Computer Methods and Programs in Biomedicine

A multiple channel QRS detector is described. The detector partitions raw signal segments into peak domains, extracts parameters associated with the peak domains, and scores peaks based on these parameters. A multi-layer perceptron (MLP) with 11 inputs generates provisional peak scores, which are refined through application of rules involving 20-30 parameters. An optimal sequence of supra threshold peaks is determined. Separately, combinatorial optimization determines an optimal structured heart rhythm sequence. Adjudication between the general supra threshold sequence and the structured sequence depends on noise level, peak quality, and rhythm structure quality. For multiple channel fusion, peak scores are determined as a noise weighted function of channel peak scores. The MLP was trained on approximately 70% of channel 1 of the MIT-BIH Arrhythmia Database. The supplementary rules were heuristically chosen over all channel 1 records. Sensitivity (SE) and positive predictive value (PPV) of the detector applied to channel 2 were a function of the noise threshold used to discard segments. At a noise level that would exclude 2.2% of channel 1 data, the SE and PPV were 99.67% and 99.75% respectively. Importantly, even in high noise, the detector was able to track large scale features of heart rhythm. Fused channel 1 and channel 2 SE and PPV were 99.96% and 99.98% respectively. The present algorithm points the way toward maximal extraction of heart rhythm information from noisy signals, and the potential to reduce false alarms generated by automated rhythm analysis software.

Coronary artery disease is a leading cause of morbidity and mortality. Invasive coronary angiography is currently the gold standard in disease diagnosis. Several studies have attempted to use artificial intelligence (AI) to automate their interpretations with varying levels of success. However, most existing studies cannot generate detailed angiographic reports beyond simple classification or segmentation. This study aims to fine-tune and evaluate the performance of a Vision-Language Model (VLM) in coronary angiogram interpretation and report generation. Using twenty-thousand angiogram keyframes of 1987 patients collated across four unique datasets, we finetuned InternVL2-4B model with Low-Rank Adaptor weights that can perform stenosis detection, anatomy labelling, and report generation. The fine-tuned VLM achieved a precision of 0.56, recall of 0.64, and F1-score of 0.60 for stenosis detection. In anatomy segmentation, it attained a weighted precision of 0.50, recall of 0.43, and F1-score of 0.46, with higher scores in major vessel segments. Report generation integrating multiple angiographic projection views yielded an accuracy of 0.42, negative predictive value of 0.58 and specificity of 0.52. This study demonstrates the potential of using VLM to streamline angiogram interpretation to rapidly provide actionable information to guide management, support care in resource-limited settings, and audit the appropriateness of coronary interventions. AUTHOR SUMMARYCoronary artery disease has heavy disease burden worldwide and coronary angiogram is the gold standard imaging for its diagnosis. Interpreting these complex images and producing clinical reports require significant expertise and time. In this study, we fine-tuned and investigated an open-source VLM, InternVL2-4B, to interpret and report coronary angiogram images in key tasks including stenosis detection, anatomy identification, as well as full report generation. We also referenced the fine-tuned InternVL2-4B against state-of-the-art segmentation model, YOLOv8x, which was evaluated on the same test sets. We examined how machine learning metrics like the intersection over union score may not fully capture the clinical accuracy of model predictions and discussed the limitations of relying solely on these metrics for evaluating clinical AI systems. Although the model has not yet achieved expert-level interpretation, our results demonstrate the potential and feasibility of automating the reporting of coronary angiograms. Such systems could potentially assist cardiologists by improving reporting efficiency, highlightning lesions that may require review, and enabling automated calculations of clinical scores such as the SYNTAX score.

