Artificial Intelligence in Medicine

Accurate recognition and deidentification of sensitive health information (SHI) in spoken dialogues requires multimodal algorithms that can understand medical language and contextual nuance. However, the recognition and deidentification risks expose sensitive health information (SHI). Additionally, the variability and complexity of medical terminology, along with the inherent biases in medical datasets, further complicate this task. This study introduces the SREDH/AI-Cup 2025 Medical Speech Sensitive Information Recognition Challenge, which focuses on two tasks: Task-1: Speech transcription systems must accurately transcribe speech into text; and Task-2: Medical speech de-identification to detect and appropriately classify mentions of SHI. The competition attracted 246 teams; top-performing systems achieved a mixed error rate (MER) of 0.1147 and a macro F1-score of 0.7103, with average MER and macro F1-score of 0.3539 and 0.2696, respectively. Results were presented at the IW-DMRN workshop in 2025. Notably, the results reveal that LLMs were prevalent across both tasks: 97.5% of teams adopted LLMs for Task 1 and 100% for Task 2. Highlighting their growing role in healthcare. Furthermore, we finetuned six models, demonstrating strong precision ([~]0.885-0.889) with slightly lower recall ([~]0.830-0.847), resulting in F1-scores of 0.857-0.867.

Background. Adult diffuse glioma is a representative class of primary brain tumors for which accurate MRI-based tumor segmentation is indispensable for treatment planning. Conventional automated segmentation methods have relied primarily on image information and spatial prompts, and auxiliary clinical information that is routinely acquired in clinical practice has not been sufficiently exploited as an input. Objective. Building on a dual-prompt-driven Segment Anything Model (SAM) extension framework that fuses visual and language reference prompts, we propose a method that integrates patient demographics, unsupervised molecular cluster variables derived from TCGA high-throughput profiling, and histopathological parameters as learnable prompt embeddings, and we evaluate its effect on the accuracy of lower-grade glioma (LGG) MRI segmentation. Methods. An auxiliary prompt encoder converts clinical metadata into high-dimensional embeddings that are fused with the prompt representations of Segment Anything Model (SAM) ViT-B through a cross-attention fusion mechanism. The TCGA-LGG MRI Segmentation dataset (Kaggle release by Buda et al.; n = 110 patients; WHO grade II-III) was split at the patient level (train/val/test = 71/17/22) using three different random seeds, and the three slices with the largest tumor area were extracted from each patient. To avoid pseudo-replication arising from multiple slices per patient and repeated measurements across seeds, our primary analysis aggregated Dice and 95th-percentile Hausdorff distance (HD95) to the patient x seed unit (n = 66); secondary analyses at the unique-patient level (n = 22) and at the per-slice level (n = 198) are also reported. Pairwise comparisons used paired t-tests with Bonferroni correction (k = 3) and Wilcoxon signed-rank tests, and a permutation test (K = 30) served as an auxiliary check of effective use of the auxiliary information. Results. At the patient x seed level (n = 66), Proposed (full clinical) achieved a Dice gain of +0.287 over the zero-shot SAM ViT-B baseline (paired-t p = 4.2 x 10^-15, Cohen's d_z = +1.25, Bonferroni-corrected p << 0.001; Wilcoxon p = 2.0 x 10^-10), and HD95 improved from 218.2 to 64.6. Because zero-shot SAM is not designed for domain-specific medical segmentation, the large absolute HD95 gap largely reflects the expected domain gap rather than a competitive baseline. The additional contribution of the full clinical configuration over the demographics-only configuration was Dice = +0.023 (paired-t p = 0.057, Bonferroni-corrected p = 0.172), which did not reach statistical significance at the patient level and is reported as a directional trend. The permutation test (K = 30, seed 2025) yielded real-metadata Dice = 0.819 versus a shuffled-metadata mean of 0.773, giving an empirical p = 0.032 = 1/(K + 1), which is at the resolution limit of this test and should therefore be interpreted as preliminary evidence. Conclusions. Integrating auxiliary clinical information as multimodal prompts produced a large improvement over the zero-shot SAM baseline on this LGG cohort. More importantly, a robustness analysis showed that Proposed (full clinical) outperformed the trained Base (no auxiliary information) under all tested spatial-prompt conditions, including perfect centroid (+0.014), and that the advantage was most pronounced in the prompt-free regime (+0.231, p = 0.039), where the base model collapsed but the proposed model maintained meaningful segmentation by leveraging clinical metadata alone. The additional contribution of molecular and histopathological information beyond demographics was not statistically resolved at the patient level (+0.023, n.s.). Establishing clinical utility will require external validation on larger multi-center cohorts and direct comparisons with established segmentation methods. Keywords: brain tumor segmentation; Segment Anything Model (SAM); vision-language prompt-driven segmentation; auxiliary clinical prompts; multimodal learning; TCGA-LGG; deep learning

