Manuel Casal-Guisande

This study developed and validated a digital twin of patient flow and waiting list dynamics in a Sleep‐Disordered Breathing Unit, based on a discrete‐event simulation (DES) model in the context of growing diagnostic demand mainly driven by obstructive sleep apnoea. The stochastic model, implemented in MATLAB, reproduces six care stages (e‐Consultation, face‐to‐face consultation, overnight pulse oximetry, respiratory polygraphy, polysomnography and follow‐up consultation), using operational data from year 2024 for parameterisation and data from year 2025 for correlation and validation. Key performance indicators included mean waiting time and mean queue length per stage, with model fidelity assessed via Mean Absolute Error, Root Mean Squared Error and Relative Percent Difference (RPD), both by stage and globally across the full patient pathway. The model accurately reproduced the aggregate system workload, with a weighted RPD of 3.6% for mean queue length and 34.6% for mean waiting time. Respiratory polygraphy showed the best agreement in terms of service load (RPD 2.7%), whereas the follow‐up consultation exhibited the largest discrepancies in waiting times (RPD 73.7%), likely related to prioritisation rules and organisational variability not explicitly modelled. Overall, this digital twin provides an operationally useful representation to support ‘what‐if’ analyses of organisational scenarios aimed at reducing diagnostic delays, representing one of the first validated applications of DES model in a dedicated Sleep‐Disordered Breathing Unit and provides a basis for future applications incorporating urgency stratification and prospective analysis.

The apnoea–hypopnoea index (AHI) remains central to the diagnosis of obstructive sleep apnoea (OSA), but it incompletely captures both clinical heterogeneity and the physiological burden of respiratory events. We propose a phenotype-guided framework with a common initial clinical assessment followed by two complementary pathways according to the patient’s predominant clinical need: a symptom-oriented pathway based on pre-test clinical phenotyping, and a cardiovascular risk-oriented pathway in which directly measured hypoxic burden (HB), complemented by the respiratory event-related pulse-rate response, may eventually support physiological stratification. In our single-centre cohort of 5385 referrals, three predominantly symptomatic phenotypes accounted for 56.5% of patients and two paucisymptomatic phenotypes (A and E) for 43.5%; these groups had overlapping AHI distributions but different clinical profiles and needs. The reported AUC of 0.87–0.95 reflects supervised reproduction of internally derived cluster labels, not prediction of clinical outcomes or CPAP benefit. HB and pulse-rate response remain candidate physiological markers: evidence for treatment-effect modification derives from post hoc analyses and does not establish validated treatment-allocation criteria. Pre-test phenotyping is sufficiently developed to justify prospective, external, multicentre validation, whereas physiological integration requires methodological standardisation and prospective evaluation before individual treatment decisions can be guided by these markers.

Background: Obstructive sleep apnea (OSA) is a highly prevalent chronic respiratory disorder with important consequences for cardiovascular health and quality of life. Its often non-specific clinical presentation contributes to substantial underdiagnosis, while the logistical, technical, and economic complexity of conventional diagnostic procedures can lead to significant delays in diagnosis. Methods: We propose a convolutional neural network (CNN)-based intelligent system for the automatic detection of OSA events from single-lead electrocardiogram (ECG) signals. Using the PhysioNet Apnea-ECG dataset, each ECG time segment was converted into a two-dimensional time–frequency representation through the Continuous Wavelet Transform (CWT), enabling OSA detection to be formulated as an image-classification problem. Three widely used CNN architectures—AlexNet, ResNet-18, and MobileNetV2—were trained using a transfer learning strategy and subsequently integrated through a soft-voting ensemble. The classification threshold was optimized using the Matthews Correlation Coefficient (MCC). Based on the resulting event classifications, the system estimates the patient’s Apnea–Hypopnea Index and assigns the corresponding OSA severity level. Results: The CNN ensemble achieved an AUC of 0.95, an F1-score of 0.87, a sensitivity of 89.17%, and a specificity of 87.06%. Conclusions: Transforming single-lead ECG signals into time–frequency images and combining general-purpose CNN architectures provides an effective, simple, and reproducible approach for automatic OSA detection. The proposed system has been integrated into a graphical clinical decision support interface, providing a basis for future deployment-oriented developments, although external validation will be required before its application in real-world clinical settings.