Conflicting evidence on scatter correction (SC) methods plagues quantitative myocardial perfusion SPECT (MPI), hindering standardized clinical protocols. This simulation study, utilizing the SIMIND Monte Carlo program and a highly realistic 4D XCAT phantom, systematically evaluates Dual Energy Window (DEW, with k=0.5) and Triple Energy Window (TEW) SC techniques. We uniquely investigate their performance across various photopeak window widths (2, 4, and 6 keV) and novel overlapped/non overlapped configurations specifically for Tc 99m MPI parameters largely unexplored in realistic cardiac models. Images were reconstructed with OSEM under uncorrected (UC), SC, and combined attenuation and scatter corrected (ACSC) conditions. Quantitative analysis focused on signal to noise ratio (SNR), contrast to noise ratio (CNR), defect contrast, and relative noise to background (RNB). Our findings consistently show ACSC's superior performance in CNR, SNR, and defect contrast, confirming its critical role. Interestingly, SC alone reduced noise but compromised defect contrast relative to UC, highlighting a potential trade-off without attenuation correction. Crucially, this study reveals minimal influence of photopeak window width and overlap configuration on image quality, and no significant difference between DEW and TEW across most metrics. These results provide essential evidence for optimizing quantitative MPI protocols, suggesting that for Tc 99m, the choice between DEW and TEW, and specific window settings, may be less critical than ensuring robust attenuation correction.

A Wearable Monitoring System (WMS), comprising a chest patch, wrist-worn pulse oximeter, and arm-worn blood pressure device, was developed in preparation for a pilot Randomised Controlled Trial (RCT) on a UK surgical ward. The system was designed to support continuous physiological monitoring and early detection of deterioration. An initial prototype user interface was developed by the research team based on prior clinical experience and engineering knowledge. To ensure suitability for clinical practice, iterative user-centred refinement was undertaken through a series of clinician focus groups and wearability assessments. Six focus groups were conducted between November 2019 and May 2021 involving multidisciplinary healthcare professionals. Feedback from these sessions informed successive interface and system modifications. System development spanned the COVID-19 pandemic, during which the WMS was rapidly adapted and deployed to support clinical care on isolation wards. Feedback obtained during this period was incorporated into later versions of the system and provided a unique opportunity to examine changes in clinician priorities under pandemic conditions. Clinicians consistently prioritised alert visibility, alarm fatigue mitigation, parameter flexibility, and centralised monitoring. Notably, preferences regarding alert modality and access mechanisms evolved over time: early enthusiasm for mobile or smartphone-type devices shifted towards a preference for fixed, ward-based displays and audible alerts at the nurses station following pandemic deployment. Building on previous wearability testing in healthy volunteers, wearability testing using a validated questionnaire was completed by 169 patient participants during the RCT. The chest patch and pulse oximeter demonstrated high tolerability, whereas the blood pressure cuff showed poor wearability and was removed from the final system. These findings demonstrate the importance of iterative, clinician-led design for wearable WMS and highlight how extreme clinical contexts such as the COVID-19 pandemic can significantly reshape perceived requirements for safety-critical monitoring technologies.

The recently published EHRA/EACVI consensus statement on a standardized bi-atrial regionalization provides new opportunities for consistent regional analyses across patients, imaging modalities and clinical centers. To make this standardized regionalization widely accessible, we developed the open-source software DIVAID, which automatically divides bi-atrial geometries according to the proposed regions, ensuring consistency, reproducibility and operator independence. We evaluated the accuracy of the algorithm by comparing its results to manual expert annotations across 140 geometries from multiple modalities and centers. Veins were automatically clipped correctly in 81% and orifices annotated correctly in 100% of cases. The median (interquartile range; IQR) Dice similarity coefficient (DSC) for left atrial regions was 0.98 (0.96-1.00) for DIVAID-expert and 0.98 (0.94-1.00) for inter-expert comparisons. For right atrial geometries, DSC was higher for DIVAID-expert than for inter-expert comparisons at 0.90 (0.80-0.95) and 0.88 (0.74-0.94), respectively. To assess the accuracy of regional boundaries, we computed the mean average surface distance (MASD) for boundaries derived from automatic or manual annotations. The median (IQR) MASD between DIVAID and experts was 0.17 mm (0.03-0.78) and 1.93 mm (0.65-3.96) in the left and right atrium, respectively. To conclude, DIVAID robustly divides anatomically diverse bi-atrial geometries according to the 15-segment model, while outperforming cardiac experts in both speed and consistency, and demonstrating an accuracy of regional boundaries comparable to the spatial resolution of cardiac imaging modalities. By providing automated, consistent atrial regionalization, DIVAID enables large-scale, standardized regional analyses and data-driven investigation of harmonized, multi-dimensional datasets, which may advance atrial arrhythmia research and personalized treatment strategies.

