Intelligence-based medicine最新文献

{"title":"A conceptual explainable multi-modal breast cancer intelligence framework with feasibility analysis, computational characterisation, and clinical translation roadmap","authors":"Ahmed Kateb Jumaah Al-Nussairi,&nbsp;Mohammed Kadhim Rahma,&nbsp;Ahmed Shakir Al-Hiti,&nbsp;Farrukh Bakhritdinov,&nbsp;Bakhodir Rakhimov,&nbsp;Khayala Mammadova,&nbsp;Maytham T. Qasim,&nbsp;Mohammad Khishe","doi":"10.1016/j.ibmed.2026.100465","DOIUrl":"10.1016/j.ibmed.2026.100465","url":null,"abstract":"<div><div>Breast cancer continues to be one of the highest killers among women in the world, therefore, there is a pressing need for intelligent clinical decision-support systems that are not only accurate, but also transparent, semantically interpretable and deployable in heterogeneous healthcare environments. Despite the recent progress with multimodal AI systems in enhancing diagnostic accuracy, current methods face limitations in the integration of heterogeneous data, their semantic reasoning, the lack of explainability, and collaborative learning limitations due to privacy concerns. To overcome these shortcomings, this research introduces the concept of OncoGraph-LLM, a conceptual multimodal explainable breast cancer intelligence framework that integrates deep imaging analysis, language intelligence, semantic knowledge representation, explainability by design and distributed privacy-preserving learning, in a single clinically-centric architecture. The framework leverages cooperative CNN–Vision Transformer feature extraction to utilize complementary local and global representations of images, the integration of contextual clinical information via ClinicalBERT and BioGPT, knowledge graph guided semantic inference to enhance reasoning, and an integrated explainability strategy that integrates feature attribution, semantic reasoning, uncertainty-aware interpretation, and clinician-oriented explanation generation to support transparent clinical decision-making. In addition, a federated learning architecture is implemented to support collaborative secure intelligence among distributed healthcare institutions without compromising patient privacy. To support the conceptual feasibility of the proposed framework, the design traceability, the computational complexity, deployment feasibility, technology readiness, comparative architectural analysis, and qualitative module contribution assessments are provided to illustrate the expected capabilities and the potential for translation of the integrated system. These analyses do not involve empirical performance testing, but offer a theoretical rationale for the proposed architecture, and highlight important implementation points. In conclusion, the proposed OncoGraph-LLM framework provides a holistic conceptual framework for reliable, interpretable, privacy-sensitive, and clinically actionable multimodal artificial intelligence for precision breast cancer diagnosis and decision-making.</div></div>","PeriodicalId":73399,"journal":{"name":"Intelligence-based medicine","volume":"15 ","pages":"Article 100465"},"PeriodicalIF":0.0,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148951241","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":"Triplet learning-based generalized additive model leveraging explainable convolutional neural network for the early detection of autism in children","authors":"Rucha Agrawal,&nbsp;Renuka Agrawal,&nbsp;Saiprasad Potharaju","doi":"10.1016/j.ibmed.2026.100466","DOIUrl":"10.1016/j.ibmed.2026.100466","url":null,"abstract":"<div><div>Autism belongs to a group of neurodevelopmental disorders in toddlers and is one of the fastest-growing conditions worldwide. Although it is a lifelong condition, early detection is crucial to ensure timely intervention and therapy, which can dramatically improve long-term outcomes and quality of life. The existing detection frameworks often struggle with various drawbacks, including generalizability issues, limited datasets, computational demands, and poor scalability. To address these limitations, the research proposes a Triplet Learning-based Generalized Additive Model-enabled Explainable Convolutional Neural Network (TLGXCN) for Autism Spectrum Disorder (ASD) detection. The research incorporates a triplet learning mechanism to improve the learning of eye gaze patterns using past and present knowledge. The