbiomed.2025.109838","url":null,"abstract":"<div><div>In Human Activity Recognition (HAR), understanding the intricacy of body movements within high-risk applications is essential. This study uses SHapley Additive exPlanations (SHAP) to explain the decision-making process of Graph Convolution Networks (GCNs) when classifying activities with skeleton data. We employ SHAP to explain two real-world datasets: one for cerebral palsy (CP) classification and the widely used NTU RGB+D 60 action recognition dataset. To test the explanation, we introduce a novel perturbation approach that modifies the model’s edge importance matrix, allowing us to evaluate the impact of specific body key points on prediction outcomes. To assess the fidelity of our explanations, we employ informed perturbation, targeting body key points identified as important by SHAP and comparing them against random perturbation as a control condition. This perturbation enables a judgment on whether the body key points are truly influential or non-influential based on the SHAP values. Results on both datasets show that body key points identified as important through SHAP have the largest influence on the accuracy, specificity, and sensitivity metrics. Our findings highlight that SHAP can provide granular insights into the input feature contribution to the prediction outcome of GCNs in HAR tasks. This demonstrates the potential for more interpretable and trustworthy models in high-stakes applications like healthcare or rehabilitation.</div></div>","PeriodicalId":10578,"journal":{"name":"Computers in biology and medicine","volume":"188 ","pages":"Article 109838"},"PeriodicalIF":7.0,"publicationDate":"2025-02-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143464356","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-02-22DOI: 10.1016/j.compbiomed.2025.109893
Shakhnoza Muksimova , Sabina Umirzakova , Sevara Mardieva , Nargiza Iskhakova , Murodjon Sultanov , Young Im Cho
Brain tumors are regarded as one of the most lethal, devastating, and aggressive diseases, significantly reducing the life expectancy of affected individuals. For this reason, in pursuit of advancing brain tumor diagnostics, this study introduces a significant enhancement to the YOLOv5m model by integrating an Enhanced Spatial Attention (ESA) layer, tailored specifically for the analysis of magnetic resonance imaging (MRI) brain scans. Traditional brain tumor detection methods, heavily reliant on expert interpretation of MRI, are fraught with challenges such as high variability and the risk of human error. Our innovative approach leverages the ESA layer to acutely focus on salient features, significantly improving the method ability to differentiate between common classes of brain tumors—meningioma, pituitary, and glioma tumors. By processing spatial features with enhanced precision, the model minimizes false positives and maximizes detection reliability. Validated against a comprehensive dataset of 3064 T1-weighted contrast-enhanced MRI images from 233 patients, our modified YOLOv5m architecture demonstrates superior performance metrics compared to the standard model, highlighting its potential as a robust tool in clinical applications for automated and precise brain tumor diagnosis.
{"title":"A lightweight attention-driven YOLOv5m model for improved brain tumor detection","authors":"Shakhnoza Muksimova , Sabina Umirzakova , Sevara Mardieva , Nargiza Iskhakova , Murodjon Sultanov , Young Im Cho","doi":"10.1016/j.compbiomed.2025.109893","DOIUrl":"10.1016/j.compbiomed.2025.109893","url":null,"abstract":"<div><div>Brain tumors are regarded as one of the most lethal, devastating, and aggressive diseases, significantly reducing the life expectancy of affected individuals. For this reason, in pursuit of advancing brain tumor diagnostics, this study introduces a significant enhancement to the YOLOv5m model by integrating an Enhanced Spatial Attention (ESA) layer, tailored specifically for the analysis of magnetic resonance imaging (MRI) brain scans. Traditional brain tumor detection methods, heavily reliant on expert interpretation of MRI, are fraught with challenges such as high variability and the risk of human error. Our innovative approach leverages the ESA layer to acutely focus on salient features, significantly improving the method ability to differentiate between common classes of brain tumors—meningioma, pituitary, and glioma tumors. By processing spatial features with enhanced precision, the model minimizes false positives and maximizes detection reliability. Validated against a comprehensive dataset of 3064 T1-weighted contrast-enhanced MRI images from 233 patients, our modified YOLOv5m architecture demonstrates superior performance metrics compared to the standard model, highlighting its potential as a robust tool in clinical applications for automated and precise brain tumor diagnosis.</div></div>","PeriodicalId":10578,"journal":{"name":"Computers in biology and medicine","volume":"188 ","pages":"Article 109893"},"PeriodicalIF":7.0,"publicationDate":"2025-02-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143464359","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-02-22DOI: 10.1016/j.compbiomed.2025.109821