We are pleased to submit our Original article entitled "Assessing medication-related burden and medication adherence among older patients from Central Nepal: A machine learning approach" for consideration in your esteemed journal. In this paper, we assessed medication burden using validated Living with medicines Questionnaire (LMQ-3) and medication adherence using Adherence to Medication refills (ARMS) Scale. In this paper we analysed our result through machine learning approach in spite of traditional statistical approach to identify the complex factors influencing both. Six ML architectures (Ordinary Least Square, LightGBM, Random Forest, XGBoost, SVM, and Penalized linear regression) were employed to predict ARMS and LMQ scores using various socio-demographic, clinical and medication-related predictive features. Model explainability was provided through SHAP (Shapley Additive exPlanations). Our study identified the moderate medication burden with moderate non-adherence among older adults. Requiring assistance for medication and polypharmacy were the strongest drivers for the medication burden and non-adherence. The high predictive accuracy by ML suggests the appropriate clinical intervention like deprescribing to cope with the high prevalent medication burden and non-adherence among older adults in Nepal.

Background: Medical imaging, especially computed tomography and magnetic resonance imaging, is essential in clinical care of patients with renal cell carcinoma (RCC). Artificial intelligence (AI) research into computer-aided diagnosis, staging and treatment planning needs curated and annotated datasets. Across literature, The Cancer Genome Atlas (TCGA) datasets are widely used for model training and validation. However, re-annotation is often necessary due to limited access to public annotations, raising entry barriers and hindering comparison with prior work. Methods: We screened 1915 CT scans from three TCGA-RCC databases and employed a segmentation model to annotate kidney lesion. After a meta-data-based exclusion step, we hosted a reader study with all papillary (n=56), chromophobe (n=27) and 200 randomly selected clear cell RCC cases. Two students quality checked and corrected the data as well as annotated tumors and cysts. Uncertain cases were checked by a board-certified radiologist. Results: After data exclusion and quality control a total of 142 annotated CT scans from 101 patients (26 female, 75 male, mean age 56 years) remained. This includes 95 CTs with clear cell RCC, 29 with papillary RCC and 18 with chromophobe RCC. Images and voxel-level annotations of kidneys and lesions are open sourced at https://zenodo.org/records/19630298. Conclusion: By making the annotations open-source, we encourage accessible and reproducible AI research for renal cell carcinoma. We invite other researchers who have previously annotated any of these cohorts to share their annotations.

