Informatics in Medicine Unlocked最新文献

{"title":"Comparative development and evaluation of multilayer fuzzy and adaptive neuro-fuzzy inference systems for early diagnosis of chronic kidney disease","authors":"Arvind Sharma ,&nbsp;Dalwinder Singh ,&nbsp;Arun Singh ,&nbsp;MVV Prasad Kantipudi ,&nbsp;Saiprasd Potharaju ,&nbsp;B. Suresh Babu","doi":"10.1016/j.imu.2025.101724","DOIUrl":"10.1016/j.imu.2025.101724","url":null,"abstract":"<div><div>Chronic Kidney Disease (CKD) is a gradual yet life-threatening condition that often remains undetected until significant renal impairment has occurred. Early recognition is essential to slow its progression and reduce the risk of associated complications. Conventional diagnostic procedures have limitations; they are subjective and may be ineffective when symptoms take longer to manifest. In this paper, we propose and compare two intelligent diagnostic models—a Multilayer Fuzzy Inference System (MLFIS) and an Adaptive Neuro-Fuzzy Inference System (ANFIS)—that aim to diagnose early-stage CKD using clinical symptoms and laboratory indicators.</div><div>The model of MLFIS is constructed in dual form i.e. the first layer determines the probable presence of CKD, which is based on symptoms like hypertension, diabetes, and family history, and the second layer specifies the disease stage, which is done based on biochemical parameters attached with or without or increasing glomerular filtration rate, blood urea nitrogen, serum creatinine, albumin, and urinary pus cells. The ANFIS model, however, combines the strengths of both fuzzy logic and neural networks, providing automatic learning of diagnosis rules based on data, thereby increasing accuracy and flexibility.</div><div>The models had been constructed and benchmarked on a retrospective dataset consisting of 500 anonymised patient records retrieved from a publicly available resource, and then tested using standard performance measures. Although the ANFIS had a classification accuracy of 99.5 % which is a higher percentage than the 99.33 per cent accuracy of the MLFIS, the difference is not that great. Findings demonstrate the benchmarking performance of fuzzy and neuro-fuzzy systems for CKD staging, providing a foundation for future validation on hospital datasets. However, it is important to note that the high accuracies reported are based on a relatively small UCI dataset and should be interpreted as internal validation only. Broader clinical validation is necessary before deployment in real-world settings. The paper also compares these two systems and determines their performance based on current machine learning models. This study seeks to fill the gap between the laboratory-based diagnostic and real clinical practice by presenting a scalable and understandable solution to screening of CKD.</div></div>","PeriodicalId":13953,"journal":{"name":"Informatics in Medicine Unlocked","volume":"60 ","pages":"Article 101724"},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145788410","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 joint frailty model to assess the relationship between time to curative treatment and biochemical recurrence in prostate cancer patients","authors":"Abderrahim Oussama Batouche ,&nbsp;Denis Rustand ,&nbsp;Eugen Czeizler ,&nbsp;Håvard Rue ,&nbsp;Tuomas Mirtti ,&nbsp;Antti Rannikko","doi":"10.1016/j.imu.2025.101727","DOIUrl":"10.1016/j.imu.2025.101727","url":null,"abstract":"<div><h3>Objective:</h3><div>Conflicting evidence exists regarding the effect of delaying prostate cancer (PCa) treatment on outcomes after curative treatment. Ideally, modelling this would require a joint analysis of the time to treatment initiation and the time to PCa recurrence. However, traditional Cox models are limited to analysing a single time-to-event outcome. In this study, we build a joint frailty model to assess the effect of delayed or expedited primary curative treatment (surgery or radiotherapy) on the risk of biochemical recurrence (BCR).</div></div><div><h3>Methods:</h3><div>We used HUS (Helsinki Metropolitan Hospital District) data lake for data mining and to categorise PCa patients by Gleason grade group (1-5), treatment, and biochemical recurrence for a final sample size of n=9934 patients (1993–2019). A broader definition of BCR was established considering secondary treatments as an additional indicator of relapse alongside traditional PSA cut-offs. We applied the INLA method for Bayesian inference, utilising the <span>INLAJoint</span> R package, to fit a shared frailty joint survival model. This model was used to analyse the relationship between the time from diagnosis to curative treatment (treatment risk) and the time from diagnosis to biochemical recurrence (relapse risk).