Muhammad Nabeel Asim , Tayyaba Asif , Faiza Mehmood , Andreas Dengel
Peptides are gaining significant attention in diverse fields such as the pharmaceutical market has seen a steady rise in peptide-based therapeutics over the past six decades. Peptides have been utilized in the development of distinct applications including inhibitors of SARS-COV-2 and treatments for conditions like cancer and diabetes. Distinct types of peptides possess unique characteristics, and development of peptide-specific applications require the discrimination of one peptide type from others. To the best of our knowledge, approximately 230 Artificial Intelligence (AI) driven applications have been developed for 22 distinct types of peptides, yet there remains significant room for development of new predictors. A Comprehensive review addresses the critical gap by providing a consolidated platform for the development of AI-driven peptide classification applications. This paper offers several key contributions, including presenting the biological foundations of 22 unique peptide types and categorizes them into four main classes: Regulatory, Therapeutic, Nutritional, and Delivery Peptides. It offers an in-depth overview of 47 databases that have been used to develop peptide classification benchmark datasets. It summarizes details of 288 benchmark datasets that are used in development of diverse types AI-driven peptide classification applications. It provides a detailed summary of 197 sequence representation learning methods and 94 classifiers that have been used to develop 230 distinct AI-driven peptide classification applications. Across 22 distinct types peptide classification tasks related to 288 benchmark datasets, it demonstrates performance values of 230 AI-driven peptide classification applications. It summarizes experimental settings and various evaluation measures that have been employed to assess the performance of AI-driven peptide classification applications. The primary focus of this manuscript is to consolidate scattered information into a single comprehensive platform. This resource will greatly assist researchers who are interested in developing new AI-driven peptide classification applications.
{"title":"Peptide classification landscape: An in-depth systematic literature review on peptide types, databases, datasets, predictors architectures and performance","authors":"Muhammad Nabeel Asim , Tayyaba Asif , Faiza Mehmood , Andreas Dengel","doi":"10.1016/j.compbiomed.2025.109821","DOIUrl":"10.1016/j.compbiomed.2025.109821","url":null,"abstract":"<div><div>Peptides are gaining significant attention in diverse fields such as the pharmaceutical market has seen a steady rise in peptide-based therapeutics over the past six decades. Peptides have been utilized in the development of distinct applications including inhibitors of SARS-COV-2 and treatments for conditions like cancer and diabetes. Distinct types of peptides possess unique characteristics, and development of peptide-specific applications require the discrimination of one peptide type from others. To the best of our knowledge, approximately 230 Artificial Intelligence (AI) driven applications have been developed for 22 distinct types of peptides, yet there remains significant room for development of new predictors. A Comprehensive review addresses the critical gap by providing a consolidated platform for the development of AI-driven peptide classification applications. This paper offers several key contributions, including presenting the biological foundations of 22 unique peptide types and categorizes them into four main classes: Regulatory, Therapeutic, Nutritional, and Delivery Peptides. It offers an in-depth overview of 47 databases that have been used to develop peptide classification benchmark datasets. It summarizes details of 288 benchmark datasets that are used in development of diverse types AI-driven peptide classification applications. It provides a detailed summary of 197 sequence representation learning methods and 94 classifiers that have been used to develop 230 distinct AI-driven peptide classification applications. Across 22 distinct types peptide classification tasks related to 288 benchmark datasets, it demonstrates performance values of 230 AI-driven peptide classification applications. It summarizes experimental settings and various evaluation measures that have been employed to assess the performance of AI-driven peptide classification applications. The primary focus of this manuscript is to consolidate scattered information into a single comprehensive platform. This resource will greatly assist researchers who are interested in developing new AI-driven peptide classification applications.</div></div>","PeriodicalId":10578,"journal":{"name":"Computers in biology and medicine","volume":"188 ","pages":"Article 109821"},"PeriodicalIF":7.0,"publicationDate":"2025-02-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143464360","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-02-22DOI: 10.1016/j.compbiomed.2025.109878
