International Journal of Medical Informatics最新文献

{"title":"Topic modeling and sentiment analysis of Greek clinician–patient conversations in hematologic malignancies","authors":"Maria Evangelia Chatzimina ,&nbsp;Helen A. Papadaki ,&nbsp;Charalampos Pontikoglou ,&nbsp;Manolis Tsiknakis","doi":"10.1016/j.ijmedinf.2025.106071","DOIUrl":"10.1016/j.ijmedinf.2025.106071","url":null,"abstract":"<div><h3>Background</h3><div>Conversations in clinical settings, especially those involving hematologic cancers or palliative care are not only informational but also emotionally charged. Understanding how these conversations are structured could help develop AI systems that support not only information exchange but also personalized emotional care.</div></div><div><h3>Objective</h3><div>This study explored the themes discussed and emotional tones in real conversations between Greek speaking clinicians and patients. Our aim was to contribute toward building digital tools for low-resource languages such as Greek that are not only linguistically appropriate but also emotionally aware.</div></div><div><h3>Methods</h3><div>We analyzed over 52,000 anonymized utterances from real Greek clinical conversations using BERTopic for topic modeling and a domain-specific sentiment classifier fine-tuned on Greek medical data. Topics were labeled using both a large language model (Gemma-3) and extractive keyword methods (KeyBERT).</div></div><div><h3>Results</h3><div>The analysis revealed 35 thematic clusters, covering areas like symptoms, diagnosis, emotional states, treatment choices, and end-of-life planning. Sentiment analysis revealed that patients expressed more negative sentiment in utterances related to pain, uncertainty or personal loss. Clinicians often responded with a more neutral or even empathetic positive tone, especially when offering support or medical advice.</div></div><div><h3>Conclusions</h3><div>Our findings show that Natural Language Processing (NLP) can reveal both the content and emotional tone of conversations in medical settings. This approach could support the development of digital health tools such as conversational agents that are more emotionally aware in Greek and other low-resource languages, especially in emotionally charged domains like palliative care.</div></div>","PeriodicalId":54950,"journal":{"name":"International Journal of Medical Informatics","volume":"204 ","pages":"Article 106071"},"PeriodicalIF":4.1,"publicationDate":"2025-07-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144749834","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}

{"title":"Predicting nonsurgical treatment outcomes in lumbar disc herniation: leveraging sparse electronic health records for patient phenotyping","authors":"Ye-Seul Lee,&nbsp;Yoon Jae Lee,&nbsp;In-Hyuk Ha","doi":"10.1016/j.ijmedinf.2025.106056","DOIUrl":"10.1016/j.ijmedinf.2025.106056","url":null,"abstract":"<div><h3>Objective</h3><div>Electronic health records (EHRs) offer a wealth of patient data but often fail to capture the dynamic progression of symptoms, particularly in non-continuously monitored conditions like lumbar disc herniation (LDH). This study aimed to enhance the prediction of nonsurgical treatment outcomes for LDH using latent class trajectory modeling (LCTM) combined with machine learning, leveraging sparse EHRs to inform patient care.</div></div><div><h3>Materials and methods</h3><div>The EHRs of 6,732 patients (2017–2021) were obtained and divided into training (2017–2019) and prediction (2020–2021) datasets. LCTM identified symptom progression trajectories using the Oswestry Disability Index (ODI), which were incorporated into machine learning models as patient phenotypes. Predictions of achieving the minimum clinically important difference (MCID) in ODI at discharge were compared between a baseline naïve model and a combined model utilizing phenotypes as intermediate variables.</div></div><div><h3>Results and discussion</h3><div>Three distinct trajectories were identified, reflecting varied recovery patterns. Patient characteristics were classified into three phenotypes. Incorporating these phenotypes into the predictive model improved the area under the receiver operating characteristic curve (AUROC) from 0.78 in the naïve model to 0.82 in the combined model. Precision, recall, and F1 scores also improved with the combined model, underscoring its robustness.