Abstract: Background: Vascular access is essential in treating patients undergoing prolonged endovenous therapy such as chemotherapy, antibiotics, and parenteral nutrition. Since the 1990s, when PICCs (peripherally inserted central catheters) appeared, vascular access options have expanded significantly, revolutionizing the treatment landscape for all types of patients. Objective: To analyze and describe the profile of the use of PICCs in a Brazilian quaternary hospital over 10 years with data collected by the infusion therapy team. Evaluating the number of PICCs implanted over the years, patients epidemiology and clinical characteristics, insertion details, associated complications, and the reason for removal. Methods: A retrospective cohort study that employs a quantitative, non-experimental approach to classify and statistically analyze past events associated with 21,652 PICCs implanted from January 2012 to December 2021 in a quaternary hospital at Sao Paulo - Brazil. All the catheters were implanted, and the data was collected by a team of nurses specializing in infusion therapy. We analyzed the number of catheters implanted over the years, insertion characteristics, patients epidemiology and clinical data, possible associated complications, and the reason for removal. Statistical analyses were conducted using R software (version 4.4.1) and SPSS (version 29) for Windows (IBM Corp, Armonk, NY). Results: During the specified period, 21,652 catheters were analyzed. The patients gender distribution was nearly balanced (48.2% versus 51.8%), and the average age was 66 years. Cardiovascular and metabolic issues were the most common comorbidities, and between 2020 and 2021, 29.3% of the sample tested positive for COVID-19. The most common location of hospitalization and implantation was the medical-surgical clinic (31.6% - 41.4%), and the most used type of catheter was the Power Picc (83.9%). The estimated complication incidence density is 2.94 complications per 1,000 catheter-days. Almost all the PICCs (98,2%) were adequately located at the cavo-atrial junction after the first attempt, 82.2% of catheters were removed after therapy, and the median duration of catheter use was 12 days. Conclusion: PICCs are widely employed for drug infusion, with their use growing progressively due to specialized teams greater availability and training. The high efficiency of these devices with a relatively low risk of complications already observed in previous studies was reinforced by the findings of this study of more than 20,000 catheters.

Background: Even for experienced operators, endovascular treatment of unruptured intracranial aneurysms involves intraoperative uncertainty that may lead to adjustments in strategy, prolong the procedure, and potentially cause inefficiency and device waste. This study aimed to evaluate whether pre-procedural testing (PPT) of endovascular treatment using patient-specific models was associated with increased operator confidence and perceived clinical utility, including improvements in procedural efficiency and reduced resource waste. Methods: We enrolled a cohort of patients who underwent PPT before endovascular treatment for complex unruptured intracranial aneurysms and compared their outcomes with a control group treated without PPT. The primary outcome was the Training Fidelity Score, a composite of three operator-reported Likert items defined a priori. Secondary outcomes included perceived clinical utility, intraoperative strategy changes, procedural time, radiation exposure, device waste and safety. Results: A total of 85 patients met the inclusion criteria (PPT=40; control=45). The Training Fidelity Score was high across the PPT group (median, 4.33/5). Perceived clinical utility was high and further increased significantly after the procedure. A significant reduction was observed in intraoperative strategy changes, with no changes recorded in the PPT group, compared to 6/45 in the control group (RR 0.09; p=0.027). Reductions in treatment time, radiation exposure and device waste were also noted. Conclusion: PPT using patient-specific models was associated with increased operator confidence, fewer intraoperative strategy changes, improved procedural efficiency, and reduced device waste without compromising safety. These findings support its use in pre-interventional preparation, but require prospective multicenter validation.

Background: Increased epicardial adipose tissue volume (EATV) is a potentially important risk marker for coronary artery disease (CAD) available from cardiac computed tomography (CT) images. Sex-differences and effects of age and body size on EATV have been insufficiently explored, and no reliable reference values exist. Consequently, EATV has yet to find its deserved use in clinical practice. Objectives: To define normal values by sex and age, the best normalization procedure for EATV to neutralize effects of body-size, explore the relationship between normalized EATV and cardiac risk, and propose a clinically meaningful cut-off. Methods: AI-based automated EATV data from the general population (n=25,155) and a clinical cohort (n=2,482) with suspected CAD was normalized to height, BSA and heart volumes. Correlation between EATV and EAT attenuation was tested with Spearman?s rank correlation and linear regression to find the optimal normalization. Normalized EATV was compared to high-risk by SCORE2 and obstructive CAD in the population cohort. A cut-off including 95% of cases with obstructive CAD was defined in the general population and tested in the clinical cohort. Results: EATV varied with sex and age across cohorts. Normalization of EATV to total heart volume (EATVh) was superior by all metrics and neutralized the effects of sex. High-risk by SCORE2 and the prevalence of obstructive CAD increased over quartiles of EATVh in the population cohort, and significantly higher EATVh was seen with obstructive CAD in both cohorts. A cut-off of 0.1 in EATVh had a negative predictive value for obstructive CAD of 97.1% in the general population and 88.9% in the clinical cohort. Conclusions: EATV varies considerably with sex, age and body size. Normalization to heart volume outperformed other procedures, and EATVh is a useful marker of obstructive CAD in both the general population and symptomatic patients.

