Radu Tanacli

Aims: Risk scores are essential to evidence-based cardiovascular care, but manual calculation is labour intensive and error prone. Large language models (LLMs) could automate this process, yet LLMs are limited by their propensity for calculation errors and factual hallucinations. Pipelines separating LLM-based data extraction from deterministic score computation may improve reliability and transparency. Methods and results: We conducted a retrospective diagnostic study at a quaternary heart centre in Germany (January 2020 to July 2023). Patients with atrial fibrillation (n = 179) from an ablation registry and patients with severe aortic stenosis (n = 76) evaluated by a heart team were included. Six LLMs (GPT-5.2, Gemini 3.1 Pro, DeepSeek-R1, Qwen3, GPT-OSS 120B, and Kimi K2.5) were tested in standalone, retrieval-augmented generation (RAG), and pipeline configurations to compute HAS-BLED, CHA2DS2-VASc, and EuroSCORE II scores from routine clinical reports. Accuracy was assessed against expert-adjudicated ground truth using root mean squared error (RMSE) and Krippendorff’s α to evaluate numerical deviation and categorical agreement, respectively. Pipeline-generated scores showed substantially higher agreement with expert adjudication than standalone LLMs, LLMs with RAG, and treating physicians (mean Krippendorff’s α: 0.78 vs. 0.32 vs. 0.39 vs. 0.31) and lower deviation from ground truth (mean RMSE: 0.89 vs. 5.81 vs. 1.85 vs. 1.34). Conclusion: Pipelines combining expert-curated knowledge injection, LLM-based clinical data extraction, and deterministic score calculation enable accurate and scalable cardiovascular risk score computation from unstructured real-world clinical data, outperforming physician-documented scores. Such pipelines could form the basis for clinical decision-support systems that automate routine risk assessment, reduce clinician workload, and promote more consistent evidence-based care.

Background: Risk scores are essential to evidence-based cardiovascular care, but manual calculation is labor-intensive and error-prone. Large language models (LLMs) could automate this process, yet LLMs are limited by their propensity for calculation errors and factual hallucinations. Pipelines that separate LLM-based data extraction from deterministic score computation may improve reliability and transparency. Methods: We conducted a retrospective diagnostic study at a quaternary heart center in Germany (January 2020 - July 2023). Patients with atrial fibrillation (n=179) from an ablation registry and patients with severe aortic stenosis (n=76) evaluated by a heart team were included. Five LLMs (DeepSeek-R1, Qwen3, GPT-4 Turbo, Llama 3.1, and PaLM 2) were tested in standalone and pipeline configurations to compute HAS-BLED, CHA₂DS₂-VASc, and EuroSCORE II scores from routine clinical reports. Accuracy was assessed by comparing predictions to expert-adjudicated ground truth, using root mean squared error (RMSE), Krippendorff's alpha for categorical agreement, and calibration analysis. Results: Pipeline-generated scores showed substantially higher agreement with expert adjudication than standalone LLMs and treating clinicians (mean Krippendorff's alpha: 0.79 vs 0.32 vs 0.31) and demonstrated superior calibration. The Qwen3-based pipeline, achieved the highest accuracy with lower RMSEs than clinicians for HAS-BLED (0.20 vs 0.87), CHA₂DS₂-VASc (0.53 vs 1.08), and EuroSCORE II (1.99 vs 2.05). Conclusion: LLM-based pipelines enable accurate, well-calibrated, and scalable cardiovascular risk score computation from unstructured real-world clinical data, outperforming clinicians and standalone LLMs with the potential to reduce clinician workload and support evidence-based care.