utilization of a Generative Additive Model (GAM) within the Convolutional Neural Network (CNN) model provides more sophisticated and realistic modeling for the disorder by capturing the intricate, non-linear interplay between multiple risk factors, thereby enhancing reliability. Moreover, the research employs Shapley Additive Explanations to enhance model interpretability, effectively linking high predictive power with transparent feature insights. Besides, the development of the Hybrid Structural Saliency Textural Pattern-based feature extraction technique learns eye-gaze patterns by generating fixation maps and improves the models’ overall performance in autism detection. Therefore, the combination of these advanced methods offers accurate ASD detection by analyzing various eye-gaze patterns and ensures high generalizability with less computational complexity. Furthermore, the experimental achievements show a greater accuracy of 98.11%, 98.11% for F1-Score, 98.69% for precision, 97.53% for sensitivity, and 98.40% for specificity for 90% of training, and outperforming other existing frameworks using the Autism_Image_Dataset. The proposed TLGXCN framework demonstrates significant potential for precise and comprehensible early detection of autism in children, providing a clinically useful decision-support tool that can expedite diagnosis and treatment.</div></div>","PeriodicalId":73399,"journal":{"name":"Intelligence-based medicine","volume":"15 ","pages":"Article 100466"},"PeriodicalIF":0.0,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148951242","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":"Stability-aware hybrid intelligence for interpretable and scalable diabetes risk prediction","authors":"Yasser Taha Alzubaidi ,&nbsp;Ahmed Kateb Jumaah Al-Nussairi ,&nbsp;Ali k. Abdul Raheem ,&nbsp;Deepak Thakur ,&nbsp;Kabul Khudaybergenov ,&nbsp;Ahmed Shakir Al-Hiti ,&nbsp;Bakhodir Rakhimov ,&nbsp;Kasim Sakran Abass ,&nbsp;Saleem Malik ,&nbsp;Mohammad Khishe","doi":"10.1016/j.ibmed.2026.100388","DOIUrl":"10.1016/j.ibmed.2026.100388","url":null,"abstract":"<div><div>AI models must be effective for diabetes risk prediction, but present techniques converge prematurely and lack transparency. This study proposes the DiaMetaHybridOptimizer (DMHO), an improved multi-phase adaptive hybrid framework to enhance prediction performance and clinical interpretability. DMHO utilizes a feedback-driven architecture with the Lemur Optimizer, Marine Predators Algorithm, and Manta Ray Foraging Optimization to maintain population diversity through fitness-ranked dynamic mutation. Based on five benchmark datasets and classifiers, DMHO exceeded baseline approaches with 96.8% accuracy. By reducing feature dimensionality by 75%, the framework identified compact subsets of clinically significant predictors including HbA1c, Glucose, and BMI. ANOVA analysis showed significant improvements in convergence and efficiency. DMHO's interpretable solution supports AI-assisted decision-making by combining computational outputs with real-world diagnostic reasoning.</div></div>","PeriodicalId":73399,"journal":{"name":"Intelligence-based medicine","volume":"14 ","pages":"Article 100388"},"PeriodicalIF":0.0,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148178176","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":"Improving trust in AI diagnosis of pulmonary hypertension with patient-specific insight","authors":"Farhad Fathieh ,&nbsp;Navid Nemati ,&nbsp;Timothy Burton ,&nbsp;Horace R. Gillins ,&nbsp;Ian Shadforth ,&nbsp;Charles R. Bridges ,&nbsp;Shyam Ramchandani","doi":"10.1016/j.ibmed.2026.100390","DOIUrl":"10.1016/j.ibmed.2026.100390","url":null,"abstract":"<div><div>Although AI diagnostic models can achieve high accuracy, their black-box nature limits clinical adoption due to poor interpretability and transparency. To address this, we employed Local Interpretable Model-Agnostic Explanations (LIME) to generate patient-specific insight into the results from the PH-Model, a machine-learned model for detecting pulmonary hypertension (PH). Our approach improves explanation fidelity using bootstrapped sampling, kernel width optimization, and local model performance analysis. We introduce PHysiological Local EXplanations (PHLEX), which aggregates feature importance into six physiologically meaningful categories, and PHysiological Global EXplanations (PHGEX), derived from true positive and negative subgroups. Together, PHLEX and PHGEX enhance interpretability and robustness, reducing complexity from 217 input features to 6 categories. Fidelity of explanations was confirmed through statistical testing, and PHGEX generalized well to unseen datasets. This framework provides reliable, patient-specific explanations without requiring prior knowledge of model inputs, supporting the white-boxing of diagnostic models and aiding clinical decision-making in cardiovascular care.