Introduction: The translation of biomarkers into binary clinical decisions requires the determination of precise cut-off points. This study validates the TholdStormDX v0.0.1 tool, a mathematical engine that employs Dual Annealing, 2- and 4-parameter logistic fitting, and vectorized Monte Carlo simulations for panel optimization under Boolean OR logic. Methods: The tool was evaluated using datasets from four diagnostic domains (Pulmonary Nodules, Hepatocellular Carcinoma [HCC], Cervical Cancer, and Breast Cancer), along with a prognosis-oriented analytical context (Breast Cancer). Validation followed a strict workflow: characterization and selection of the best individual and combined thresholds in the Training (Train) and Validation (Val) sets, using the Test set in a completely independent manner, solely to assess the model s performance and generalizability. Results: The tool enabled precise derivation of cut-off points for both individual biomarkers and multivariable combinations. Evaluation on the Test set objectively demonstrated in which scenarios a single biomarker outperforms a complex panel, promoting clinical parsimony. For example, in Breast Cancer diagnosis, an individual predictor outperformed the optimized panel (Sensitivity: 0.953 / Specificity: 0.952 in Test); conversely, in Hepatocellular Carcinoma, the multivariable combination showed superior performance compared to the single marker (Sens: 0.707 / Spe: 0.718 in Test). Additionally, the self-auditing system effectively flagged metric degradation when noisy variables were included, preventing potential issues. Conclusion: TholdStormDX v0.0.1 proves to be a robust and transparent bioinformatics platform for deriving clinical thresholds. Its main contribution lies in mitigating local minima and promoting clinical parsimony, enabling researchers to objectively identify when a single biomarker is sufficient and when a panel provides real added value. Furthermore, it transforms the problem of biological noise into a safety feature: by systematically warning about algorithmic instability, it prevents overfitting and ensures the clinical viability of medical decisions. Availability: The software is free and distributed under the GNU GPLv3 license. TholdStormDX v0.0.1 is written in Python, and its source code is available at the following GitHub address: https://github.com/roberto117343/TholdStormDX.

Optimizing radiotherapy dose distributions remain a resource-intensive bottleneck. Existing AI-based dose prediction methods often have limited generalizability because they rely on small, heterogeneous datasets. We present nnDoseNetv2, an auto-configured, end-to-end framework for dose prediction across diverse disease sites (head and neck, prostate, breast, and lung), prescription levels (1.5-84 Gy), and treatment modalities (IMRT, VMAT, and 3D-CRT). By integrating machine-specific beam geometry with 3D structural information, the framework is designed to generalize across varied clinical scenarios. A single multi-site model was trained on 1,000 clinical plans. On sites seen during training, performance was comparable to specialized site-specific models. On unseen sites (liver and whole brain), the model outperformed site-specific models, with mean absolute errors of 2.46% and 6.97% of prescription, respectively. These results suggest that geometric awareness can bridge disparate anatomical domains while eliminating the need for site-specific model maintenance, providing a scalable and high-fidelity approach for personalized radiotherapy planning.