</div></div><div><h3>Conclusion</h3><div>This data-driven approach enhanced interpretability and prediction accuracy, demonstrating the potential to guide clinical decision-making. By addressing data sparsity in EHRs and minimizing temporal data leakage, the integration of LCTM and machine learning enables robust prediction models for treatment outcomes, facilitating personalized care and improved outcomes in nonsurgical LDH treatment. In clinical settings with sparse longitudinal data, this approach facilitates the development of robust prediction models for treatment outcomes, enabling personalized care strategies and achieving better outcomes in nonsurgical LDH treatment.</div></div>","PeriodicalId":54950,"journal":{"name":"International Journal of Medical Informatics","volume":"204 ","pages":"Article 106056"},"PeriodicalIF":3.7,"publicationDate":"2025-07-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144713891","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}

{"title":"Leveraging open-source large language models (LLMs) in scoping reviews: a case study on disability and AI applications","authors":"Azadeh Bayani ,&nbsp;Leandre Parfait Epoh Ewane ,&nbsp;Davllyn Santos Oliveira dos Anjos ,&nbsp;Muriel Mac-Seing ,&nbsp;Jean Noel Nikiema","doi":"10.1016/j.ijmedinf.2025.106048","DOIUrl":"10.1016/j.ijmedinf.2025.106048","url":null,"abstract":"<div><h3>Background</h3><div>Large language models (LLMs) have the potential to offer solutions for automating many of the manual tasks involved in scientific reviews, including data extraction, literature screening, summarization, and quality assessment.</div></div><div><h3>Objectives</h3><div>This study aims to evaluate the performance of LLMs in the task of title and abstract screening and full-text data extraction of a scoping review study, by identifying their effectiveness, efficiency, and potential integration into human-based and manual tasks.</div></div><div><h3>Materials and Method</h3><div>The following key three steps of a scientific scoping review were automated: 1) Title and Abstract Screening, 2) Full-Text Screening, and 3) Data Extraction based on nine study dimensions. The four most recent lightweight open-source LLMs −Mistral, Vicuna, and Llama 3.2 with 1B and 3B parameters- were applied and evaluated through the steps.</div></div><div><h3>Results</h3><div>Llama 3.2-3B demonstrated the best performance in the title and abstract screening, achieving an accuracy of 66 %, excelling in the exclusion of papers. For full-text screening, it maintained the highest overall accuracy of 65 %, effectively identifying excluded papers. In data extraction, the Mistral model outperformed others across most dimensions, though Llama 3.2-3B excelled in extracting objectives and study implications.</div></div><div><h3>Discussion and conclusion</h3><div>The present study underscores both the potential and limitations of LLMs in automating scoping reviews. Automating the entire scoping review without human intervention is sub-optimal. Using a more controlled approach balances the strengths of LLMs with the need for human judgment, supporting not only the replication of scientific reviews but also their continuous refinement and follow-up over time.</div></div>","PeriodicalId":54950,"journal":{"name":"International Journal of Medical Informatics","volume":"204 ","pages":"Article 106048"},"PeriodicalIF":4.1,"publicationDate":"2025-07-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144722528","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}

{"title":"Digital divide and health professional shortages: telehealth access for chronic disease management in rural Florida","authors":"Di Shang ,&nbsp;Cynthia Williams ,&nbsp;Aishwarya Joshi ,&nbsp;Christopher Baynard","doi":"10.1016/j.ijmedinf.2025.106054","DOIUrl":"10.1016/j.ijmedinf.2025.106054","url":null,"abstract":"<div><h3>Background</h3><div>This study examines telehealth utilization among Medicare and Medicaid beneficiaries with chronic diseases in geographically vulnerable areas of Florida during COVID-19.</div></div><div><h3>Methods</h3><div>Using a retrospective design, we retrieved visit-level data from a State of Florida Medicaid and Medicare Managed Care Organization from 2019 and 2020; we included 54,927 unique patients. We aggregated telehealth use at the county level and calculated the percentage of patients who used video telehealth in each county during the study period. We used data from the Digital Divide Index (DDI), Health Professional Shortage Area Designations (HPSA), and Florida Health Outcome Rankings to investigate the intersectionality of rurality, digital divide, health professional shortage, health outcome, and telehealth utilization during the pandemic. T-test and regression analyses were used to examine the study objectives.