Šakalytė, Gintarė

Vainikinių arterijų liga (VAL) išlieka viena svarbiausių sergamumo ir mirtingumo nuo širdies ir kraujagyslių ligų priežasčių. Vidutinės (50–69 proc.) vainikinių arterijų (VA) stenozės klinikinėje praktikoje kelia sunkumų pasirenkant tolesnę gydymo taktiką, nes anatominis spindžio susiaurėjimas ne visuomet atspindi stenozės funkcinį reikšmingumą ir būsimų nepageidaujamų įvykių riziką. Kompiuterinės tomografijos angiografija (KTA) leidžia įvertinti ne tik stenozės dydį, bet ir aterosklerozinės plokštelės morfologiją bei sudėtį. Šio darbo tikslas buvo įvertinti KTA nustatytų vidutines VA stenozes lemiančių aterosklerozinių plokštelių morfologinius požymius bei jų sąsajas su VAL rizikos veiksniais, invazinių tyrimų rezultatais ir didžiųjų nepageidaujamų širdies ir kraujagyslių įvykių (MACE) išsivystymu, taip pat įvertinti šių požymių prognozinę vertę. Atliktame darbe nustatyti aterosklerozinės plokštelės geometriniai ir sandaros parametrai, susiję su VAL rizikos veiksniais ir invazinių tyrimų rezultatais. Derinant plokštelės morfologinius ir sandaros parametrus, sudarytas patikimas MACE prognozinis modelis. Gauti rezultatai pagrindžia kompleksinį vidutinių VA stenozių vertinimą, apimantį stenozės dydį, kraujagyslės geometriją ir plokštelės sudėtį, siekiant tiksliau individualizuoti pacientų rizikos vertinimą, stebėseną ir gydymo taktikos pasirinkimą.

Širdies nepakankamumas (ŠN) - viena svarbiausių visuomenės sveikatos problemų pasaulyje. Jo paplitimas nuolat didėja dėl gyventojų senėjimo, pažangos gydant širdies ir kraujagyslių ligas bei rizikos veiksnių, tokių kaip cukrinis diabetas, nutukimas, fizinio aktyvumo stoka ir rūkymas, dažnėjimo. ŠN pasaulyje serga apie 64 mln. žmonių, o kasmet Europos šalyse užfiksuojama apie 2,4 mln. naujų atvejų ir daugiau nei 2 mln. hospitalizacijų dėl širdies nepakankamumo paūmėjimo [1]. Širdies nepakankamumas, kai išlikusi išstūmio frakcija, (ŠNilF) diagnozuojamas, kai reiškiasi ŠN požymiai ir simptomai, esant kairiojo skilvelio išstūmio frakcijai (KS IF) ≥ 50 proc., ir yra objektyvių širdies struktūrinių ir / ar funkcinių pokyčių, būdingų kairiojo skilvelio diastolinei disfunkcijai arba padidėjusiam prisipildymo spaudimui, taip pat yra padidėjusi natriuretinių peptidų koncentracija [2].

Vožtuvinės širdies ligos yra dažniausia po išeminės širdies ligos širdies nepakankamumo priežastis ir nekoreguotos lemia progresuojančią širdies perkrovą spaudimu ir tūriu, remodeliaciją, fibrozę bei dekompensaciją. Remiantis širdies nepakankamumo kohortomis, reikšminga vožtuvinė patologija yra nustatoma maždaug kas šeštam širdies nepakankamumu (ŠN) sergančiam pacientui [1], o pažengusio ŠN populiacijoje vožtuvinės patologijos pasireiškia dar dažniau ir gali siekti net iki 40-60 procentų. Šie duomenys rodo ankstyvos širdies vožtuvinių ligų diagnostikos ir intervencijų būtinybę, nes nekoreguotos šios ligos siejamos su didesne stacionarizavimo bei širdinės mirties rizika [2]. 2025 m. Europos kardiologų draugijos (EKD) vožtuvinių širdies ligų gairės dar labiau pabrėžia, kokie svarbūs specializuoti vožtuvinių ligų centrai ir daugiadalykės širdies komandos [5]. įtariant reikšmingą širdies vožtuvo ydą, pacientai turėtų būti siunčiami į vožtuvinių širdies ligų centrus, kuriuose dirba patyrę echokardiografijos, širdies kompiuterinės tomografijos, širdies magnetinio rezonanso, intervencinės kardiologijos, širdies chirurgijos, širdies nepakankamumo ir elektrofiziologijos specialistai.