Prehabilitation (PRH) is a preoperative process aimed at optimizing patients functional capacity to improve surgical outcomes and overall well-being. While its physical and cognitive benefits are increasingly documented, its emotional impact, particularly in neuro-oncology patients, remains less explored. This study assessed the psychological effects of a PRH program on 29 brain tumor patients. The primary outcome, emotional well-being, was measured using quality of life and emotional distress metrices. Secondary outcomes included perceived stress levels and control attitudes. Additionally, qualitative data from structured interviews provided further insights into the psychological effects of the intervention. The results indicated significant improvements in quality of life and reductions in emotional distress, particularly among women. While perceived stress levels remained stable, control attitudes showed an increase. Qualitative analysis further highlighted the positive changes in the control sense and identified additional factors, such as the importance of social support sources during the PRH process. Overall, these findings suggest that PRH interventions play a significant role in enhancing emotional well-being among neuro-oncological patients in the preoperative phase. These results underscore the importance of implementing comprehensive and personalized PRH approaches to optimize clinical status both before and after surgery, thereby promoting sustained psychological benefits in this population. This study is based on data collected at Institut Guttmann in Barcelona in the context of the Prehabilita project (ClinicalTrials.gov identifier: NCT05844605; registration date: 06/05/2023).

Background and Objectives: We report the first intraoperative deployment of a real-time machine vision system in neurosurgery, derived from our previous anatomical detection work, automatically identifying structures during endoscopic endonasal surgery. Existing systems demonstrate promising performance in offline anatomical recognition, yet so far none have been implemented during live operations. Methods: A real-time anatomy detection model was trained using the YOLOv8 architecture (Ultralytics). Following training completion in the PyTorch environment, the model was exported to ONNX format and further optimized using the NVIDIA TensorRT engine. Deployment was carried out using the NVIDIA Holoscan SDK, the system ran on an NVIDIA Clara AGX developer kit. We used the model for real-time recognition of intraoperative anatomical structures and compared it with the same video labelled manually as reference. Model performance was reported using the average precision at an intersection-over-union threshold of 0.5 (AP50). Furthermore, end-to-end delay from frame acquisition to the display of the annotated output was measured. Results: A mean AP50 of 0.56 was achieved. The model demonstrated reliable detection of the most relevant landmarks in the transsphenoidal corridor. The mean end-to-end latency of the model was 47.81 ms (median 46.57 ms). Conclusion: For the first time, we demonstrate that clinical-grade, real-time machine-vision assistance during neurosurgery is feasible and can provide continuous, automated anatomical guidance from the surgical field. This approach may enhance intraoperative orientation, reduce cognitive load, and offer a powerful tool for surgical training. These findings represent an initial step toward integrating real-time AI support into routine neurosurgical workflows.