</div></div><div><h3>Results:</h3><div>Conditional on covariates, including age at diagnosis and Gleason grade group, our joint survival model revealed a significant association <span><math><mi>γ</mi></math></span> = -1.32 [-1.50, -1.14] between the risk of treatment and risk of BCR. As an example, regardless of the grade group, patients within the top 5% with the lowest risk of receiving treatment (i.e., the longest time to treatment) exhibited an HR=5.27 [4.28, 6.62] fold (increased) risk of recurrence compared to the average patient.</div></div><div><h3>Conclusion:</h3><div>We successfully employed a joint frailty model to simultaneously model the effect of time to curative primary treatment on time to biochemical recurrence. We show that the time to curative treatment is associated with the risk of relapse. Patients of the same age, same diagnostic PSA, and same grade group who are treated early are less likely to develop biochemical recurrence.</div></div>","PeriodicalId":13953,"journal":{"name":"Informatics in Medicine Unlocked","volume":"60 ","pages":"Article 101727"},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145921551","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":"PAL-Net: A point-wise CNN with patch-attention for 3D anatomical facial landmark localization","authors":"Ali Shadman Yazdi ,&nbsp;Annalisa Cappella ,&nbsp;Benedetta Baldini ,&nbsp;Riccardo Solazzo ,&nbsp;Gianluca Tartaglia ,&nbsp;Chiarella Sforza ,&nbsp;Giuseppe Baselli","doi":"10.1016/j.imu.2025.101729","DOIUrl":"10.1016/j.imu.2025.101729","url":null,"abstract":"<div><div>Manual annotation of anatomical landmarks on 3D facial scans is a time-consuming and expertise-dependent task, yet it remains critical for clinical assessments, morphometric analysis, and craniofacial research. While several deep learning methods have been proposed for facial landmark localization, most focus on pseudo-landmarks or require complex input representations, limiting their clinical applicability. This study presents a fully automated deep learning pipeline (PAL-Net) for localizing 50 anatomical landmarks on facial models acquired via stereo-photogrammetry. The method combines coarse alignment, region-of-interest filtering, and an initial landmark approximation with a patch-based pointwise CNN enhanced by attention mechanisms. Trained and evaluated on 214 annotated scans from healthy adults, PAL-Net achieved a mean localization error of 3.686 mm and preserved relevant anatomical distances with an average error of 2.822 mm. While the geometric error exceeds expert intra-observer variability, the distance-wise error maintains structural integrity sufficient for high-throughput anthropometric analysis. To assess generalization, the model was further evaluated on 700 subjects from the FaceScape dataset, achieving a mean localization error of 0.41 mm and a distance error of 0.38 mm. Comparing with existing methods, PAL-Net offers a favorable trade-off between accuracy and computational cost. While performance degrades in regions with poor mesh quality (e.g., ears, hairline), the method demonstrates consistent accuracy across most anatomical regions. PAL-Net generalizes effectively across datasets and facial regions, outperforming existing methods in both point-wise and structural evaluations. It provides a lightweight, scalable solution for high-throughput 3D anthropometric analysis, with potential to support clinical workflows and reduce reliance on manual annotation. Source code can be accessed at <span><span>https://github.com/Ali5hadman/PAL-Net-A-Point-Wise-CNN-with-Patch-Attention</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":13953,"journal":{"name":"Informatics in Medicine Unlocked","volume":"60 ","pages":"Article 101729"},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145921552","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 medical image captioning system for TeleOTIVA: Supporting SDGs-oriented cervical precancer screening in Indonesia","authors":"Firdaus Firdaus ,&nbsp;Siti Nurmaini ,&nbsp;Rizki Ayunda Pratama ,&nbsp;Elly Matul Imah ,&nbsp;Muhammad Naufal Rachmatullah ,&nbsp;Ade Iriani Sapitri ,&nbsp;Annisa Darmawahyuni ,&nbsp;Anggun Islami ,&nbsp;Akhiar Wista Arum ,&nbsp;Bambang Tutuko ,&nbsp;Patiyus Agustiansyah ,&nbsp;Rizal Sanif ,&nbsp;Radiyati Umi Partan","doi":"10.1016/j.imu.2025.101719","DOIUrl":"10.1016/j.imu.2025.101719","url":null,"abstract":"<div><div>Cervical cancer screening using Visual Inspection with Acetic Acid (VIA) remains a critical strategy in resource-limited settings. However, its effectiveness is often hindered by diagnostic variability arising from subjective interpretation. To address this challenge, we introduce TeleOTIVA, an AI-powered system designed to automatically detect and describe cervical lesions from VIA images. The system integrates YOLOv11-based lesion detection and segmentation with a Dense Residual Network and an embedding LSTM-based image captioning module, enabling it to generate clinically meaningful descriptions encompassing lesion borders, surface texture, and anatomical location. The performance of TeleOTIVA demonstrates