Background Inter­-individual variability in LDL cholesterol (LDL-­C) reduction after ST-­elevation myocardial infarction (STEMI) remains incompletely explained. Although pharmacogenetic variation has been proposed as a determinant of atorvastatin response, the role of acute inflammatory burden in the setting of STEMI is less well defined. Purpose To evaluate clinical and genetic determinants of LDL-­C reduction 6 months after STEMI. Methods We studied 93 consecutive STEMI patients with lipid assessment at baseline and 6 months. Absolute LDL-­C reduction (ΔLDL = LDL₆M − LDL_baseline) and achievement of ESC guideline targets were analysed. Multivariable linear and logistic regression models included baseline LDL­-C, age, sex, smoking status, statin intensity, and preprocedural white blood cell count (WBC). Comprehensive genetic profiling (>300 lipid­ and pharmacokinetic-­related variants) was assessed using candidate gene scores, genetic burden models, unsupervised clustering, and penalized regression (Elastic Net with nested cross­-validation). Results Median LDL­-C reduction was −1.49 mmol/L (range −4.09 to −0.01). However, 33% of patients failed to achieve ≥30% LDL reduction, 19% showed an absolute reduction of <1.0 mmol/L, and 44% remained above 1.4 mmol/L at 6 months. Baseline LDL­-C was the strongest predictor of LDL reduction (β = −0.511, p < 0.001). Importantly, preprocedural WBC independently predicted an attenuated LDL response (β = +0.069 per 1×10⁹/L increase, p < 0.001), with the model explaining 42.6% of the variability in ΔLDL (R² = 0.426). WBC consistently predicted poor response across all definitions (e.g., OR 1.16 for <30% reduction, p = 0.004). In contrast, none of the genetic approaches improved prediction beyond that provided by clinical variables (all p > 0.60). No stable genetic signature of impaired statin response was identified. Conclusion In STEMI patients, systemic inflammatory burden at presentation independently predicts attenuated LDL-­C reduction at 6 months, whereas extensive genetic variation does not. These findings support a state-­dependent, inflammation-­modulated model of lipid-­lowering response in acute coronary disease.

Dyslipidemia and elevated low-density lipoprotein cholesterol (LDL-C) are major drivers of atherosclerotic cardiovascular disease, yet response to statin therapy varies between patients. We investigated clinical and genetic factors associated with LDL-C response to atorvastatin in patients with ST-elevation myocardial infarction (STEMI). A total of 93 patients underwent targeted next-generation sequencing of a pharmacogene panel, and LDL-C was measured at baseline and after 6 months. The mean ΔLDL-C was − 1.46 ± 0.97 mmol/L. After exclusion of singleton variants, 176 variants remained for analysis, including 68 rare variants (minor allele frequency < 0.05). In single-variant analyses adjusted for age, sex, smoking status, and baseline LDL-C, the CYP2C19 phenotype showed the strongest nominal association with ΔLDL-C, followed by KIF6 rs20455, ABCB1 rs3842, and APOB rs72654423; however, none remained significant after false discovery rate correction. Exploratory aggregate analyses showed associations between the transport pathway score and the weighted polygenic risk score and LDL-C response. In the burden model, CYP2C19 phenotype, transport gene variation, age, and baseline LDL-C were associated with ΔLDL-C. For the binary treatment endpoint, the full model combining clinical and genetic variables showed higher apparent in-sample discrimination than the genetic-only model (AUC 0.829 vs. 0.710; DeLong test p = 0.0044). These exploratory findings require independent validation before clinical application.