Background: Artificial intelligence (AI) has the potential to improve the efficiency of evidence synthesis and reduce human error. However, robust methods for evaluating rapidly evolving AI tools within the practical workflows of evidence synthesis remain underdeveloped. This protocol describes a study design for assessing the effectiveness, efficiency, and usability of AI tools in comparison to traditional human-only workflows in the context of Cochrane systematic reviews. Methods: Members of the Cochrane Evaluation of (Semi-) Automated Review (CESAR) Methods Project developed an adaptive platform study-within-a-review (SWAR) design, modeled after clinical platform trials. This design employs a master protocol to concurrently evaluate multiple AI tools (interventions) against a standard human-only process (control) across three key review tasks: title and abstract screening, full-text screening, and data extraction. The adaptive framework allows for the addition or removal of AI tools based on interim performance analyses without necessitating a restart of the study. Performance will be assessed using metrics such as accuracy (sensitivity, specificity, precision), efficiency (time on task), response stability, impact of errors, and usability, in alignment with Responsible use of AI in evidence SynthEsis (RAISE) principles. Results: The study will generate comparative data about the performance and usability of specific AI tools employed in a semi- or fully automated manner relative to standard human effort. The protocol provides a flexible framework for the assessment of AI tools in evidence synthesis, addressing the limitations of static, one-time evaluations. Discussion: This study protocol presents a novel methodological approach to addressing the challenges of evaluating AI tools for evidence syntheses. By validating entire workflows rather than individual technologies, the findings will establish an evidence base for determining the viability of integrating AI into evidence-synthesis workflows. The adaptive design of this study is flexible and can be adopted by other investigators, ensuring that the evaluation framework remains relevant as new tools emerge.

BackgroundSystematic reviews are essential for evidence-based decision making in health sciences but require substantial time and resource for manual processes, particularly title and abstract screening. Recent advances in machine learning and large language models (LLMs) have demonstrated promise in accelerating screening with high recall but are often limited by modest gains in efficiency, mostly due to the absence of a generalisable stopping criterion. Here, we introduce and report preliminary findings on the performance of a novel semi-automated active learning system, JARVIS, that integrates LLM-based reasoning using the PICOS framework, neural networks-based classification, and human decision-making to facilitate abstract screening. MethodsDatasets containing author-made inclusion and exclusion decisions from six published systematic reviews were used to pilot the semi-automated screening system. Model performance was evaluated across recall, specificity and area under the curve precision-recall (AUC-PR), using full-text inclusion as the ground truth. Estimated workload and financial savings were calculated by comparing total screening time and reviewer costs across manual and semi-automated scenarios. ResultsAcross the six review datasets, recall ranged between 98.2% and 100%, and specificity ranged between 97.9% and 99.2% at the defined stopping point. Across iterations, AUC-PR values ranged between 83.8% and 100%. Compared with human-only screening, JARVIS delivered workload savings between 71.0% and 93.6%. When a single reviewer read the excluded records, workload savings ranged between 35.6 % and 46.8%. ConclusionThe proposed semi-automated system substantially reduced reviewer workload while maintaining high recall, improving on previously reported approaches. Further validation in larger and more varied reviews, as well as prospective testing, is warranted.

Background: Generative artificial intelligence (GenAI) chatbots have shown utility in assisting with various research tasks. Traditional, complementary, and integrative medicine (TCIM) is a patient-centric approach that emphasizes holistic well-being. The integration of TCIM and GenAI presents numerous key opportunities. However, TCIM researchers' attitudes toward GenAI tools remain less understood. This large-scale, international cross-sectional survey aimed to elucidate the attitudes and perceptions of TCIM researchers regarding the use of GenAI chatbots in the scientific process. Methods: A search strategy in Ovid MEDLINE identified corresponding authors who were TCIM researchers. Eligible authors were invited to complete an anonymous online survey administered via SurveyMonkey. The survey included questions on socio-demographic characteristics, familiarity with GenAI chatbots, and perceived benefits and challenges of using GenAI chatbots. Results were analysed using descriptive statistics and thematic content analysis. Results: The survey received 716 responses. Most respondents reported familiarity with GenAI chatbots (58.08%) and viewed them as very important to the future of scientific research (54.37%). The most acknowledged benefits included workload reduction (74.07%) and increased efficiency in data analysis/experimentation (71.14%). The most frequently reported challenges involved bias, errors, and limitations. More than half of the respondents (57.02%) expressed a need for training to use GenAI chatbots in the scientific process, alongside an interest in receiving training (72.07%). However, 43.67% indicated that their institutions did not offer these programs. Discussion: By developing a deeper understanding of TCIM researchers' perspectives, future AI applications in this field can be more informed, and guide future policies and collaboration among researchers.