promising results. Evaluations of the generated captions, compared to expert-annotated ground truth, yielded high scores across multiple metrics: BLEU (0.5711), METEOR (0.6726), and ROUGE-L (0.6929). These results indicate a high degree of n-gram similarity, semantic relevance, grammatical accuracy, and structural alignment with human-generated descriptions. In other words, the model not only mirrors expert-level vocabulary but also captures the clinical essence of VIA image interpretation. This synergy between advanced lesion detection and automated caption generation significantly enhances the accuracy, efficiency, and accessibility of cervical cancer screening. TeleOTIVA thus offers a powerful and scalable diagnostic aid, particularly impactful for improving early detection efforts in underserved and low-resource regions.</div></div>","PeriodicalId":13953,"journal":{"name":"Informatics in Medicine Unlocked","volume":"60 ","pages":"Article 101719"},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145921550","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":"Review of machine learning models in short- and long-term glucose forecasting and hypoglycemia classification","authors":"Beyza Cinar ,&nbsp;Louisa van den Boom ,&nbsp;Maria Maleshkova","doi":"10.1016/j.imu.2025.101723","DOIUrl":"10.1016/j.imu.2025.101723","url":null,"abstract":"<div><div>Type 1 Diabetes (T1D) is a chronic autoimmune disorder that requires lifelong insulin therapy. A common side effect is hypoglycemia, characterized by decreased blood glucose levels (BGL) below 70 mg/dL. Diabetes care can be optimized using machine learning (ML) models that can predict and alert patients to potential glycemic abnormalities. The ML models can be classified into regression-based, in which glucose levels are forecasted, and classification-based, in which adverse events are classified. This review analyzes the performance of ML models applied to T1D and compares these in terms of short- and long-term prediction horizons (PHs), defined as 15–120 min and 3 to more than 24 h, respectively. This review investigates: 1) How much in advance can glucose values or a hypoglycemic event be accurately predicted? 2) Which ML methods have the best performance? 3) Which factors impact the performance? And 4) Does personalization increase performance? The results indicate that 1) a PH of up to 1 h provides the best results. 2) Conventional ML methods yield the best results for classification and deep learning (DL) for regression. A single model cannot adequately classify across multiple PHs. 3) The model performance is influenced by multivariate datasets and the input sequence length (ISL). 4) Finally, personal data enhances performance, but due to limited data quality, population-based models are preferred.</div></div>","PeriodicalId":13953,"journal":{"name":"Informatics in Medicine Unlocked","volume":"60 ","pages":"Article 101723"},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145735871","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":"SuSastho.AI: A multimodal medical copilot for adolescents using evidence-based medicine and large language models","authors":"Shamim Ahamed ,&nbsp;Moinul H. Chowdhury ,&nbsp;Marzia Zaman ,&nbsp;Abhishek Agarwala ,&nbsp;Md Abdullah Al Mashud ,&nbsp;Tariqul ,&nbsp;Tareq Al Muntasir ,&nbsp;Rifat Rahman ,&nbsp;Rifat Shahriyar ,&nbsp;Farhana Sarker ,&nbsp;Khondaker A. Mamun","doi":"10.1016/j.imu.2025.101720","DOIUrl":"10.1016/j.imu.2025.101720","url":null,"abstract":"<div><div>Adolescents in Bangladesh face serious sexual, reproductive, and mental health challenges due to cultural stigma, poverty, and limited healthcare infrastructure. Within the country, 63% of adolescents are deprived of essential sexual and reproductive health services, and only 13% receive mental health support. Adolescents living in urban slums and with disabilities face additional challenges in receiving reliable health information. This limited access exposes them to a high risk of sexually transmitted infections (STIs), unintended pregnancies, and serious mental health issues. Addressing these challenges, our study introduces SuSastho.AI, a healthcare copilot providing access to reliable health information to adolescents. We utilized large language models, along with Evidence-Based Medicine, retrieval-augmented generation, and a clinically validated dataset to provide evidence-based responses, supporting both voice and text-based interactions. Clinical evaluation of a pilot study shows our method reduces incorrect responses by 26.9% and increases response correctness by 16.1% compared to other methods. It achieved an accuracy rate of 86.7% when specifically evaluated based on available knowledge. While the responses are mostly consistent with up-to-date medical practices, occasional, less precise responses highlight the need for further refinement. Participants reported overall positive feedback, where 87% found answers to their questions, and 90.7% found responses relevant. Our results show that SuSastho.AI can provide reliable and evidence-based information while being an affordable way to support traditional healthcare systems with a high potential to transform digital health. The study sets an example as an evidence-based healthcare copilot to support adolescents and lays the foundation for future research where evidence-based tools overcome social barriers.