Background & Aims: Accurate assessment of clinical malnutrition using anthropometric and functional indicators could improve the care of elderly trauma patients in intensive care units (ICUs). This study aimed to develop an AI-driven malnutrition assessment toolbox based on a minimal set of clinically feasible indicators. Methods: Multiple machine learning models, including logistic regression, support vector machines, k-nearest neighbors, decision trees, random forests, XGBoost, and neural-network-based ensemble models, were developed using different indicator configurations from a clinically collected patient dataset. Models were trained using baseline and longitudinal measurements to predict malnutrition risk. SHAP analysis was used to interpret the importance of selected indicators. Results: Baseline (Day 1) data alone did not provide a reliable prediction, whereas longitudinal measurements substantially improved performance. Models based on a minimal indicator set, including bilateral mid-upper arm circumference, calf circumference, and key static variables, outperformed models using the full indicator set. Tree-based methods consistently outperformed linear and distance-based models, with the three-time-point XGBoost achieving the best individual performance. Neural-network-based ensemble models further improved predictive stability. The best overall performance was achieved by the ensemble model using the minimal indicator set from Day 1 and Day 3. SHAP analysis confirmed the importance of the selected indicators. Conclusions: This AI-driven toolbox provides an efficient and clinically feasible approach for early malnutrition assessment in elderly trauma patients in the ICU. Its strong performance with a minimal indicator set supports its potential for integration into clinical workflows and future digital twin systems for intelligent nutritional management.

To evaluate whether hemodynamic responses to acute cognitive stress, measured via pulse-wave harmonic analysis, can characterize cardiovascular regulatory coherence and differentiate older adults with mild cognitive impairment (MCI) from cognitively intact individuals, this exploratory cross-sectional observational study utilized a within-session pre-post cognitive task design. A total of 101 community-dwelling older adults in southern Taiwan were stratified by Montreal Cognitive Assessment (MoCA) scores into Reference (MoCA [&ge;]26, n=12, paired n=10), MCI (MoCA 18-25; n=50, paired n=45), and dementia-level (MoCA <18; n=39) groups, the latter being excluded from task-evoked analyses. The primary outcome was the Harmonic Response Consistency Score (HRCS), quantifying the directional uniformity of cardiovascular regulatory responses, alongside secondary measures of harmonic amplitudes (Cn) and phase angles (Pn). Although mean pre-post changes were subtle, response organization differed by cognitive status. The Reference group exhibited high response consistency (mean HRCS = 9.00), characterized by coordinated harmonic down-modulation. Conversely, the MCI group showed attenuated, directionally heterogeneous responses. Compared to the Reference group, the MCI group demonstrated significantly lower HRCS values for the Cn domain (Mean difference: 2.60, 95% CI 0.29-4.91; p=0.020) and PnSD domain (Mean difference: 1.98, 95% CI 0.04-3.92; p=0.030), indicating a breakdown in regulatory coherence. These findings suggest that acute cognitive stimulus reveals coherent harmonic down-modulation in cognitively intact older adults but fragmented responses in MCI. Pulse-harmonic profiling thus serves as a robust physiological index of cardiovascular regulatory coherence, which, when integrated with neuropsychological assessments, may enhance the sensitivity of non-invasive, community-based screening frameworks for early cognitive aging.

MedSafe-Dx (v0), introduces a new safety-focused benchmark for evaluating large language models in clinical diagnostic decision support using a filtered subset of the DDx Plus dataset (N=250). MedSafe-Dx evaluates three dimensions: escalation sensitivity, avoidance of false reassurance, and calibration of uncertainty. Models were tasked with providing a ranked differential (ICD-10), an escalation decision (Urgent vs. Routine), and a confidence flag. Performance was measured via a "Safety Pass Rate," a composite metric penalizing three hard failure modes: missed escalations of life-threatening conditions, overconfident incorrect diagnoses, and unsafe reassurance in ambiguous cases. Eleven models were evaluated and revealed a significant disconnect between diagnostic recall and safety. GPT-5.2 achieved the highest Safety Pass Rate (97.6%), while several models exhibited high rates of missed escalations or unsafe reassurance. MedSafe-Dx provides a robust stress test for identifying high-risk failure modes in diagnostic decision support and shows that high diagnostic accuracy does not guarantee clinical safety. While the benchmark is currently limited by synthetic data and proxy labels, it provides a reproducible, auditable framework for testing AI behavior before clinical deployment. Our findings suggest that interventions such as safety-focused prompting and reasoning-token budgets could be essential components for the safe deployment of LLMs in clinical workflows.