Loneliness is clinically important but under-documented in electronic health records (EHRs), posing challenges for secondary use and computational phenotyping. This study evaluated whether natural language processing (NLP) methods can detect and classify loneliness severity from clinical notes. Patients with a loneliness survey (mild, moderate, severe) were identified, and notes within six months prior to the survey were retrieved. An expert-expanded lexicon was applied, and transformer models (RoBERTa, ClinicalBERT, Longformer) were fine-tuned for loneliness severity classification. Large language model-based summarization of social and psychiatric history was also tested as an alternative input representation. Performance was evaluated using accuracy, weighted-F1, and per-class F1. All models achieved modest accuracy (0.3 to 0.7), and struggled to identify severe loneliness, reflecting sparse and inconsistent documentation even among surveyed patients. While summarization marginally improved accuracy, gains primarily reflected mild predictions. Manual review of 100 social worker notes from severely lonely patients found explicit mentions of loneliness in only two cases, confirming that relevant documentation is exceedingly rare. These findings demonstrate that model performance is constrained by the sparse and inconsistent documentation of loneliness in EHRs, rather than by deficiencies in the modeling approach itself.

Inadequate discharge communication is a well-documented contributor to medication non-adherence, missed follow-ups, and preventable readmissions across healthcare systems worldwide. In resource-limited oncology settings, where patients are often low-literate, speak non-dominant languages, and manage complex multi-drug regimens, this problem is acute and largely unaddressed. We present Aakhyan, a vernacular patient communication platform that addresses the full post-discharge arc: from converting English-language discharge summaries into structured, voice-based vernacular explanations, through medication adherence support, to proactive follow-up management - all delivered via WhatsApp. The architecture is novel in its strict separation of concerns: a vision-language model performs structured JSON extraction from discharge images; all patient-facing content is generated deterministically from clinician-approved templates with community-sensitive vocabulary registers. This design eliminates the hallucination risk inherent in generative AI patient communication (documented at 18-82% in prior studies) while preserving the extraction capability of large language models. The platform supports four language registers, Bengali, Hindi, simplified English for tribal populations, and Assamese, with text-to-speech synthesis across all registers, including a custom grapheme-to-phoneme engine developed for Assamese phonology. Beyond discharge communication, the platform includes scheduled medication adherence nudges, interactive follow-up reminders, and a Daily Availability and Patient Notification System (DAPNS) that notifies patients the evening before their follow-up whether their doctor and required investigations are available, preventing wasted trips by rural patients who travel 2-6 hours to reach the centre. A 100-patient stratified randomised controlled study is planned at Silchar Cancer Centre, with structured teach-back assessment at 48-72 hours post-discharge as the primary comprehension outcome and preliminary clinical efficacy as a secondary objective. This paper describes the clinical rationale, technical architecture, safety framework, and positioning of Aakhyan within the existing literature on mHealth patient communication interventions.

Medical concept extraction from electronic health records underpins many downstream applications, yet remains challenging because medically meaningful concepts, such as diagnoses, are frequently implied rather than explicitly stated in medical narratives. Existing benchmarks with human-annotated evidence spans underscore the importance of grounding extracted concepts in medical text. However, they predominantly focus on explicitly stated concepts and provide limited coverage of cases in which medically relevant concepts must be inferred. We present MedicalBench, a new benchmark for medical concept extraction with evidence grounding that evaluates implicit medical reasoning. MedicalBench formulates medical concept extraction as a verification task over medical note concept pairs, coupled with sentence level evidence identification. Built from MIMIC-IV discharge summaries and human verified ICD-10 codes, the dataset is curated through a multi stage large language model (LLM) triage pipeline followed by medical annotation and expert review. It deliberately includes implicit positives, semantically confusable negatives, and cases where LLM judgments disagree with medical expert assessments. Annotators provide sentence level evidence spans and concise medical rationales. The final dataset contains 823 high quality examples. We define two complementary evaluation tasks: (1) medical concept extraction and (2) sentence level evidence retrieval, enabling assessment of both correctness and interpretability. Benchmarking state-of-the-art LLMs and a supervised baseline reveals that performance remains modest, highlighting the difficulty of extracting implicitly expressed concepts. We further show that explicitly incorporating reasoning cues and prompting to extract implicit evidence substantially improves medical concept extractions, while performance is largely invariant to note length, indicating that MedicalBench isolates reasoning difficulty rather than superficial confounders. MedicalBench provides the first systematic benchmark for implicit, evidence-grounded medical concept extraction, offering a foundation for developing medical language models that can both identify medically relevant concepts and justify their predictions in a transparent and medically faithful manner.