Background: Catheter ablation (CA) is the leading rhythm control strategy for atrial fibrillation (AF). Despite its effectiveness, AF recurrences, driven by underlying structural and electrical remodeling in both the atria and the ventricle, remain prevalent. Cardiac magnetic resonance (CMR) imaging has emerged as a key modality for assessing myocardial remodeling and potentially identifying predictors of AF recurrence. This study aimed to investigate atrial and biventricular remodeling following CA of AF and its impact on AF recurrence. Methods and Results: Fifty patients with AF (52% male, age 68±10 years) and twenty healthy controls (45% male, age 68±10 years) underwent CMR at least three months following CA. Recurrence of AF was evaluated over 1-year follow-up and defined as at least two unsuccessful attempts to restore SR. Atrial function was measured using phasic volumetric and strain parameters. Ventricular function was evaluated through volumetry, global left ventricular (LV) and right ventricular (RV) strain and LV structure through parametric mapping (native T1/T2 mapping and extracellular volume (ECV)). Patients with AF recurrence at 1 year showed significantly impaired LA active emptying function (14±12% vs 26±14%, P=0.009) and strain (−4.5±4.3% vs −10.6±6.8%, P=0.004), prolongation of both LV T1 (1040±55ms vs 985±43ms, P=0.007) and T2 (53.6±1.6ms vs 50.6±2.5ms, P=0.002) times and elevated LV ECV (28.0±3.0% vs 26.2±1.7%, P=0.027) compared with those without recurrence. A combined model incorporating LA active strain, LV T1 native, ECV and T2 further improved prediction of AF recurrence at 1-year follow-up (HR: 17.70, 95% CI=3.89 - 80.61, log-rank P value −8%, T1 > 1017.3ms, ECV > 27% and T2 > 52.4ms). RV free wall longitudinal strain was more negative in AF non-recurrence group compared with controls but not in the AF recurrence group, suggesting the compensatory importance of RV function in post-procedural hemodynamic recovery. Conclusion: AF recurrence after CA appears linked to persistent atrial dysfunction and ventricular remodelling, potentially fostering a pro-arrhythmic milieu. Incorporating CMR-derived measures into AF care might help improve personalized risk stratification for CA outcomes.

Background: Despite a phenylalanine (Phe) restrictive diet, most adult patients with ‘classical’ phenylketonuria (PKU) maintain life‐long Phe concentrations above the normal range and receive tyrosine (Tyr) and protein‐enriched diets to maintain acceptable concentrations and ensure normal development. While these interventions are highly successful in preventing adverse neuropsychiatric complications, their long‐ term consequences are incompletely explored. We observed early cardiomyopathic characteristics and associated hemodynamic changes in adult PKU patients and present here the results of a longitudinal evaluation of cardiac phenotype. Methods: Fifteen adult patients with PKU (age: 39.8 ± 8.1 years, 9 males and 6 females) underwent a comprehensive follow‐up cardiac magnetic resonance (CMR) imaging assessment after a mean follow‐up interval of 8.3 ± 0.3 years from the initial baseline visit. The CMR protocol included left (LV) and right (RV) ventricular and left atrial (LA) volumetric assessment, LV parametric mapping (precontrast and postcontrast T1 and T2 maps, extracellular volume [ECV]), multilayer LV myocardial strain, systolic and diastolic hemodynamic forces and RV and LA strain and aortic distensibility evaluation. Plasma concentrations of Phe, tyrosine (Tyr) and other biochemical markers of disease were retrospectively collected. For comparison, a group of 20 matched control subjects undergoing an identical CMR protocol was included. Results: On average, the LV end‐diastolic volume (EDV) (158 ± 29 vs. 143 ± 29 mL, p = 0.013) and end‐systolic volume (ESV) (68 ± 18 vs. 62 ± 18 mL, p = 0.011) were lower at follow‐up. In contrast, LV mass (LVM) (72 ± 25 vs. 82 ± 29 g, p < 0.001) and the ratio LVM/EDV (0.46 ± 0.12 vs. 0.58 ± 0.23 g/mL, p = 0.005) were increased, and T1 times were longer (940 ± 42 vs. 1010 ± 35 ms, p < 0.001). LV EF (57 ± 6 vs. 57 ± 7%, p = 0.90), longitudinal (GLS) and circumferential (GCS) systolic strain remained unchanged, but early diastolic hemodynamic (HD) forces were more markedly negative (−19.4 ± 7.0 vs. −26.5 ± 12.2%, p = 0.012), while LA strain 43.8 ± 11.3 vs. 37.3 ± 9.6%, p = 0.031) and aortic distensibility (6.38 ± 1.75 vs. 5.21 ± 1.17 10 ⁻³ mmHg ⁻¹ , p = 0.008) decreased at follow‐up. Compared with controls, PKU patients maintain reduced systolic function with lower LV EF and impaired GCS and have more markedly negative early diastolic HD pressures. A higher decrease in Phe concentration (ΔPhe) was associated with longer T1 times, ΔT1 ( β = −0.78, p < 0.001), increased ECV, ΔECV ( β = −0.61, p = 0.016) and a decrease in systolic function, ΔEF ( β = 0.61, p = 0.017). In contrast, variations in Tyr concentrations did not affect the cardiac phenotype. Conclusions: At long‐term follow‐up, a marked drop in Phe plasma concentration was associated with detrimental cardiac remodelling consisting of decreased LV systolic function and increased diffuse fibrosis, in PKU patients. These new data prompt further investigation into the effects of large Phe variability over time and underline the usefulness of periodic cardiovascular assessment in adults with PKU.