Backgound: Accurate assessment of diastolic function and left ventricular (LV) filling pressure is central to heart failure diagnosis and risk stratification. Contemporary guideline algorithms rely on complex parameters that are not consistently available in routine clinical practice. Objective: To compare the diagnostic and prognostic performance of the 2016 American Society of Echocardiography/European Association of Cardiovascular Imaging (ASE/EACVI) and 2025 ASE guidelines with a deep learning model based on routinely acquired echocardiographic variables. Methods: This study evaluated the guideline-based algorithms and a deep learning model in participants from the Atherosclerosis Risk in Communities (ARIC) cohort (n=5450) for prognostication and two invasive hemodynamic validation cohorts from the United States (n=83) and Japan (n=130) for detection of elevated left ventricular filling pressure. Results: In the ARIC cohort, the deep learning model demonstrated superior prognostic performance compared with the 2016 and 2025 guidelines (C-index: 0.676 vs. 0.638 and 0.602, respectively; both p<0.001). Similar findings were observed among participants with preserved ejection fraction (C-index: 0.660 vs. 0.628 and 0.590; both p<0.001), with improved performance compared with the H2FPEF score (C-index: 0.660 vs. 0.607; p<0.001). In the US hemodynamic validation cohort, the deep learning model showed higher diagnostic performance than the 2025 guidelines (AUC: 0.879 vs. 0.822; p=0.041) and similar performance compared with the 2016 guidelines (AUC: 0.879 vs. 0.812; p=0.138). In the Japanese hemodynamic validation cohort, the deep learning model outperformed both guidelines (AUC: 0.816 vs. 0.634 and 0.694; both p<0.05). Conclusions: A deep learning model leveraging routinely available echocardiographic parameters demonstrated improved diagnostic and prognostic performance compared with contemporary guideline-based approaches, potentially offering a scalable alternative for assessing diastolic function and left ventricular filling pressures.

Objectives: Large language models (LLMs) have shown promise in creating clinical codelists for research purposes, a time-consuming task requiring expert domain knowledge. Here, we evaluate the performance and assess failure modes of a retrieval augmented generation (RAG) approach to creating clinical codelists for the large and complex medical terminology used by the Clinical Practice Research Datalink (CPRD). Materials & Methods: We set up a RAG system using a database of word embeddings of the medical terminology that we created using a general-purpose word embedding model (gemini-embedding). We developed 7 reference codelists presenting different challenges and tagged required and optional codes. We ran 168 evaluations (7 codelists, 2 different database subsets, 4 models, 3 epochs each). Scoring was based on the omission of required codes, and inclusion of irrelevant codes. We used model-grading (i.e., grading by another LLM with the reference codelists provided as context) to evaluate the output codelists (a score of 0% being all incorrect and 100% being all correct). Results: We saw varying accuracy across models and codelists, with Gemini 3 Pro (Score 43%) generally performing better than Claude Sonnet 4.6 (36%), Gemini 3 Flash, and OpenAI GPT 5.2 performing worst (14%). Models performed better with shorter target codelists (e.g., Eosinophilic esophagitis with four codes, and Hidradenitis suppurativa with 14 codes). For example, all models consistently failed to produce a complete Wrist fracture codelist (with 214 required codes). We further present evaluation summaries, and failure mode evaluations produced by parsing LLM chat logs. Discussion: Besides demonstrating that a single-shot RAG approach is currently not suitable for codelist generation, we demonstrate failure modes including hallucinations, retrieval failures and generation failures where retrieved codes are not used. Conclusions: Our findings suggest that while RAG systems using current frontier LLMs may create correct clinical codelists in some cases, they still struggle with large and complex terminologies and codelists with a large number of codes. The failure mode we highlight can inform the creation of future workflows to avoid failures.

