This study explores the potential of unsupervised machine learning algorithms to identify latent cardiac risk profiles by analyzing ECG-derived parameters from two general groups: clinically healthy individuals (Norm dataset, n = 14,863) and patients hospitalized with heart failure (patients’ dataset, n = 8220). Each dataset includes 153 ECG and heart rate variability (HRV) features, including both conventional and novel diagnostic parameters obtained using a Universal Scoring System. The study aims to apply unsupervised clustering algorithms to ECG data to detect latent risk profiles related to heart failure, based on distinctive ECG features. The focus is on identifying patterns that correlate with cardiac health risks, potentially aiding in early detection and personalized care. We applied a combination of Uniform Manifold Approximation and Projection (UMAP) for dimensionality reduction and Hierarchical Density-Based Spatial Clustering (HDBSCAN) for unsupervised clustering. Models trained on one dataset were applied to the other to explore structural differences and detect latent predispositions to cardiac disorders. Both Euclidean and Manhattan distance metrics were evaluated. Features such as the QRS angle in the frontal plane, Detrended Fluctuation Analysis (DFA), High-Frequency power (HF), and others were analyzed for their ability to distinguish different patient clusters. In the Norm dataset, Euclidean distance clustering identified two main clusters, with Cluster 0 indicating a lower risk of heart failure. Key discriminative features included the “ALPHA QRS ANGLE IN THE FRONTAL PLANE” and DFA. In the patients’ dataset, three clusters emerged, with Cluster 1 identified as potentially high-risk. Manhattan distance clustering provided additional insights, highlighting features like “ST DISLOCATION” and “T AMP NORMALIZED” as significant for distinguishing between clusters. The analysis revealed distinct clusters that correspond to varying levels of heart failure risk. In the Norm dataset, two main clusters were identified, with one associated with a lower risk profile. In the patients’ dataset, a three-cluster structure emerged, with one subgroup displaying markedly elevated risk indicators such as high-frequency power (HF) and altered QRS angle values. Cross-dataset clustering confirmed consistent feature shifts between groups. These findings demonstrate the feasibility of ECG-based unsupervised clustering for early risk stratification. The results offer a non-invasive tool for personalized cardiac monitoring and merit further clinical validation. These findings emphasize the potential for clustering techniques to contribute to early heart failure detection and personalized monitoring. Future research should aim to validate these results in other populations and integrate these methods into clinical decision-making frameworks.

Дана стаття націлена на аналіз та вдосконалення методів попередньої обробки електрокардіосигналів для задачі виявлення низькоамплітудних регулярних компонент. В даному дослідженні було проаналізовано основні переваги та недоліки існуючих підходів до попередньої обробки ЕКГ сигналу для задачі виявлення пізніх потенціалів шлуночків та передсердь. На основі проведеного аналізу, з метою підвищення точності виявлення пізніх потенціалів різними алгоритмами, та покращення якості попередньої обробки ЕКГ сигналу для виявлення низькоамплітудних компонент, було запропоновано удосконалений метод усереднення кардіоциклів. Головною особливістю запропонованого методу є розбиття великої кількості кардіоциклів для усереднення на менші сукупності (епохи), та подальше застосування лінійного матричного розкладання для придушення нерегулярних включень. Також за необхідності у разі розбиття на епохи можна використовувати перекриття. Це дозволить зменшити розбіжності між епохами, та збільшити кількість кардіоциклів. Використання даного підходу дає можливість мінімізувати нерегулярні включення в ЕКГ сигнал та підвищити точність виділення низькоамплітудних пізніх потенціалів. Крім того, розбиття на епохи та перекриття дозволяє уникати розмиття низькоамплітудних високочастотних компонент під час усереднення в результаті варіабельності серцевого ритму, а також покращувати якість усередненого сигналу при використанні зменшеної кількості кардіоциклів. Для тестування запропонованого методу використовувались різні підходи для оцінки попередньої обробки ЕКГ сигналу. Переважно було проведено порівняння кардіоциклів, які були отримані в результаті роботи різних алгоритмів усереднення та запропонованоого методу, з еталоном. Для тестування методу усереднення, було розроблено штучний ЕКГ сигнал з наявним шумом, наявними пізніми потенціалами шлуночків та передсердь, варіабельністю серцевого ритму, а також з високоамплітудною компонентою, яка виникає у випадковому місці через кожні два серцеві скорочення. Еталонний кардіоцикл було отримано з вихідного штучного сигналу без наявних спотворень та шуму. По перше, було проведено візуальне порівняння та оцінка різних методів усереднення з еталоном. По друге, було розраховано метрики подібності пізніх потенціалів на усередненому кардіоциклі з пізніми потенціалами на еталонному сигналі. На основі даних метрик було побудовано криві залежностей значень метрик подібності від амплітуди пізніх потенціалів на ЕКГ сигналі. По третє, було проведено оцінку впливу запропонованого методу усереднення на результати класифікації реальних ЕКГ сигналів з наявними пізніми потенціалами за допомогою різних алгоритмів машинного навчання. Загальний результат тестування показав, що запропоний метод усереднення за результатами дослідження здатний відтворювати морфологію низькоамплітудних регулярних складових на 10-30% точніше, та покращувати точність класифікації на 5-12%.