Kajsa Tunedal , Tino Ebbers , Gunnar Cedersund
Cardiovascular digital twins and mechanistic models can be used to obtain new biomarkers from patient-specific hemodynamic data. However, such model-derived biomarkers are only clinically relevant if the uncertainty of the biomarkers is smaller than the variation between timepoints/patients. Unfortunately, this uncertainty is challenging to calculate, as the uncertainty of the underlying hemodynamic data is largely unknown and has several sources that are not additive or normally distributed. This violates normality assumptions of current methods; implying that also biomarkers have an unknown uncertainty. To remedy these problems, we herein present a method, with attached code, for uncertainty calculation of model-derived biomarkers using non-normal data. First, we estimated all sources of uncertainty, both normal and non-normal, in hemodynamic data used to personalize an existing model; the errors in 4D flow MRI-derived stroke volumes were 5–20 % and the blood pressure errors were 0 ± 8 mmHg. Second, we estimated the resulting model-derived biomarker uncertainty for 100 simulated datasets, sampled from the data distributions, by: 1) combining data uncertainties 2) parameter estimation, 3) profile-likelihood. The true biomarker values were found within a 95 % confidence interval in 98 % (median) of the cases. This shows both that our estimated data uncertainty is reasonable, and that we can use profile-likelihood despite the non-normality. Finally, we demonstrated that e.g. ventricular relaxation rate has a smaller uncertainty (∼10 %) than the variation across a clinical cohort (∼40 %), meaning that these biomarkers have clinical potential. Our results take us one step closer to the usage of model-derived biomarkers for cardiovascular patient characterization.
{"title":"Uncertainty in cardiovascular digital twins despite non-normal errors in 4D flow MRI: Identifying reliable biomarkers such as ventricular relaxation rate","authors":"Kajsa Tunedal , Tino Ebbers , Gunnar Cedersund","doi":"10.1016/j.compbiomed.2025.109878","DOIUrl":"10.1016/j.compbiomed.2025.109878","url":null,"abstract":"<div><div>Cardiovascular digital twins and mechanistic models can be used to obtain new biomarkers from patient-specific hemodynamic data. However, such model-derived biomarkers are only clinically relevant if the uncertainty of the biomarkers is smaller than the variation between timepoints/patients. Unfortunately, this uncertainty is challenging to calculate, as the uncertainty of the underlying hemodynamic data is largely unknown and has several sources that are not additive or normally distributed. This violates normality assumptions of current methods; implying that also biomarkers have an unknown uncertainty. To remedy these problems, we herein present a method, with attached code, for uncertainty calculation of model-derived biomarkers using non-normal data. First, we estimated all sources of uncertainty, both normal and non-normal, in hemodynamic data used to personalize an existing model; the errors in 4D flow MRI-derived stroke volumes were 5–20 % and the blood pressure errors were 0 ± 8 mmHg. Second, we estimated the resulting model-derived biomarker uncertainty for 100 simulated datasets, sampled from the data distributions, by: 1) combining data uncertainties 2) parameter estimation, 3) profile-likelihood. The true biomarker values were found within a 95 % confidence interval in 98 % (median) of the cases. This shows both that our estimated data uncertainty is reasonable, and that we can use profile-likelihood despite the non-normality. Finally, we demonstrated that e.g. ventricular relaxation rate has a smaller uncertainty (∼10 %) than the variation across a clinical cohort (∼40 %), meaning that these biomarkers have clinical potential. Our results take us one step closer to the usage of model-derived biomarkers for cardiovascular patient characterization.</div></div>","PeriodicalId":10578,"journal":{"name":"Computers in biology and medicine","volume":"188 ","pages":"Article 109878"},"PeriodicalIF":7.0,"publicationDate":"2025-02-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143464396","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号