Early detection of breast cancer remains essential for improving clinical outcomes, and complementary non-invasive approaches are needed to support existing screening methods, particularly for women with dense breast tissue. We have previously reported plasma lipid biomarker discovery using untargeted high-resolution liquid chromatography tandem mass spectrometry (LC-MS/MS). In this study, we performed biomarker confirmation and developed machine-learning models applied to targeted plasma lipid measurements for the non-invasive detection of early-stage breast cancer across international cohorts with independent external validation. Targeted LC-MS/MS was used to quantify candidate lipid panels in plasma samples from European discovery cohorts (n = 554) and an independent Australian cohort (n = 266) used for external validation. Data-driven feature selection identified a 15-lipid panel with strong performance in European cohorts (AUC >= 0.94). External validation prior to confidence stratification yielded 76% sensitivity, 64% specificity, and an AUC of 0.81 in the Australian validation cohort. Clinical assay development requires iterative panel and model testing to support translational feasibility and performance in the intended-use population. An analytically viable panel, excluding lipids requiring complex and costly synthesis, achieved comparable accuracy with improved assay robustness. Confidence-based analysis showed enhanced performance for predictions made with moderate to high confidence, with sensitivity up to 89% and AUC up to 0.85, suggesting that ongoing research should focus on strategies to enhance diagnostic model confidence. Importantly, model predictions were independent of breast density, tumour size, grade, subtype, and morphology, indicating biological specificity of the lipid signature. These results demonstrate that calibrated machine-learning models applied to plasma lipid biomarkers can support non-invasive breast cancer detection. Expanding training datasets to include greater diversity will further improve performance in the ongoing development of this lipid-based detection approach.

Interpreting machine learning models typically relies on feature attribution methods that quantify the contribution of individual variables to model predictions. However, it remains unclear whether attribution magnitude reflects the true functional importance of features for model performance. Here, we present a unified interpretability framework integrating permutation-based attribution, feature ablation, and stability under perturbation across multiple feature spaces. Using nested cross-validation and permutation-based null diagnostics, we systematically evaluate the relationship between attribution magnitude and functional dependence in clinical and biomarker-based prediction models. Attribution magnitude is frequently misaligned with functional importance, with weak to strong negative correlations observed across feature spaces (Spearman {rho} ranging from -0.374 to -0.917). Features with high attribution often have limited impact on model performance when removed, whereas features with low attribution can be essential for maintaining predictive accuracy. These discrepancies define distinct classes of interpretability failure, including attribution excess and latent dependence. Interpretability further depends on feature space composition, and stable, functionally relevant features are not necessarily those with the highest attribution scores. By integrating attribution, functional impact, and stability into a composite Feature Reliability Score, we identify features that remain informative across perturbations and analytical contexts. These findings indicate that interpretability does not arise from attribution magnitude alone but is better characterized from stability under perturbation. This framework provides a basis for more robust model interpretation and highlights limitations of attribution-centric approaches in high-dimensional and correlated data settings.

Ambulatory electrocardiograms (ECG) provides continuous monitoring of the hearts electrical activity. However, many existing machine learning and artificial intelligence models for analyzing ambulatory ECG traces are often unimodal and do not incorporate patient clinical context. In this study, we propose a multimodal framework integrating ambulatory ECG-derived representations with clinical text embeddings to predict two cardiac outcomes: sudden cardiac death and pump failure death. Ambulatory ECG traces are preprocessed, segmented, and encoded via a multiple instance learning and temporal convolutional neural network framework. In parallel, patient clinical features are parsed into structured prompts, which are passed through a large language model to generate clinical reasoning; this reasoning passes through a biomedical language encoder to generate a text embedding. With the ECG and text embeddings, we systematically evaluate multiple fusion strategies, including concatenation- and gating-based approaches, to integrate these two data modalities. Our results demonstrate that multimodal models consistently outperform unimodal baselines, with adaptive fusion mechanisms providing the greatest improvements in predictive performance. Decision curve analysis highlights the potential clinical utility of the proposed framework for risk stratification. Finally, we visualize model attention across modalities, including ECG attention patterns, segment-level saliency, heart rate variability features, and clinical reasoning, to contextualize patient-specific predictions.