The article is focused on the development of Siamese neural network models for ECG signals classification that reflect cardiovascular pathologies, including arrhythmias, in the context of a limited amount of training data. The problem of a shortage of training data in machine learning for diagnosing heart disease is associated with a wide variety of pathological states and insufficient information for certain classes in open medical databases. The study aims to develop a complex method based on the combination of high-resolution electrocardiography and vectorcardiography with Siamese neural network architectures and training methods, which makes it possible to improve the accuracy of cardiac arrhythmias classification. The peculiarity of the proposed method, which is based on the ability of Siamese NNs to compare, is to detect and analyse the differences between the ECG signal under study and the generated reference feature vector of signals with pathology, which allows to effectively identify signal changes even for those diseases that are limitedly represented in the training dataset. In addition, to improve the efficiency of training, a method for generating a reference feature vector for pathological signals was developed. This vector is used by the Siamese neural network for comparison. The application of the principal component analysis (PCA) method allowed to extract key features from 100 ECG signals with pathologies, which contributed to the creation of a reference feature vector with a minimum number of training samples. Additionally, for each input ECG signal and reference feature vector, an average cardiac cycle was calculated, which helped to identify low-amplitude ECG components and features of the QRS complex. To implement the developed complex method, the PTB-XL database was used, which contains 12-channel ECG records classified into 70 disease categories. To reduce the impact of data imbalance, augmentation methods, as well as preprocessing methods were used to remove noisy signals and selectively reduce overrepresented classes. Two models of Siamese neural networks (NN) were developed as part of the study. The first model is focused on detecting low-amplitude pathological components of ECG signals, in particular, late atrial and ventricular potentials. The second model is designed to classify 18 types of arrhythmias and 19 associated pathologies, such as coronary heart disease, hypertrophy, and myocardial infarction. The effectiveness of the proposed NN specialised ECGnet network models was evaluated by comparing them with the specialised ECGnet network in the task of recognising late atrial and ventricular potentials. The first model exceeded the accuracy of ECGnet by an average of 10% and reduced the probability of false negative predictions. The second NN model for multi-class classification, which covered 37 diagnostic classes with rare diseases with less than 200 observations, exceeded the average accuracy of ECGnet by 10%, reaching a maximum increase of 28%. The obtained results allow to outline further ways to improve the complex method. In particular, improving the accuracy of ECG signal classification with pathologies is possible by using additional transformations of input features and methods of amplifying low-amplitude signal components.

This study describes the development of a model for detection erroneously identified cardiac cycles as a classification algorithm that is capable of differentiating the signal of QRS complexes from noise. To solve this task, about 8 public databases were analyzed. From these databases 13,712,521 normal cardiac cycles, as well as 678,641 cardiac cycles of various types were prepared. This amount of data was chosen to increase the generalization ability of the trained models and the ability to correctly work with cardiac cycles of various shapes and types. In this study 2 neural network architectures were developed - a fully connected and a deep recurrent neural network. The new loss function was proposed, which minimizes the false positive rate, that is, the probability of identifying a noisy signal as a QRS complex. This function also showed an improvement in the learning speed and also significantly reduced the problem of the algorithm falling into local minima. The deep convolutional neural network was able to achieve higher classification accuracy than the fully connected one. The classification accuracy of the convolutional neural network was 99% with a false positive rate of 1%. The trained models have been used to solve various tasks that require accurate determination of QRS complexes. Thus, the proposed models were able to improve the quality of detection of cardiac cycles by 5-70%, and were also able to improve the quality indicator for identifying low-amplitude late potentials by approximately 20%, reducing the dynamic time warping distance from 0.72 to 0.58 with a reference value of 0.52.

The paper considers the actual problem of improving neural network models for the classification of cardiovascular pathologies by compressing the information contained in electrocardiographic (ECG) signals. Due to the active implementation of artificial intelligence in medical diagnostics, the study focuses on creating a reduced ECG feature vector that allows for a significant reduction in training data volume without losing important diagnostic information. The research is dedicated to the development of an autoencoder model with a specialized architecture that combines convolutional and fully connected layers, attention layers, residual connections, and a symmetric structure with shared weights. This approach enables not only to compress the input multichannel signal but also to form a latent space from which the signal can be restored or used as a feature vector for classification. In the proposed autoencoder architecture, the latent representation is combined with the layer weight vector into a single reduced vector that contains information about both the shape and the structural features of the ECG signal. The model was trained using a composite loss function, which allows for balancing between signal reconstruction quality and classification accuracy. To evaluate the model, several alternative dimensionality reduction methods were used, including downsampling, transformation to Frank's orthogonal leads, calculation of a resultant vector, and principal component analysis (PCA). A number of alternative dimensionality reduction methods were used to test the model: reduction of the sampling frequency; transformation to Frank orthogonal derivations; calculation of the resulting vector; principal component analysis (PCA). All methods were tested on the PTB-XL dataset, which contains 12-lead ECG recordings with a wide range of cardiovascular pathologies. The effectiveness of the reduced feature vector was assessed using two models—a convolutional neural network (EcgNet) and a fully connected network with two hidden layers. The analysis of the results showed that, on average, classification accuracy using the reduced vector decreased by only 2% for EcgNet and increased by 3–6% for the fully connected network, indicating preservation of diagnostic information even with a high compression ratio of 25. In contrast, traditional dimensionality reduction methods such as PCA and orthogonal transformation showed a significant deterioration in classification quality (up to –16%). While downsampling to 75 Hz significantly reduces data volume, it also leads to the loss of high-frequency information critical for detecting ischemia and arrhythmias. Particular attention is given to ventricular and atrial late potentials, which are low-amplitude and high-frequency in nature. The constructed reduced feature vector preserved the informative characteristics of these classes, resulting in improved classification accuracy compared to traditional dimensionality reduction methods. Thus, the developed reduced feature vector demonstrates the ability to retain diagnostic information under significant ECG data compression, achieving a balance between computational efficiency, classification accuracy, and application versatility. The obtained results suggest that the use of the proposed reduced feature vector may serve as a promising solution for compact and efficient systems for automated ECG analysis, particularly in mobile or computing-limited systems, while also enabling faster development and testing of new neural network models for diagnostics.