Objective: To identify distinct clinical groups within the Pi*ZZ and Pi*SZ genotypes using unsupervised Machine Learning (ML) methods applied to a large multicenter national cohort, with the aim of describing disease heterogeneity and exploring potential patterns for future stratification. Methods: A cross-sectional multicentre study was conducted using data from the Spanish population enrolled in the EARCO international registry. Patients with severe deficiency (serum AAT < 11 µM or genotypes Pi*ZZ or Pi*SZ) included between February 2020 and December 2024 were analysed. Clinical, functional, and demographic variables were examined using the k-prototypes clustering algorithm, which integrates continuous and categorical data. Differences between clusters were evaluated with Mann–Whitney U and Chi-square tests ( p < 0.05). Results: A total of 697 patients were analyzed (253 Pi*ZZ, 444 Pi*SZ). Four distinct groups were identified within each genotype. In Pi*ZZ, two groups showed predominant parenchymal lung disease with emphysema and variable obstruction severity, one consisted mainly of middle-aged non-smoking women with bronchiectasis, and another included young, asymptomatic individuals with normal lung function. Among Pi*SZ, the phenotypic spectrum ranged from asymptomatic subjects identified by family screening to patients with advanced COPD, emphysema, and high comorbidity burden. Cardiovascular and metabolic comorbidities contributed to the characterization of disease expression, particularly in Pi*SZ. Conclusions: Unsupervised ML allowed the identification of clinically coherent descriptive profiles within severe alpha-1 antitrypsin deficiency, highlighting the wide heterogeneity of its respiratory manifestations. This stratification may contribute to a more individualized understanding of disease expression and provide a framework for future longitudinal validation.

Background and Objectives: Non-invasive mechanical ventilation is the first-line treatment for managing acute respiratory failure. However, patient variability and complex pulmonary mechanics complicate therapy adjustments, frequently leading to ventilator-induced lung injuries. This study aims to propose and define a simulation platform and decision support prototype, named VENTILA2, to optimize pressure-controlled ventilation strategies. Methods: The system integrates a bicompartmental series model of the respiratory system incorporating severity-stratified physiological profiles of chronic obstructive pulmonary disease and acute respiratory distress syndrome, and it is coupled with a Mamdani fuzzy inference system. This architecture maps inspiratory time adjustments based on pressure errors and their derivatives across predefined clinical profiles within a scenario-based feedforward parameter-mapping framework. Results: Evaluated through quantitative operational verification across all profiles and proof-of-concept case studies, the platform successfully recreates complex clinical scenarios, accurately simulating phenomena such as accelerated lung emptying in severe acute respiratory distress syndrome and air trapping in moderate chronic obstructive pulmonary disease. Conclusions: VENTILA2 provides a controlled simulation environment for evaluating pathology-specific ventilatory configurations across simulated profiles prior to clinical implementation, though it remains an early-stage prototype whose clinical effectiveness, safety, and robustness remain to be rigorously evaluated.

Background/Objectives: Most cases of non-small cell lung cancer (NSCLC) are diagnosed at advanced stages, where prognosis remains poor. Machine learning (ML) offers new opportunities for patient stratification. This study aimed to identify distinct subgroups of patients with advanced-stage NSCLC using unsupervised ML techniques and to evaluate their association with survival outcomes and biomarker expression. Methods: 400 patients with advanced-stage NSCLC were analyzed using the k-prototypes algorithm. Clinical, demographic, and analytical variables, including smoking history and comorbidities, were incorporated. Identified clusters were compared in terms of molecular biomarkers and overall survival. A multivariable Cox proportional hazards model was performed to assess the association between cluster membership and overall survival. Results: Five patient profiles were identified. Cluster F, characterized by a predominance of women, relatively low smoking exposure, and a higher frequency of epidermal growth factor receptor (EGFR) mutations, showed the most favorable survival profile. Cluster S comprised mainly male heavy smokers with poorer performance status and high metastatic burden, whereas Cluster Y consisted predominantly of younger men without comorbidities but with frequent M1c disease. Clusters E and O showed intermediate outcomes and were characterized by older age with pleural effusion and by an older predominantly male smoking profile, respectively. In the multivariable Cox model, compared with Cluster F, a higher risk of death was observed for Cluster S (HR 1.62, 95% CI 1.01–2.60; p = 0.045) and Cluster Y (HR 1.55, 95% CI 1.09–2.21; p = 0.015), although the association for Cluster S should be interpreted cautiously. Differences in molecular biomarker distribution were also observed across clusters, particularly for EGFR mutations and programmed death-ligand 1 (PD-L1) expression. Conclusions: In this single-centre retrospective pilot study, unsupervised ML identified distinct patient profiles associated with differences in survival outcomes and molecular characteristics. These findings support the potential of data-driven approaches to characterize heterogeneity in advanced-stage NSCLC, although external validation is required before clinical application.

