Atrial fibrillation (AF) remains a significant health challenge as the number of patients increases with the aging of the population worldwide. AF detection continues to be a significant challenge, particularly in the long term, to improve the quality of life of people suffering from the disease and reduce treatment costs. In recent decades, machine learning and deep learning have provided promising results in classifying heart conditions with moderate interpretability. We hypothesized that the complexity of AF could be handled in low-dimensional latent spaces in terms of non-Gaussian manifolds and that feature selection techniques on an audio-inspired feature set could improve the detection and interpretability of the AF phenomenon. Therefore, we used long-term monitoring databases in which noise-reduction filters preprocessed the signals and extracted their RR-value sequences. From these datasets, several audio spectral characteristics were calculated and used as input vectors for different types of manifold estimators. We also used the least absolute shrinkage and selection operator (LASSO) regression, a feature selection technique. Our experiments showed that these spectral representations of AF segments can yield better-defined low-dimensional embedding manifolds and acceptable intrinsic separability of AF from sinus rhythm. Moreover, including supervised stages and LASSO regression improved the overall performance of this proposal. The performance obtained in classifying AF signals versus normal ones reaches a precision of 91.2%, a recall of 95.5%, and an accuracy of 94.5%. These results pave the way for manifold learning and low-dimensional latent spaces in AF applications, and they support the advantage of using machine learning methods in different AF problems.

The causes of ventricular fibrillation (VF) are not yet elucidated, and it has been proposed that different mechanisms might exist. Moreover, conventional analysis methods do not seem to provide time or frequency domain features that allow for recognition of different VF patterns in electrode-recorded biopotentials. The present work aims to determine whether low-dimensional latent spaces could exhibit discriminative features for different mechanisms or conditions during VF episodes. For this purpose, manifold learning using autoencoder neural networks was analyzed based on surface ECG recordings. The recordings covered the onset of the VF episode as well as the next 6 min, and comprised an experimental database based on an animal model with five situations, including control, drug intervention (amiodarone, diltiazem, and flecainide), and autonomic nervous system blockade. The results show that latent spaces from unsupervised and supervised learning schemes yielded moderate though quite noticeable separability among the different types of VF according to their type or intervention. In particular, unsupervised schemes reached a multi-class classification accuracy of 66%, while supervised schemes improved the separability of the generated latent spaces, providing a classification accuracy of up to 74%. Thus, we conclude that manifold learning schemes can provide a valuable tool for studying different types of VF while working in low-dimensional latent spaces, as the machine-learning generated features exhibit separability among different VF types. This study confirms that latent variables are better VF descriptors than conventional time or domain features, making this technique useful in current VF research on elucidation of the underlying VF mechanisms.

Micro-earthquakes are frequently associated with volcanic activity and are vital indicators of volcanic processes. These minor seismic events occur within or near volcanic systems, yielding valuable insights into subsurface activities. Geologists meticulously record and analyze these events to monitor volcanoes and forecast eruptions. While recent years have seen several studies proposing automated detection and classification systems of seismic events, approaches based on Manifold Learning techniques could be beneficial in terms of information interpretability and transfer learning to other Machine Learning tasks. This paper presents a novel approach employing audio features and psychoacoustic scales to represent micro-earthquakes at Cotopaxi and Llaima Volcanoes, and these representations are then transformed into low-dimensional latent spaces. We implemented a multi-class classification system for events generated by these volcanoes, incorporating feature selection techniques based on audio-inspired features. This approach enhances the detection of volcanic phenomena triggering eruptions and improves interpretability. Our results indicated high accuracy, with rates of 94.44% for Llaima Volcano and 95.45% for Cotopaxi Volcano when utilizing mutual information to select the most relevant features. Spectral Roll-off Point and Spectral Flux dominate in classifying events for both volcanoes. These findings suggest that low-dimensional latent spaces, particularly when utilizing spectral features, can be a promising foundation for developing transfer learning schemes in general, and new multi-class classification systems in particular, for detecting volcanic micro-earthquakes.