</div></div>","PeriodicalId":73399,"journal":{"name":"Intelligence-based medicine","volume":"14 ","pages":"Article 100390"},"PeriodicalIF":0.0,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148178179","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":"Multimodal artificial intelligence for early cancer detection via liquid biopsy, imaging, and clinical records","authors":"M. Rajesh ,&nbsp;S. Prema ,&nbsp;R.V. Siva Balan ,&nbsp;U. Nilabar Nisha ,&nbsp;Silambarasu Chinnu ,&nbsp;Usha Moorthy","doi":"10.1016/j.ibmed.2026.100395","DOIUrl":"10.1016/j.ibmed.2026.100395","url":null,"abstract":"<div><div>Tumours are diverse and multiscale, making it difficult for modern medicine to diagnose early cancer. Using structured clinical data, radiologic imaging features, and liquid samples, this research presents a multimodal AI framework for the early and reliable detection of cancer. The proposed approach surpasses single-modality approaches by integrating signals from various domains, including cancer genetic, anatomical, and physiological data. Using attention-based fusion, representation learning, and better preprocessing, we developed a prediction model that fine-tuned the weights of different modes. The results of the experiments demonstrated that it outperformed unimodal models on all datasets in terms of sensitivity, specificity, and generalisation. The framework has potential for screening purposes because of its ability to detect cancer at an early stage. Clinical confidence and interpretability were both boosted by the results of explainability tests, which revealed substantial feature contributions. The suggested multimodal framework outperformed unimodal baselines across all assessment cohorts with an AUC of 0.94, sensitivity of 0.91, and specificity of 0.88. Experimental results confirm multimodal fusion's clinically interpretable early cancer detection and precision oncology decision assistance.</div></div>","PeriodicalId":73399,"journal":{"name":"Intelligence-based medicine","volume":"14 ","pages":"Article 100395"},"PeriodicalIF":0.0,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148178175","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":"ctDrug: Conditional target-aware drug design based on decoder transformer and cross-attention","authors":"Sedigheh Vahidi Ferdosi,&nbsp;Amir Lakizadeh","doi":"10.1016/j.ibmed.2026.100399","DOIUrl":"10.1016/j.ibmed.2026.100399","url":null,"abstract":"<div><div>Target-aware de novo drug generation remains challenging due to the difficulty of effectively incorporating protein information without sacrificing molecular diversity or syntactic validity. Most current models are either target-agnostic or use oversimplified conditioning (e.g., discrete labels), limiting their relevance. We propose ctDrug, an autoregressive framework that conditions molecule generation on continuous protein embeddings via cross-attention, eliminating the need for 3D structures. A key contribution is the systematic comparison of sequence-based (ProtT5) and molecule-centric (Chem-BERTa) target representations within the same architecture. On DRD2 and HTR1A, ctDrug outperforms label-based baselines in distributional fidelity (FCD/SNN) while achieving high validity, novelty, and drug-like properties. Results suggest that continuous embeddings can provide more flexible target conditioning compared to discrete label-based approaches.