</div></div><div><h3>Results</h3><div>Our results suggested that telehealth visits in non-metro counties were higher than in metro counties before the pandemic (2.84 % vs. 0.61 %, p &lt; 0.001). However, telehealth usage in non-metro counties was lower during the pandemic than in metro counties (35.17 % vs. 40.6 %, p &lt; 0.001). The digital divide was higher in non-metro counties (p-value &lt; 0.001) and negatively associated with telehealth usage in 2020 (p &lt; 0.001). The analysis of health outcomes shows that the digital divide (p &lt; 0.001) and health professional shortage (p = 0.042) were negatively associated with county health outcome rankings.</div></div><div><h3>Conclusion</h3><div>Enhancing telehealth usage in rural regions requires a holistic approach that includes multilevel collaborations, incentives to providers, digital literacy programs, and broadband subsidies.</div></div>","PeriodicalId":54950,"journal":{"name":"International Journal of Medical Informatics","volume":"204 ","pages":"Article 106054"},"PeriodicalIF":3.7,"publicationDate":"2025-07-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144685659","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}

{"title":"Comparative study of LOINC and SNOMED CT in panel mapping: enhancing interoperability in laboratory testing","authors":"Hyejin Ryu ,&nbsp;Sumi Sung ,&nbsp;Kuenyoul Park ,&nbsp;Min-Sun Kim ,&nbsp;YeJin Oh ,&nbsp;Shinae Yu ,&nbsp;Eun-Jung Cho ,&nbsp;Sollip Kim","doi":"10.1016/j.ijmedinf.2025.106055","DOIUrl":"10.1016/j.ijmedinf.2025.106055","url":null,"abstract":"<div><h3>Background</h3><div>We aimed to evaluate and compare the applicability of Logical Observation Identifiers Names and Codes (LOINC) and SNOMED CT in mapping frequently requested panel tests.</div></div><div><h3>Method</h3><div>Frequently requested panel tests were identified from the test records of two major referral laboratories. Subsequently, LOINC and SNOMED CT mappings were cross-validated, and the results were classified based on pre-defined criteria. A consensus was reached among the teams. A comparative structural analysis was performed by aligning the mapped SNOMED CT concepts with LOINC codes and visualizing the selected items with conditionality to evaluate the differences in representation.</div></div><div><h3>Results</h3><div>We conducted 23 panel tests. Exact mapping was achieved for 87.0% of the panel tests using LOINC: two failures were recorded due to the lack of a suitable code, and one panel test was classified as narrowly mapped. In contrast, SNOMED CT achieved the exact mapping for 78.2% of the panel tests, with 12 mappings requiring post-coordination to represent the test concepts. SNOMED CT mapping failures stemmed from missing primitive concepts and limited pre-coordinated codes. LOINC offered detailed component specifications and predefined panel codes. In contrast, SNOMED CT relied on post-coordination to address gaps in the pre-defined codes, allowing the addition of new combinations where necessary.</div></div><div><h3>Conclusion</h3><div>LOINC demonstrated advantages in mapping frequently performed panel tests with higher exact mapping rates and structured panel codes. However, leveraging SNOMED CT’s flexibility through initiatives such as the LOINC Ontology project could enhance interoperability and standardization in laboratory data exchange.</div></div>","PeriodicalId":54950,"journal":{"name":"International Journal of Medical Informatics","volume":"204 ","pages":"Article 106055"},"PeriodicalIF":3.7,"publicationDate":"2025-07-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144703970","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}