In this study, we investigate gender bias in a Retrieval-Augmented Generation (RAG) based AI assistant developed for Finnish wellbeing services counties. We tested the system using 36 clinically relevant queries, each rendered in three gendered variants (male, female, gender-neutral), and evaluated responses using both an LLM-as-a-judge approach and a human expert panel consisting of a physician and a sociologist specializing in ethics. We observed substantial and clinically significant differences across gendered variants, including differential treatment urgency, inappropriate symptom associations, and misidentification of clinical context. Female variants disproportionately framed responses around childcare and reproductive health regardless of clinical relevance, reflecting societal stereotypes rather than medical reasoning. Bias manifested both at the LLM generation stage and the RAG retrieval stage, in several cases causing the model to hallucinate responses entirely. Some bias patterns were persistent across repeated runs, while others appeared inconsistently, highlighting the challenge of distinguishing systematic bias from stochastic variation.

Automating Hierarchical Condition Category (HCC) assignment directly from unstructured electronic health record (EHR) notes remains an important but understudied problem in clinical informatics. We present HCC-Coder, an end to end NLP system that maps narrative documentation to 115 Centers for Medicare & Medicaid Services(CMS) HCC codes in a multi-label setting. On the test dataset, HCC-Coder achieves a macro-F1 of 0.779 and a micro-F1 of 0.756, with a macro-sensitivity of 0.819 and macro-specificity of 0.998. By contrast, Generative Pre-trained Transformer (GPT)-4o achieves highest score of a macro-F1 of 0.735 and a micro-F1 of 0.708 under five-shot prompting. The fine-tuned model demonstrates consistent absolute improvements of 4%-5% in F1-scores over GPT-4o. To address severe label imbalance, we incorporate inverse-frequency weighting and per-label threshold calibration. These findings suggest that domain-adapted transformers provide more balanced and reliable performance than prompt-based large language models for hierarchical clinical coding and risk adjustment.

Neurodegenerative diseases such as Mild Cognitive Impairment (MCI), Multiple Sclerosis (MS), Parkinson s Disease (PD), and Amyotrophic Lateral Sclerosis (ALS) are becoming more prevalent. Each of these diseases, despite its specific pathophysiological mechanisms, leads to widespread reorganization of brain activity. However, the corresponding neurophysiological signatures of these changes have been elusive. As a consequence, to date, it is not possible to effectively distinguish these diseases from neurophysiological data alone. This work uses Magnetoencephalography (MEG) resting-state data, combined with interpretable machine learning techniques, to support differential diagnosis. We expand on previous work and design a Riemannian geometry-based classification pipeline. The pipeline is fed with typical connectivity metrics, such as covariance or correlation matrices. To maintain interpretability while reducing feature dimensionality, we introduce a classifier-independent feature selection procedure that uses effect sizes derived from the Kruskal-Wallis test. The ensemble classification pipeline, called REDDI, achieved a mean balanced accuracy of 0.81 (+/-0.04) across five folds, representing a 13% improvement over the state-of-the-art, while remaining clinically transparent. As such, our approach achieves reliable, interpretable, data-driven, operator-independent decision-support tools in Neurology.