</div></div>","PeriodicalId":73399,"journal":{"name":"Intelligence-based medicine","volume":"14 ","pages":"Article 100399"},"PeriodicalIF":0.0,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148178177","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":"Designing AI tools for Pathology: A mixed-method study on user interface design for breast cancer lymph node metastases detection","authors":"Sofia Jarkman ,&nbsp;Martin Lindvall ,&nbsp;Claes Lundström ,&nbsp;Darren Treanor ,&nbsp;Jeroen van der Laak","doi":"10.1016/j.ibmed.2026.100396","DOIUrl":"10.1016/j.ibmed.2026.100396","url":null,"abstract":"<div><h3>Background</h3><div>Histopathological assessment of lymph nodes is essential in breast cancer care. Artificial intelligence (AI) has shown promise in supporting this process, yet successful clinical adoption requires attention to human-AI interaction and user interface design. This study explored the user experience of an AI-based clinical decision support tool assisting pathologists in diagnosing breast cancer metastases.</div></div><div><h3>Methods</h3><div>We conducted a within-subjects mixed-methods study in which ten pathologists evaluated two AI-based clinical decision support user interface prototypes for breast cancer metastasis detection in sentinel nodes. <em>ANNO</em> displayed threshold based on-image annotations with minimal automation, whereas <em>RAVS</em> presented four likelihood regions of interest (ROI) per slide with automated measurement and case summaries. Quantitative measures included review time, diagnostic outcomes, and System Usability Scale (SUS) scores. Qualitative data were gathered via semi-structured interviews and analysed thematically.</div></div><div><h3>Results</h3><div>ANNO achieved higher usability scores (mean SUS 90.3) and was associated with a sense of control and resembled current diagnostic workflow, whereas RAVS (mean SUS 80.0) was preferred by participants valuing rapid navigation and automated measurements. Interview findings highlighted trade-offs: ANNO preserved diagnostic overview and autonomy, while RAVS improved navigational efficiency but reduced whole-slide context. ROI presentation strategy seemed to influence preference. Trust in AI was generally cautious and linked to needs for transparency and long-term clinical exposure.</div></div><div><h3>Conclusion</h3><div>Both interfaces were perceived as potentially clinically valuable but differed in their usability and workflow integration. Our exploratory findings indicates that for future development of AI-based decision aiding tools for this diagnostic task, a combination of high human control with targeted automation is advisable, combined with visual clarity, and an effective ROI presentation strategy.</div></div>","PeriodicalId":73399,"journal":{"name":"Intelligence-based medicine","volume":"14 ","pages":"Article 100396"},"PeriodicalIF":0.0,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148178178","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":"Rider-based CNN with AResU-Net Segmentation for Brain Tumor in 3D MRI image","authors":"V. Anitha","doi":"10.1016/j.ibmed.2026.100394","DOIUrl":"10.1016/j.ibmed.2026.100394","url":null,"abstract":"<div><div>Magnetic Resonance Imaging is most widely utilized option for diagnosing the brain tumors, however accurate analysis of the brain tumors using 3D MRI scans is challenging because of the irregular tumor boundaries, poorly contrasted images and diverse shapes. Hence a novel Rider-based CNN with AResU-Net is introduced for the precise segmentation and classification of the brain tumors in 3D MRI image. Traditional methods struggles due to infiltrative tumors, diffused tumor boundaries, unclear and over-segmentation, leads to reduced classification accuracy. To address this challenge, a novel Attention-Based Residual U-Net with fuzzy clustering is introduced to enhance the segmentation by reducing the voxel errors in unclear 3D boundaries. The fuzzy clustering in the U-Net assigns membership values to pixels for ensuring smoother transitions across tissue boundaries and reducing the segmentation errors with unclear borders. Moreover, the existing CNN lacks in predicting the long-range characteristics because of fine-grained spatial data loss. To maintain long-term dependencies and enhance accuracy, a Rider optimization-based CNN-LSTM is introduced. During optimization, the CNN extracts the spatial features, while the LSTM handles long-term dependencies, with the Adaptive Rider Optimization (ARO) algorithm optimizing LSTM weights. The proposed model achieved superior performance, with a classification accuracy of 99.85% and sensitivity of 99.8%, thus mitigating the segmentation errors and addressing challenges in the brain tumor analysis.