</div></div>","PeriodicalId":13953,"journal":{"name":"Informatics in Medicine Unlocked","volume":"60 ","pages":"Article 101720"},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145788038","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 systematic review on computer vision-based methods for cervical cancer detection","authors":"Hope Mbelwa ,&nbsp;Judith Leo ,&nbsp;Crispin Kahesa ,&nbsp;Elizabeth Mkoba","doi":"10.1016/j.imu.2025.101726","DOIUrl":"10.1016/j.imu.2025.101726","url":null,"abstract":"<div><div>Cervical cancer is a leading cause of mortality among women globally, especially in regions where access to timely screening remains a challenge. With such concerns, accurate detection of cervical lesions is essential for effective diagnosis and treatment. This review aimed to explore the application of computer vision-based methods for detecting cervical cancer, identifying their potential, setbacks and areas for future development. A comprehensive literature search across Scopus, IEEE Xplore, PubMed, and Google Scholar identified 96 relevant studies published between 2014 and August 2025. These studies applied computer vision methods including CNNs, Vision Transformers, and multimodal models to cervical cancer detection using Pap smear, colposcopy, and histopathology images. They were analyzed based on the techniques employed, datasets used, evaluation metrics adopted, and reported results. This review highlights significant advancements in the field, particularly in lesion classification, precise segmentation of affected regions, and accurate detection of cancerous regions. However, some challenges were identified, including limited image datasets with insufficiently distributed normal and abnormal cases, aggravated by privacy issues and accurate labeling of medical images, which is critical and rigorous, often leading to annotation inconsistencies. Lastly, this study revealed that integrating Natural Language Processing and Computer Vision can enhance cervical cancer diagnosis through multi-modal models that combine both clinical text and imaging data. Additionally, this study proposes the use of techniques like annotation-efficient learning to manage limited labeled datasets using methods such as semi-supervised and transfer learning as well as the use of federated learning to ensure privacy in computer-aided diagnostic systems.</div></div>","PeriodicalId":13953,"journal":{"name":"Informatics in Medicine Unlocked","volume":"60 ","pages":"Article 101726"},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145788409","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":"Automated identification of left ventricular hypertrophy using cardiac ultrasound imaging: A systematic review of artificial intelligence driven approaches","authors":"Jimcymol James ,&nbsp;Anjan Gudigar ,&nbsp;U. Raghavendra ,&nbsp;Jyothi Samanth ,&nbsp;M. Maithri ,&nbsp;Aryaman Kaprekar ,&nbsp;Mukund A. Prabhu ,&nbsp;Massimo Salvi ,&nbsp;Filippo Molinari ,&nbsp;Edward J. Ciaccio ,&nbsp;U. Rajendra Acharya","doi":"10.1016/j.imu.2025.101730","DOIUrl":"10.1016/j.imu.2025.101730","url":null,"abstract":"<div><div>Left Ventricular Hypertrophy (LVH) is a significant cardiovascular risk marker that manifests in several clinical conditions, including Hypertension (HTN), Chronic Kidney Disease (CKD), and Hypertrophic Cardiomyopathy (HCM). This systematic review examines Artificial Intelligence (AI) approaches for the automated identification of these conditions using cardiac Ultrasound (US) imaging. Following the PRISMA guidelines, 37 relevant articles (7 reviews, 30 research papers) published between 2010 and 2025 were analysed. The analysis revealed three primary methodological approaches: feature learning pipelines, end-to-end Deep Learning (DL), and hybrid methods that combine both techniques. For CKD detection, only one study using cardiac US was identified, which achieved 99.09 % classification accuracy using Support Vector Machine (SVM) with steerable Gaussian filters and entropy features. HTN classification studies have demonstrated high performance across different approaches: traditional Machine Learning (ML) classifiers (decision trees with transform features: 99.11 %, weighted k-nearest neighbors: 98 %) and DL methods (Area Under Curve (AUC): 0.92–0.94). HCM studies ranged from binary classification (42.3 % of studies) to multi-class problems of increasing complexity (3-class: 38.4 %, 4-class: 11.5 %, 5-class: 7.6 %), with SVM achieving 95.2 % average sensitivity and DL models reaching an average AUC of 0.94. Current limitations include a predominant focus on binary classification problems, limited research on cardiac-based CKD detection, and a lack of standardized datasets. Future research directions include developing hybrid methodologies that combine traditional and DL approaches, creating standardized multimodal databases, implementing explainable AI techniques, and integrating Internet of Things technologies for continuous monitoring.