Ongoing clinical trials investigate the therapeutic value of stereotactic cardiac radioablation (cRA) in heart failure patients with ventricular tachycardia. Animal data indicate an effect on local cardiac conduction properties. However, the exact mechanism of cRA in patients remains elusive. Aim of the current study was to investigate in vivo and in vitro myocardial properties in heart failure and ventricular tachycardia upon cRA. High-density 3D electroanatomic mapping in sinus rhythm was performed in a patient with a left ventricular assist device and repeated ventricular tachycardia episodes upon several catheter-based endocardial radio-frequency ablation attempts. Subsequent to electroanatomic mapping and cRA of the left ventricular septum, two additional high-density electroanatomic maps were obtained at 2- and 4-month post-cRA. Myocardial tissue samples were collected from the left ventricular septum during 4-month post-cRA from the irradiated and borderzone regions. In addition, we performed molecular biology and mitochondrial density measurements of tissue and isolated cardiomyocytes. Local voltage was altered in the irradiated region of the left ventricular septum during follow-up. No change of local voltage was observed in the control (i.e. borderzone) region upon irradiation. Interestingly, local activation time was significantly shortened upon irradiation (2-month post-cRA), a process that was reversible (4-month post-cRA). Molecular biology unveiled an increased expression of voltage-dependent sodium channels in the irradiated region as compared with the borderzone, while Connexin43 and transforming growth factor beta were unchanged (4-month post-cRA). Moreover, mitochondrial density was decreased in the irradiated region as compared with the borderzone. Our study supports the notion of transiently altered cardiac conduction potentially related to structural and functional cellular changes as an underlying mechanism of cRA in patients with ventricular tachycardia.

Background: Left ventricular global longitudinal strain (LV GLS) is a superior predictor of adverse cardiac events in patients with myocardial infarction and heart failure. We investigated the ability of morphological features of infarcted myocardium to detect acute left ventricular (LV) dysfunction and predict LV functional recovery after three months in patients with acute ST-segment elevation myocardial infarction (STEMI). Methods: Sixty-six STEMI patients were included in the C-reactive protein (CRP) apheresis in Acute Myocardial Infarction Study (CAMI-1). LV ejection fraction (LVEF), LV GLS, LV global circumferential strain (LV GCS), infarct size (IS), area-at-risk (AAR), and myocardial salvage index (MSI) were assessed by CMR 5 ± 3 days (baseline) and 12 ± 2 weeks after (follow-up) the diagnosis of first acute STEMI. Results: Significant changes in myocardial injury parameters were identified after 12 weeks of STEMI diagnosis. IS decreased from 23.59 ± 11.69% at baseline to 18.29 ± 8.32% at follow-up (p < 0.001). AAR and MVO also significantly reduced after 12 weeks. At baseline, there were reasonably moderate correlations between IS and LVEF (r = −0.479, p < 0.001), LV GLS (r = 0.441, p < 0.001) and LV GCS (r = 0.396, p = 0.001) as well as between AAR and LVEF (r = −0.430, p = 0.003), LV GLS (r = 0.501, p < 0.001) and weak with LV GCS (r = 0.342, p = 0.020). At follow-up, only MSI and change in LV GCS over time showed a weak but significant correlation (r = −0.347, p = 0.021). Patients with larger AAR at baseline improved more in LVEF (p = 0.019) and LV GLS (p = 0.020) but not in LV GCS. Conclusion: The CMR tissue characteristics of myocardial injury correlate with the magnitude of LV dysfunction during the acute stage of STEMI. AAR predicts improvement in LVEF and LV GLS, while MSI is a sensitive marker of LV GCS recovery at three months follow-up after STEMI.