</div></div>","PeriodicalId":73399,"journal":{"name":"Intelligence-based medicine","volume":"14 ","pages":"Article 100394"},"PeriodicalIF":0.0,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148178174","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":"A scoping review of machine learning models for predicting preterm birth and miscarriage: Mapping the landscape and the performance-validation paradox","authors":"Golnoush Shahraki ,&nbsp;Elias Mazrooei Rad ,&nbsp;Mohammad Heidari","doi":"10.1016/j.ibmed.2025.100337","DOIUrl":"10.1016/j.ibmed.2025.100337","url":null,"abstract":"<div><h3>Objective</h3><div>This scoping review aims to summarize how machine learning (ML) has been applied to predict adverse pregnancy outcomes—particularly preterm birth and miscarriage—by comparing data sources, model designs, and reported performance.</div></div><div><h3>Methods</h3><div>Evidence from 40 eligible studies was systematically reviewed. For each, we extracted details on the type of data used (e.g., clinical records, imaging, biomarkers), sample size, ML algorithms, and major performance metrics such as AUC and accuracy. Recurring strengths and weaknesses were identified through thematic analysis.</div></div><div><h3>Results</h3><div>The reviewed studies drew on a wide range of data sources—from large electronic health records (EHRs) to imaging and time-series physiological signals. Tree-based algorithms and support vector machines generally showed strong predictive performance, with several studies reporting AUC values above 0.90. However, most investigations were limited by small, single-center datasets and lacked external validation, raising concerns about generalizability and clinical interpretability.</div></div><div><h3>Conclusion</h3><div>Machine learning approaches could meaningfully improve how clinicians anticipate adverse pregnancy outcomes, but their clinical use remains premature. Progress will depend on studies that include larger and more diverse populations, apply rigorous external validation, and focus on developing models that clinicians can interpret and act upon.</div></div>","PeriodicalId":73399,"journal":{"name":"Intelligence-based medicine","volume":"13 ","pages":"Article 100337"},"PeriodicalIF":0.0,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145926922","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":"Stage-wise mutual attention network for diagnostic support of chronic ischaemic heart disease using multimodal medical information","authors":"Shoko Miyauchi ,&nbsp;Shunsuke Takashima ,&nbsp;Ken’ichi Morooka ,&nbsp;Ryo Kurazume","doi":"10.1016/j.ibmed.2026.100360","DOIUrl":"10.1016/j.ibmed.2026.100360","url":null,"abstract":"<div><div>Chronic ischaemic heart disease (CIHD) is a leading cause of death worldwide. Contributing factors include lifestyle-related diseases, such as diabetes and hypertension, as well as genetic predispositions. The diagnosis of CIHD requires physicians to perform multiple tests, including highly invasive tests, which impose a significant burden on patients. To address this issue, a diagnostic support network for CIHD has been proposed, which uses multimodal medical information obtained solely from non-invasive tests as input. However, in this system, feature integration from different modalities is performed by concatenating each feature, which may not fully account for the diagnostic relevance of each input. In this study, we propose a new diagnostic support network for CIHD that integrates multimodal features using an attention mechanism. Moreover, we introduce a stage-wise integration approach in which different feature vectors are progressively combined, two at a time. This allows features from different modalities to be gradually integrated while preserving diagnostically relevant information. We confirmed that the proposed method outperforms conventional approaches in classifying healthy subjects and patients using non-invasive multimodal data. Furthermore, we demonstrated that the classification performance improves when integrating the three modalities step-by-step, starting from pairs of closely related modalities rather than merging all three modalities simultaneously.</div></div>","PeriodicalId":73399,"journal":{"name":"Intelligence-based medicine","volume":"13 ","pages":"Article 100360"},"PeriodicalIF":0.0,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146188804","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}