</div></div>","PeriodicalId":13953,"journal":{"name":"Informatics in Medicine Unlocked","volume":"60 ","pages":"Article 101730"},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145921553","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":"Objective detection of Parkinson's disease motor states using Lasso-selected IMUs features","authors":"Mevludin Memedi","doi":"10.1016/j.imu.2026.101732","DOIUrl":"10.1016/j.imu.2026.101732","url":null,"abstract":"<div><h3>Background and objective</h3><div>Machine learning (ML) models that use data captured from Inertial Measurement Units (IMUs) are widely applied in the clinical management of Parkinson's disease (PD). However, performance and interpretability of these models can be influenced by the selection of input features, especially when dealing with multi-dimensional data. This study investigates the performance of Lasso regularization in improving performance and simplicity of supervised and unsupervised ML models using IMUs data.</div></div><div><h3>Methods</h3><div>Data were collected using IMUs placed on the wrists and ankles of 19 patients (14 males and 5 females) with advanced PD (mean years with disease of 10 years). Participants performed different motor tests, and three movement disorder specialists rated the severity of motor states (Off and dyskinesia) on a Treatment Response Scale (TRS). Sensor data were processed, and features were reduced by Lasso regularization and used as inputs to Support Vector Machines (SVM) for classification and regression. In addition, clustering methods were employed to align the sensor data to clinical labels.</div></div><div><h3>Results</h3><div>Linear SVM correctly classified Off motor state from treatment-induced dyskinesia state with an accuracy of 93.9 % and 93.3 %, respectively. The correlation coefficient between the predicted TRS score derived by gaussian SVM and mean TRS score of the three specialists was 0.91. The clusters derived by the clustering algorithms separated well the instances when the patients were in Off and dyskinesia motor states.</div></div><div><h3>Conclusions</h3><div>Using Lasso regularization as a feature selection method coupled with ML models yielded good predictive performance when fusing multi-sensor and -activity data from IMUs. This approach can be used as a tool to objectively assess PD motor states and improve the management of the disease by individualizing treatments.</div></div>","PeriodicalId":13953,"journal":{"name":"Informatics in Medicine Unlocked","volume":"60 ","pages":"Article 101732"},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145921549","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":"Noise-Aware Undersampling for imbalanced medical data (NAUS)","authors":"Zholdas Buribayev ,&nbsp;Ainur Yerkos ,&nbsp;Zhibek Zhetpisbay ,&nbsp;Markus Wolfien","doi":"10.1016/j.imu.2026.101731","DOIUrl":"10.1016/j.imu.2026.101731","url":null,"abstract":"<div><div>Advancements in medical research have increasingly relied on robust data analytics to support diagnostic and treatment decisions. However, data analysis still faces challenges when investigating datasets with severe class imbalance, often stemming from the rarity of certain conditions and uneven disease distributions. To address this issue, we propose the Noise-Aware Undersampling with Subsampling (NAUS) algorithm. NAUS integrates clustering, noise removal, and Tomek-link identification techniques to create refined subsamples that assess the significance of individual observations, while systematically removing redundant and noisy data. The proposed approach was evaluated on datasets related to chronic kidney disease, liver disease, heart disease and its performance was compared to that of traditional oversampling methods (e.g., SMOTE, ADASYN, LoRAS) and undersampling techniques (e.g., random undersampling, Tomek-links). Our experimental results, based on machine learning classifiers (e.g. Random Forest, LightGBM, and Multilayer Perceptron). Data visualization further confirmed that NAUS effectively mitigates class imbalance, making it a promising tool for enhancing the reliability of medical data analysis.</div></div>","PeriodicalId":13953,"journal":{"name":"Informatics in Medicine Unlocked","volume":"60 ","pages":"Article 101731"},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145921547","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}