Predicting response to induction chemotherapy (IC) and overall survival (OS) is critical for optimizing treatment in patients with locally advanced nasopharyngeal carcinoma (LANPC). This study aimed to develop and validate a multi-task deep learning model integrating pretreatment MRI and whole slide images (WSIs) to predict IC response and OS in LANPC. Pretreatment MRI and WSIs from 404 patients with LANPC were retrospectively collected to construct a multi-task model (MoEMIL) for the simultaneous prediction of early IC response and OS. MoEMIL employed multi-instance learning to process WSIs, PyRadiomics and a convolutional neural network (ResNet50) to extract MRI features, and fused multimodal features through a multi-gate mixture-of-experts architecture. Clustering-constrained attention multiple instance learning and gradient-weighted class activation mapping were applied for visualization and interpretation. MoEMIL effectively stratified patients into good and poor IC response groups, achieving areas under the curve of 0.917, 0.869, and 0.801 in the train, validation, and test sets, respectively, and outperformed the deep learning radiomics model, the pathomics model and TNM staging. The model also stratified patients into high- and low-risk OS groups (P < 0.05). MoEMIL shows promise as a decision-support tool for early IC response prediction and prognostication in LANPC. Author SummaryWe have developed a deep learning model that integrates two types of medical images, including magnetic resonance imaging (MRI) and digital pathological slices, to simultaneously predict response to induction chemotherapy and prognosis in patients with locally advanced nasopharyngeal carcinoma. Current treatment decisions primarily rely on traditional tumor staging (TNM), which often fails to comprehensively reflect the complexity of the disease. Our model, named MoEMIL, was trained and tested on data from 404 patients across two hospitals and consistently outperformed both single-model approaches and TNM staging methods. By identifying patients who exhibit poor response to induction chemotherapy or higher prognostic risk, our tool can assist clinicians in achieving personalized treatment, enabling intensified management for high-risk patients and avoiding unnecessary side effects for low-risk patients. Additionally, we visualize the models reasoning process through heat map generation, which highlights the image regions exerting the greatest influence on prediction outcomes. This work represents a step toward more precise treatment for nasopharyngeal carcinoma; however, larger-scale prospective studies are required before the model can be integrated into routine clinical practice.

Lung cancer screening saves lives, yet uptake remains suboptimal and inequitable. Personalised communication can improve attendance and reduce anxiety, but scaling such support is a workforce challenge. We fine-tuned Googles Gemma 2 9B using QLoRA on 5,086 synthetic screening conversations and compared it against Googles Gemini 2.5 Flash (a larger frontier model) and an unmodified baseline across 300 multi-turn conversations with 100 patient personas spanning ten clinical categories. Evaluation combined automated natural language processing metrics with independent language model judgement in two complementary modes: structured clinical rubric and simulated patient persona. The fine-tuned model achieved the highest simulated patient experience score (3.71/5 vs 3.65 for the frontier model), recorded zero boundary violations after clinician review of all flagged instances, and led on the four most safety-critical categories. A composite Patient Adaptation Index showed that the fine-tuned model led overall (0.37 vs 0.35 vs 0.35), with its clearest advantage on the two clinically specific components: empathy calibration to patient distress and selective smoking cessation signposting. These findings suggest that targeted fine-tuning of open-source models can yield clinical communication quality comparable to larger proprietary systems, with advantages in safety-critical scenarios and suitability for NHS data governance constraints. Human clinician review of these conversations is ongoing.