{"title":"LLM-powered breast cancer staging from PET/CT reports: a comparative performance study","authors":"Daniel Spitzl ,&nbsp;Markus Mergen ,&nbsp;Rickmer Braren ,&nbsp;Lukas Endrös ,&nbsp;Matthias Eiber ,&nbsp;Lisa Steinhelfer","doi":"10.1016/j.ijmedinf.2025.106053","DOIUrl":"10.1016/j.ijmedinf.2025.106053","url":null,"abstract":"<div><h3>Purpose</h3><div>Imaging reports are crucial in breast cancer management, with the tumor-node-metastasis (TNM) classification serving as a widely used model for assessing disease severity, guiding treatment decisions, and predicting patient outcomes. Large language models (LLMs) offer a potential solution by extracting standardized UICC TNM classifications and the corresponding UICC stage directly from existing PET/CT reports. This approach holds promise to enhance staging accuracy, streamline multidisciplinary discussions, and improve patient outcomes.</div></div><div><h3>Methods</h3><div>Here, we evaluated four LLMs—ChatGPT-4o, DeepSeek V3, Claude 3.5 Sonnet, and Gemini 2.0 Flash—for their capacity to determine TNM staging based on UICC/AJCC breast cancer guidelines. A total of 111 fictitious PET/CT reports were analyzed, and each model’s outputs were measured against expert-generated TNM classifications and stage categorizations.</div></div><div><h3>Results</h3><div>Among the tested models, Claude 3.5 Sonnet demonstrated superior F1 scores of 0.95%, 0.95%, 1.00% and 0.92% for T, N, M classification and UICC stage classification, respectively.</div></div><div><h3>Conclusions</h3><div>These findings underscore the ability of advanced natural language processing (NLP) technologies to support reliable cancer staging, potentially aiding clinicians. Despite the encouraging performance, prospective clinical trials and validation across diverse practice settings remain critical to confirming these preliminary outcomes. Nonetheless, this study highlights the promise of LLM-based systems in reinforcing the accuracy of oncologic workflows and lays the groundwork for broader adoption of AI-driven tools in breast cancer management.</div></div>","PeriodicalId":54950,"journal":{"name":"International Journal of Medical Informatics","volume":"204 ","pages":"Article 106053"},"PeriodicalIF":3.7,"publicationDate":"2025-07-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144685660","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}

{"title":"Towards practical federated learning and evaluation for medical prediction models","authors":"Andrei Kazlouski ,&nbsp;Ileana Montoya Perez ,&nbsp;Faiza Noor ,&nbsp;Mikael Högerman ,&nbsp;Otto Ettala ,&nbsp;Tapio Pahikkala ,&nbsp;Antti Airola","doi":"10.1016/j.ijmedinf.2025.106046","DOIUrl":"10.1016/j.ijmedinf.2025.106046","url":null,"abstract":"<div><div><em>Background</em>: Federated learning (FL) is a rapidly advancing technique that enables collaborative model training while preserving data privacy. This approach is particularly relevant in healthcare, where privacy concerns and regulatory restrictions often prevent centralized data sharing. FL has shown promise in tasks such as disease detection, achieving performance levels comparable to centralized systems. However, its practical usability in real-world applications remains underexplored.</div><div><em>Methods</em>: We evaluate the practical effectiveness of FL in predicting whether patients suspected of prostate cancer require invasive biopsy procedures. The study uses 14 publicly available prostate cancer datasets from 10 countries. We propose and benchmark a novel FL evaluation strategy, Leave-Silo-Out (LSO), which quantifies the performance gap between federated training and free-riding (utilizing the federated model without contributing data). Additionally, we investigate whether locally trained models can outperform multi-hospital FL models. The results are assessed with a focus on improving the diagnosis of local patients.</div><div><em>Results</em>: Our findings reveal that the benefits of FL vary with the amount of locally available annotated data. Hospitals with very small datasets see negligible improvements from FL compared to free-riding. Institutions with moderate datasets may achieve some gains through FL training. However, hospitals with extensive datasets often experience little to no advantage from FL and, in some cases, observe reduced performance compared to local training.</div><div><em>Conclusion</em>: Federated learning shows potential in scenarios with limited data availability. However, its practical applicability is highly context-dependent, influenced by factors such as data availability and specific task requirements.</div></div>","PeriodicalId":54950,"journal":{"name":"International Journal of Medical Informatics","volume":"204 ","pages":"Article 106046"},"PeriodicalIF":3.7,"publicationDate":"2025-07-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144685661","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}