Introduction: Hematopoietic stem cell transplantation (HSCT) is an effective treatment option for haematological malignancies (HM), although it involves potentially cardiotoxic therapies. Cardiotoxicity is currently defined mainly by reduction in left ventricular ejection fraction (LVEF) and/or global longitudinal strain (GLS), which may occur after earlier subclinical myocardial changes have already developed [1]. While transthoracic echocardiography is the first-line modality, its sensitivity for early subclinical myocardial changes may be limited. Cardiac magnetic resonance (CMR) enables more comprehensive assessment of ventricular function, deformation and geometry, yet data on CMR-detected subclinical myocardial remodelling after HSCT remains limited [2]. Purpose: To evaluate subclinical myocardial remodelling before and one year after HSCT using CMR-derived functional, deformation and geometric parameters. Methods: Thirty-nine patients with HM undergoing HSCT were prospectively enrolled. CMR was performed before HSCT and at one-year follow-up using a 3T scanner. Left ventricular (LV) and right ventricular (RV) volumes, systolic function longitudinal and circumferential strain, ventricular geometry and mitral valve-LV (MV-LV) coupling parameters were assessed. Statistical analysis was performed using IBM SPSS 29.0. Paired comparisons between baseline and one-year follow-up were performed, with correlation and subgroup analyses based on baseline values, sex and age. The study protocol was approved by our Regional Biomedical Research Ethics Committee. Results: The study population had a median age of 61 years (range 18-74). Of the 39 patients included, 21 were men (53.8%) and 18 were women (46.2%). LVEF, LV geometry, volumes, GLS and circumferential strain did not change significantly. MV-LV coupling analysis demonstrated an increase in MV end-systolic area (p=0.015) and a greater systolic circumferential strain dispersion at the mitral level (p=0.021). Overall mitral annular circumferential strain remained unchanged, but patients with worse baseline values exhibited significantly greater deterioration (p=0.033), indicating heterogeneous MV-LV coupling response. RV remodelling was characterized by a deterioration in RV GLS (p=0.023) and an increase in RV ESL (p=0.039), with basal diameter changes affecting patients with smaller baseline values (p=0.005). Changes in mitral annular circumferential strain correlated with changes in RV GLS (ρ=0.727, p<0.001). Increasing age was associated with greater LV GLS deterioration over time (ρ=0.54, p<0.001). Conclusions: Despite preserved ventricular volumes and global systolic function, CMR revealed early and heterogeneous subclinical myocardial remodelling one year after HSCT, involving MV-LV coupling and RV function. These findings suggest selective vulnerability of myocardial mechanics not detected by conventional functional parameters.

Background: Accurate segmentation of calcified plaques in coronary computed tomography angiography (CCTA) images is critical for reliable assessment of coronary atherosclerotic burden and for supporting interpretation of luminal stenosis, yet it remains a challenging task due to annotation inconsistencies, blooming artifacts and low contrast at lesion boundaries. These limitations may affect both automated model performance and the clinical trustworthiness of AI systems. This study explores the impact of annotation refinement on segmentation performance and model explainability, as well as the influence of representation learning on explanation quality. Methods: We trained a deep convolutional neural network to segment calcified plaques in coronary arteries using a dataset of expert-labeled CT slices. Initial training on radiologist-provided annotations yielded suboptimal results. To address this, annotations were manually revised and validated by radiologists. In addition to a standard ImageNet-pretrained model, we evaluated a self-supervised representation learning approach using DINOv2. Grad-CAM was used to generate visual explanations for model predictions before and after annotation refinement. Results: Models trained with refined annotations achieved notably improved segmentation accuracy, with clearer delineation of calcified regions. Grad-CAM localization analysis demonstrated improved concentration of model attention within plaque and vessel regions. Furthermore, models incorporating DINOv2 representations produced more spatially coherent attention maps, with improved anatomical localization consistent with coronary vessel regions and reduced off-target activations, as qualitatively confirmed by expert radiologists. Conclusions: Our findings emphasize the importance of high-quality, validated annotations in developing accurate and interpretable AI models for medical imaging. In addition, the results suggest that representation learning influences the reliability and clinical relevance of explainability outputs. The combination of manual annotation refinement, expert validation, and improved feature representations provides a practical workflow for human-in-the-loop AI development in cardiovascular imaging. This study demonstrates that annotation quality is a critical and often underestimated determinant of XAI reliability, and suggests that explainability methods can serve as feedback tools for iterative, clinician-guided dataset curation in cardiovascular imaging.