Background: Current deep learning models in computational pathology, radiology, and digital pathology produce opaque predictions that lack the explainable artificial intelligence (xAI) capabilities required for clinical adoption. Despite achieving radiologist-level performance in tasks from whole-slide image (WSI) classification to mammographic screening, these models function as black boxes: clinicians cannot trace predictions to specific biological features, verify outputs against established morphological criteria, or integrate AI reasoning into precision oncology workflows and tumor board decision-making. Methods: We present Virtual Spectral Decomposition (VSD), a modality-agnostic, interpretable-by-design framework that decomposes medical images into six biologically interpretable tissue composition channels using sigmoid threshold functions - the same mathematical structure as CT windowing. Unlike post-hoc xAI methods (Grad-CAM, SHAP, LIME) applied to black-box deep learning models, VSD channels have pre-defined biological meanings derived from tissue physics, providing inherent explainability without sacrificing quantitative rigor. For whole-slide image (WSI) analysis in digital pathology, we introduce the dendritic tile selection algorithm, a biologically-inspired hierarchical architecture achieving 70-80% computational reduction while preferentially sampling the tumor immune microenvironment. VSD is validated across three cancer types and imaging modalities: pancreatic ductal adenocarcinoma (PDAC) on CT imaging, lung adenocarcinoma (LUAD) on H&E-stained pathology slides using TCGA data, and breast cancer on screening mammography. Composition entropy of the six-channel vector is computed as a visual Biological Entropy Index (vBEI) - an imaging biomarker quantifying the diversity of active biological defense systems. Results: In pancreatic cancer, the fat-to-stroma ratio (a novel CT-derived radiomics biomarker) declines from >5.0 (normal) to <0.5 (advanced PDAC), enabling early detection of desmoplastic invasion before mass formation on standard imaging. In lung cancer, composition entropy from H&E whole-slide images correlates with tumor immune microenvironment markers from RNA-seq (CD3: rho=+0.57, p=0.009; CD8: rho=+0.54, p=0.015; PD-1: rho=+0.54, p=0.013) and predicts overall survival (low entropy immune-desert phenotype: 71% mortality vs 29%, p=0.032; n=20 TCGA-LUAD), providing immune phenotyping for checkpoint immunotherapy patient selection from a $5 H&E slide without molecular assays. In breast cancer, each lesion type produces a characteristic six-channel fingerprint functioning as an interpretable computer-aided diagnosis (CAD) system for quantitative BI-RADS assessment and subtype classification (IDC vs ILC vs DCIS vs IBC). A five-level xAI audit trail provides complete traceability from clinical decision support output to specific biological structures visible on the original images. Conclusion: VSD establishes a unified, interpretable-by-design mathematical framework for explainable tissue composition analysis across imaging modalities and cancer types. Unlike black-box deep learning and post-hoc xAI approaches, VSD provides inherently interpretable, clinically verifiable cancer detection and immune phenotyping from standard clinical imaging at existing costs - without requiring foundation model infrastructure, specialized hardware, or molecular assays. The open-source pipeline (Google Colab, Supplementary Material) enables immediate reproducibility and extension to additional cancer types across the pan-cancer TCGA atlas.

AI documentation frameworks remain poorly designed for point-of-care use, leaving clinicians without actionable information on how to use clinical AI models when they need it most. We developed the Clinician Model Card, an interactive, clinician-centered documentation tool, and evaluated it in a sequential exploratory mixed-methods study: interviews with 12 physicians informed iterative co-design, evaluated in a national survey of 129 physicians across Germany. The tool was well-received: 84% agreed it should be routinely available, and 66% considered its content relevant to clinical decision-making. Yet comprehensibility of statistical performance metrics remained poor despite targeted interventions: only 32% understood the Validation & Performance section well, and fewer than 54% correctly interpreted AUROC or PPV, with AI literacy as strong predictor of comprehension. We propose empirically derived design principles for clinician-centered AI documentation. Effective AI transparency requires not only clinician-friendly design and workflow integration, but sustained investment in AI literacy.

Manuscript preparation is a critical bottleneck in scientific publishing, yet existing AI writing tools require cloud transmission of sensitive content, creating data-confidentiality barriers for clinical researchers. We introduce the Paper Analysis Tool (PAT), a free, multi-agent framework that deploys 31 specialized agents powered by small language models (SLMs) to audit manuscripts across multiple quality dimensions without external data transmission. Applied to three published clinical neurological papers, PAT generated 540 evaluable suggestions. Validation by two expert reviewers (R.B., A.G.) confirmed 391 actionable, high-value revisions (90% agreement), achieving a 72.4% overall usefulness accuracy spanning methodological, statistical, and visual domains. Furthermore, deterministic re-evaluation of 126 agent-suggested rewrite pairs using Phase 0 metrics confirmed text improvement: total word count decreased by 25%, passive voice prevalence dropped sharply from 35% to 5%, average sentence length decreased by 24%, long-sentence fraction fell by 67%, and the Flesch-Kincaid grade improved by 17% . Our validation confirms that systematic, agent-driven pre-submission review drives measurable improvements, successfully converting manuscript optimization from an opaque, manual endeavor into a transparent and rigorous scientific process. Manuscript preparation is a critical bottleneck in scientific publishing, yet existing AI writing tools require cloud transmission of sensitive content, creating data-confidentiality barriers for clinical researchers. We introduce the Paper Analysis Tool (PAT), a free, multi-agent framework that deploys 31 specialized agents powered by small language models (SLMs) to audit manuscripts across multiple quality dimensions without external data transmission. Applied to three published clinical neurological papers, PAT generated 540 evaluable suggestions. Independent validation by two expert reviewers (R.B., A.G.) confirmed 391 actionable, high-value revisions (90% agreement), achieving a 72.4% overall usefulness accuracy spanning methodological, statistical, and visual domains. Furthermore, deterministic re-evaluation of 126 suggested Phase 0 rewrite pairs confirmed text improvement: total word count decreased by 25%, passive voice prevalence dropped sharply from 35% to 5%, average sentence length decreased by 24%, and long-sentence fraction fell by 67%, and the Flesch-Kincaid grade improved modestly. Our validation confirms that systematic, agent-driven pre-submission review drives measurable improvements, successfully converting manuscript optimization from an opaque, manual endeavor into a transparent and rigorous scientific process.