{"title":"Bridging the digital divide: artificial intelligence as a catalyst for health equity in primary care settings","authors":"Ayokunle Osonuga ,&nbsp;Adewoyin A. Osonuga ,&nbsp;Sandra Chinaza Fidelis ,&nbsp;Gloria C. Osonuga ,&nbsp;Jack Juckes ,&nbsp;David B. Olawade","doi":"10.1016/j.ijmedinf.2025.106051","DOIUrl":"10.1016/j.ijmedinf.2025.106051","url":null,"abstract":"<div><h3>Background</h3><div>Health inequalities remain one of the most pressing challenges in contemporary healthcare, with primary care serving as both a gateway to services and a potential source of disparities. Artificial intelligence (AI) technologies offer unprecedented opportunities to address these inequities through enhanced diagnostic capabilities, improved access to care, and personalised interventions.</div></div><div><h3>Objective</h3><div>This comprehensive narrative review aimed to synthesise current evidence on AI applications in primary care settings, specifically targeting health inequality reduction and identifying both opportunities and barriers for equitable implementation.</div></div><div><h3>Method</h3><div>Following PRISMA-ScR (Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews) guidelines, we employed a systematic approach to literature identification, selection, and synthesis across seven electronic databases covering literature from 2020 to 2024. Of 1,247 initially identified studies, 89 met inclusion criteria with 52 providing sufficient data quality for evidence synthesis.</div></div><div><h3>Results</h3><div>The review identified promising applications such as AI-powered risk stratification algorithms that improved hypertension control in low-income populations, telemedicine platforms reducing geographic barriers in rural communities, and natural language processing tools facilitating care for non-native speakers. However, significant challenges persist, including algorithmic bias that may perpetuate existing inequities, the digital divide excluding vulnerable populations, and insufficient representation in training datasets. Current evidence suggests that whilst AI holds transformative potential for advancing health equity, successful implementation requires intentional co-design with affected communities, robust bias mitigation strategies, and comprehensive digital literacy programmes.</div></div><div><h3>Conclusion</h3><div>Future research must prioritise equity-centred AI development, longitudinal outcome studies in diverse populations, and policy frameworks ensuring responsible deployment. However, careful consideration of unintended consequences, including potential overdiagnosis, erosion of human clinical judgement, and inadvertent exclusion of vulnerable populations, is essential to prevent AI from exacerbating existing health disparities. The paradigm shift towards equity-first AI design represents a critical opportunity to leverage technology for social justice in healthcare.</div></div>","PeriodicalId":54950,"journal":{"name":"International Journal of Medical Informatics","volume":"204 ","pages":"Article 106051"},"PeriodicalIF":3.7,"publicationDate":"2025-07-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144685656","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}

{"title":"A multi-task deep sequential neural network for IgA nephropathy Oxford classification and prognosis prediction","authors":"Sai Pan ,&nbsp;Yibing Fu ,&nbsp;Lai Jiang ,&nbsp;Jiaona Liu ,&nbsp;Guangyan Cai ,&nbsp;Wenge Li ,&nbsp;Weicen Liu ,&nbsp;Xiaofei Wang ,&nbsp;Zhong Yin ,&nbsp;Quan Hong ,&nbsp;Jie Wu ,&nbsp;Yong Wang ,&nbsp;Shuwei Duan ,&nbsp;Jingjing Chen ,&nbsp;Pu Chen ,&nbsp;Mai Xu ,&nbsp;Xiangmei Chen","doi":"10.1016/j.ijmedinf.2025.106052","DOIUrl":"10.1016/j.ijmedinf.2025.106052","url":null,"abstract":"<div><h3>Background</h3><div>While deep learning has advanced pathological analysis in IgA nephropathy (IgAN), the lack of integrated models that combine multi-label structural identification, Oxford classification, and prognosis prediction remains a significant clinical challenge.</div></div><div><h3>Methods</h3><div>We developed DeepSNN, a novel deep sequential neural network that serves as a multi-task model trained on multi-center multi-modal renal datasets. The architecture integrates lesion segmentation, glomerular classification, Oxford MEST-C scoring, and prognosis prediction subnets. To ensure interpretability, we conducted visualization experiments and comparative analyses with pathologists’ diagnostic patterns. Pathologist comparisons employed Cohen’s Kappa with blinded re-evaluation of test and validation sets.