Introduction: Reliable identification of calcified coronary plaques in coronary computed tomography angiography (CCTA) is essential for accurate assessment of atherosclerotic burden and stenosis severity. Plaque delineation remains challenging due to blooming artifacts, low-contrast boundaries, and interobserver variability. Limited transparency of explainable artificial intelligence (XAI) models further hinders clinical adoption. We evaluated whether expert-driven annotation refinement improves segmentation accuracy and interpretability in AI-based calcified plaque detection. Methods: A proprietary CCTA dataset of 47 patients (1,259 axial slices; 1,532 calcified plaque instances) was analysed. Initial annotations were performed by board-certified radiologists. Observed inconsistencies prompted structured manual refinement using predefined anatomical and morphological criteria, followed by independent validation to ensure consistency. A 2D SegFormer-based with an ImageNet pretrained ResNet-101 backbone was fine-tuned on the CCTA dataset using a 75/15/10 train/validation/test split. Performance was assessed with Dice coefficient and lesion-level precision, recall, and F1-score. Interpretability was evaluated using Grad-CAM, with attention maps reviewed for anatomical plausibility. Results: On the independent test set, the refined model achieved a Dice coefficient of 0.732 in this highly challenging small-structure segmentation task, with precision of 0.762, and recall of 0.718. Lesion-level precision, recall, and F1-score were 0.723, 0.773, and 0.733, respectively. Predicted lesion burden closely approximated ground truth (2.34 vs. 2.15 lesions per image), indicating stable and clinically plausible detection performance without systematic over segmentation. Most false positives were located outside coronary vessel boundaries or corresponded to borderline microcalcifications, reflecting ambiguity even at expert level. Grad-CAM maps derived from the segmentation head demonstrated the most anatomically coherent attention patterns, with improved alignment to true plaque locations following annotation refinement and reduced off-vessel attention. Conclusions: In CCTA, annotation quality significantly influences segmentation performance and anatomical plausibility of XAI explanations. Expert-validated refinement enhances model reliability and supports development of clinically interpretable AI tools for automated calcified plaque assessment.

Introduction: Assessment of the left atrial appendage (LAA) on cardiac computed tomography is essential for thrombus exclusion and evaluation of abnormal filling patterns. Differentiating normal LAA from pathological findings such as slow flow or thrombus may require multiphase imaging and expert interpretation, moreover, it can lead to further imaging. We evaluated the performance of a deep learning-based classification pipeline for distinguishing normal LAA from pathological findings using CCTA images. Methods: Following automated LAA segmentation, cropped LAA regions were used to train a convolutional neural network (ConvNeXt) for slice-level classification. The preliminary dataset included 18 patients (9 slow flow, 6 thrombus, 5 normal) as well as publicly available dataset (∼980 cases). Images were categorized into normal or pathological classes based on expert review. A subset of patients was held out for validation to assess generalization to unseen data. Model performance was evaluated using confusion matrix analysis, with emphasis on classification accuracy and error distribution. Misclassified cases were systematically reviewed, demonstrating that a subset of errors reflected borderline or ambiguous findings, with variability even among expert reviewers. Results: The classification model demonstrated high overall accuracy in distinguishing normal LAA from pathological findings. All normal cases were correctly classified in the validation cohort. Thrombus cases were identified with near-perfect accuracy. Misclassification primarily occurred in slow flow cases, which showed partial overlap with thrombus and less often with healthy features on certain slices. Confusion matrix analysis confirmed strong discrimination between normal and abnormal LAA, with minimal false-negative predictions for pathological findings. Preliminary use of a larger backbone network further reduced confusion between slow flow and thrombus categories. Conclusions: Deep learning-based classification of LAA findings on CCTA demonstrates high accuracy for identifying pathological LAA compared with normal anatomy. The model achieved excellent performance in detecting thrombus and preserved strong discrimination between normal and abnormal LAA. With larger validation cohorts, this approach may support automated thrombus screening and enhance workflow efficiency in cardiac CT practice.

Transthyretin amyloid cardiomyopathy (ATTR-CM) has undergone a shift towards noninvasive diagnostics using bone tracer scintigraphy with Perugini grading. While both grades 2 and 3 are considered diagnostic, potential phenotypic differences between these groups remain uncertain. We aimed to evaluate the diagnostic yield of technetium-99 m pyrophosphate (99mTc-PYP) scintigraphy and to compare clinical and multimodality imaging characteristics across scintigraphic grades in a single-center cohort.