Background: Large Language Models (LLMs) have demonstrated strong performance in medical question-answering tasks, highlighting their potential for clinical decision support and medical education. However, their effectiveness in subspecialty areas such as nephrology remains underexplored. In this study, we assess the performance of open-source LLMs in answering multiple-choice questions from the Nephrology Self-Assessment Program (NephSAP) to better understand their capabilities and limitations within this specialized clinical domain. Methods: We evaluated the performance of five open-source large language models (LLMs): PodGPT which a podcast-pretrained model focused on STEMM disciplines, Llama 3.2-11B, Mistral-7B-Instruct-v0.2, Falcon3-10B-Instruct, and Gemma-2-9B-it. Each model was tested on its ability to answer multiple-choice questions derived from the NephSAP. Model performance was quantified using accuracy, defined as the proportion of correctly answered questions. In addition, the quality of the models explanatory responses was assessed using several natural language processing (NLP) metrics: Bilingual Evaluation Understudy (BLEU), Word Error Rate (WER), cosine similarity, and Flesch-Kincaid Grade Level (FKGL). For qualitative analysis, three board-certified nephrologists reviewed 40 randomly selected model responses to identify factual and clinical reasoning errors, with performance summarized as average error ratios based on the proportion of error-associated words per response. Results: Among the evaluated models, PodGPT achieved the highest accuracy (64.77%), whereas Llama showed the lowest performance with an accuracy of 45.08%. Qualitative analysis showed that PodGPT had the lowest factual error rate (0.017), while Llama and Falcon achieved the lowest reasoning error rates (0.038). Conclusions: This study highlights the importance of STEMM-based training to enhance the reasoning capabilities and reliability of LLMs in clinical contexts, supporting the development of more effective AI-driven decision-support tools in nephrology and other medical specialties.

Background. Federated learning (FL) enables collaborative model training across institutions without sharing patient-level data. However, standard FL algorithms such as FedAvg degrade under non-independently and non-identically distributed (non-IID) data, a prevalent condition when patient demographics, scanner hardware, and disease prevalence differ across hospital sites. Objective. We propose iPS-MFFL (Individualized Per-Site Meta-Federated Feature Learning), a federated framework with a hierarchical local-model architecture that addresses non-IID heterogeneity through (1) a shared feature extractor, (2) multiple weak-learner classification heads that can be trained with heterogeneous training objectives to promote complementary decision boundaries, (3) independent per-learner server aggregation so that each weak learner's parameters are averaged only with its counterparts at other clients, and (4) a lightweight meta-model, itself federated, that adaptively stacks the weak-learner outputs. Methods. We evaluate on the Brain Tumor MRI Classification dataset (7,200 images; 4 classes: glioma, meningioma, pituitary tumor, no tumor) partitioned across K = 5 simulated hospital sites using Dirichlet non-IID sampling (alpha = 0.3). Four baselines are compared: Local-only training, FedAvg, FedProx, and Freeze-FT. All experiments are repeated over three random seeds (13, 42, 2025) and evaluated using paired t-tests, Cohen's d effect sizes, and post-hoc power analysis.