</div></div><div><h3>Results</h3><div>DeepSNN demonstrated exceptional lesion identification capabilities across the People’s Liberation Army General (PLAG) Hospital dataset (n = 245) and China-Japan Friendship (CJF) Hospital dataset (n = 32), achieving dice coefficients of 0.95 and 0.92, respectively. For Oxford classification, DeepSNN delivered outstanding outcomes with high Kappa values of 0.84, 0.79, 0.87, 0.87, and 0.82 for M, E, S, T, and C scores on the PLAG dataset. Notably, our method outperformed three junior pathologists and achieved comparable performance to senior pathologists across both datasets. During a median follow-up of 47.7 (IQR: 21.9–61.1) months, DeepSNN excelled in prognosis prediction (AUC: 0.810), demonstrating improvement over the International IgA Nephropathy Prediction Tool (IIPT) (AUC: 0.742, ΔAUC = +0.068) in PLAG Hospital dataset (n = 245). Furthermore, visualization maps showed consistent pathological region identification between pathologists and DeepSNN.</div></div><div><h3>Conclusions</h3><div>DeepSNN successfully integrates multiple diagnostic tasks with performance comparable to senior pathologists, demonstrating substantial potential for streamlining IgAN clinical workflows. This innovation addresses critical gaps in automated renal pathology analysis while maintaining clinical interpretability.</div></div>","PeriodicalId":54950,"journal":{"name":"International Journal of Medical Informatics","volume":"204 ","pages":"Article 106052"},"PeriodicalIF":4.1,"publicationDate":"2025-07-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144739188","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}

{"title":"Natural language processing in medical text processing: A scoping literature review","authors":"Luis B. Elvas ,&nbsp;Ana Almeida ,&nbsp;João C. Ferreira","doi":"10.1016/j.ijmedinf.2025.106049","DOIUrl":"10.1016/j.ijmedinf.2025.106049","url":null,"abstract":"<div><h3>Background</h3><div>The exponential growth of digitized medical data has created significant challenges for healthcare professionals, as medical documentation transitions from simple text records to complex, multi-dimensional data structures. Natural Language Processing (NLP), particularly Named Entity Recognition (NER), has emerged as a crucial tool for extracting and categorizing critical information from clinical texts. The development of transformer-based models like BERT and the ability to fine-tune pre-trained AI models have revolutionized the field, offering unprecedented opportunities to enhance the efficient and precise interpretation of medical data across diverse languages and healthcare contexts.</div></div><div><h3>Objective</h3><div>This literature review aimed to analyze recent NLP approaches for medical text processing, examining techniques, performance metrics, and advancements across different languages and healthcare contexts.</div></div><div><h3>Method</h3><div>Following the Preferred Reporting Items for Systematic Reviews and Meta Analyses (PRISMA) methodology, a scoping search was conducted in Scopus and PubMed databases, focusing on studies published between 2019–2024. The review included studies on language model fine-tuning and information extraction in healthcare, with a specific search query designed to capture relevant NLP techniques.</div></div><div><h3>Results</h3><div>Of 67 initial records, 31 studies were ultimately included. Bidirectional Encoder Representations from Transformers (BERT)-based approaches, neural networks, and CRF/LSTM techniques dominated, consistently achieving F1-scores above 85 %. The studies covered multiple languages, with 51.5 % in English, 27.3 % in Chinese, and smaller representations in Italian, German, and Spanish. Hybrid approaches and techniques addressing data privacy and limited labeled data were notably prevalent.</div></div><div><h3>Conclusions</h3><div>The review revealed that modern NLP techniques, particularly BERT-based models and hybrid approaches, show significant promise in medical text processing across different languages. While challenges remain in cross-lingual adaptation and data availability, these technologies demonstrate potential to enhance medical data interpretation and analysis.</div></div>","PeriodicalId":54950,"journal":{"name":"International Journal of Medical Informatics","volume":"204 ","pages":"Article 106049"},"PeriodicalIF":3.7,"publicationDate":"2025-07-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144685655","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}