Background: The calculation of extracellular volume (ECV) in cardiac magnetic resonance requires hematocrit, limiting its applicability in clinical practice. Based on the linear relationship between hematocrit and blood T1 relaxivity, a synthetic ECV could be estimated without a blood sample. We aim to develop and test regression models for synthetic ECV without blood sampling in 1.5-T and 3.0-T scanners. Methods: A total of 1101 subjects who underwent cardiac magnetic resonance scanning with native and postcontrast T1 mapping and venous hematocrit within 24 hours were retrospectively enrolled. Subjects were randomly split into derivation (n=550) and validation (n=551) subgroups for each scanner. Different regression models were derived controlling for sex, field strength, and left ventricle/right ventricle blood pool and validated in the validation group. We performed additional validation analyses in subgroups of patients with histological validation (n=17), amyloidosis (n=29), anemia (n=185), and reduced ejection fraction (n=322). Results: In the derivation group, 8 specific models and 2 common estimate models were derived. In the validation group, using specific models, synthetic ECV had high agreement with conventional ECV (R ² , 0.87; P 2 , 0.88, P 2 , 0.88; P 2 , 0.89, P <0.0001; −0.21% and −0.18%, left ventricle and right ventricle model, respectively). Histological validation demonstrated equal performance of synthetic and measured ECV. Synthetic ECV as calculated by the common model showed a bias in the anemia cohort significantly reduced by the specific model (−2.45 to −1.28, right ventricle common and specific model, respectively). Conclusions: Synthetic ECV provided a promising way to calculate ECV without blood sampling. Specific models could provide the most accurate value, while common models could be more suitable in routine clinical practice because of their simplicity while maintaining adequate accuracy.

Background: Phenylketonuria is the most common inborn error of amino acid metabolism, where oxidative stress and collateral metabolic abnormalities are likely to cause cardiac structural and functional modifications. We aim herein to characterize the cardiac phenotype of adult subjects with phenylketonuria using advanced cardiac imaging. Methods and Results: Thirty‐nine adult patients with phenylketonuria (age, 30.5±8.7 years; 10‐year mean phenylalanine concentration, 924±330 µmol/L) and 39 age‐ and sex‐matched healthy controls were investigated. Participants underwent a comprehensive cardiac magnetic resonance and echocardiography examination. Ten‐year mean plasma levels of phenylalanine and tyrosine were used to quantify disease activity and adherence to treatment. Patients with phenylketonuria had thinner left ventricular walls (septal end‐diastolic thickness, 7.0±17 versus 8.8±1.7 mm [ P P =0.004]), more dilated left ventricular cavity (end‐diastolic volume, 87±14 versus 80±14 mL/m ² [ P =0.0178]; end‐systolic volume, 36±9 versus 29±8 mL/m ² [ P P P =0.027]), and lower left ventricular mass (38.2±7.9 versus 47.8±11.0 g/m ² [ P P 1200 µmol/L (909±48 ms). Both mean phenylalanine ( P =0.013) and tyrosine ( P =0.035) levels were independently correlated with T1; and in a multiple regression model, higher phenylalanine levels and higher left ventricular mass associate with lower T1. Conclusions: Cardiac phenotype of adult patients with phenylketonuria reveals some traits of an early‐stage cardiomyopathy. Regular cardiology follow‐up, tighter therapeutic control, and prophylaxis of cardiovascular risk factors, in particular dyslipidemia, are recommended.

This case report describes the contributions of multimodality imaging, cardiac biopsy, and genetic sequencing to the diagnosis and management of heart transplant rejection in a 23-year old patient with dilated cardiomyopathy.