Context and objectives: Lung ultrasound (LUS) is a safe and cost-effective diagnostic tool. B-lines are fundamental ultrasound (US) artifacts for LUS-based diagnosis of various pulmonary conditions, especially for the evaluation of the lung parenchyma. However, the challenges in their identification and interpretation, combined with a shortage of experts and training programs, restrict the use of this tool in routine clinical practice. Methods: To overcome these limitations, this study presents a computer vision (CV)-based system for automatic B-line detection. The proposed system integrates preprocessing and feature engineering stages with an object detection module. First, LUS images are normalized to ensure interoperability across different US devices and settings. Subsequently, the Radon and inverse Radon transforms are applied to generate a mask that highlights hyperechoic vertical structures, which is then fused with the preprocessed LUS image. Finally, the resulting image serves as input for a convolutional neural network (CNN) based on the You Only Look Once (YOLO) architecture, enabling the automatic localization of B-lines. Results: The results obtained on the test set demonstrate satisfactory performance, achieving a precision of 89.13%, a recall of 80.39%, and an average precision (AP) of 0.82 at an Intersection over Union (IoU) of 0.5. Conclusions: A clinical decision support tool is proposed, aimed at improving efficiency and consistency in LUS interpretation, as well as facilitating its integration into clinical practice. Although the system is still in its conceptual stage, these findings lay the groundwork for future clinical validation processes directed toward its future implementation.

Background/Objectives: Obstructive sleep apnea (OSA) is a prevalent chronic disorder whose first-line treatment, continuous positive airway pressure (CPAP), is effective only if the patient maintains sufficient adherence. Early predictions of the risk of low adherence would make it possible to personalize follow-up and optimize healthcare resources. The aim of this study was to develop and evaluate a machine-learning-based clinical decision support system to predict CPAP adherence after one year of treatment. Methods: A cohort of 200 patients with OSA from the Sleep-Disordered Breathing Unit of Hospital Álvaro Cunqueiro in Vigo was used. The cohort was split into a training set (n = 160) and an independent test set (n = 40). Two scenarios were defined: Scenario A, with pre-treatment variables, and Scenario B, which also includes early adherence metrics. In each scenario, variables were selected through recursive feature elimination. The selected variables were apnea-hypopnea index (AHI), 3% oxygen desaturation index (ODI3%), chronic obstructive pulmonary disease and neck circumference in Scenario A, and first-month adherence, ODI3% and AHI in Scenario B. Once these subsets were defined, several classifiers were analyzed. Results: Random Forest was the model selected in both scenarios. On the test set, Scenario A reached an area under the curve (AUC) of 0.71 (sensitivity 0.83; specificity 0.45) and Scenario B an AUC of 0.91 (sensitivity 0.90; specificity 0.82). Conclusions: Early prediction of CPAP adherence using machine learning is feasible; incorporating real first-month use markedly improves discriminative ability. The system was integrated into a web prototype as a proof of concept. Given the modest sample size and the absence of external validation, these results should be interpreted as preliminary, corresponding to an exploratory, feasibility study. For future implementation, an extensive clinical validation process will be required, along with the expansion of the database, which will likely contribute to improving the system’s robustness and generalization capacity.

Context and objectives: Lung ultrasound (LUS) is a safe low-cost tool that enables diagnosis, monitoring and guidance for interventional procedures at the patient's bedside. However, its expansion is hindered by a lack of training programs and the inherent difficulty of interpreting ultrasound images. In this context, Artificial Intelligence (AI) is emerging as a supportive tool for LUS interpretation, ensuring diagnostic efficacy and mitigating the shortage of experts. This systematic review aims to summarize and analyze recent advances in AI-based tools to support LUS interpretation. Methods: A systematic literature search was conducted across Web of Science, IEEE Xplore, and PubMed databases to identify peer-reviewed original journal articles published between 2015 and November 2025 that employed AI for the identification and localization of lung artifacts, anatomical structures, and pathological findings. Risk of bias was assessed using PROBAST + AI. Results: Twenty-four studies were included, identifying three main strategies: segmentation (10 studies), object detection (4 studies), and the generation of visual explanations through saliency maps (10 studies). All employed CNN-based architectures. The evaluation metrics used were heterogeneous. The PROBAST + AI assessment showed relevant risk-of-bias concerns, mainly concentrated in the participants and analysis domains. Conclusions: The development of AI systems to support LUS interpretation shows high potential; however, current studies exhibit significant heterogeneity in their objectives, methodologies, and evaluation metrics. It is necessary to move towards solutions designed for specific clinical environments and to adopt standardized protocols and evaluations that facilitate their implementation in clinical practice. Systematic Review Registration: https://www.crd.york.ac.uk/PROSPERO/view/CRD420261322517 , PROSPERO CRD420261322517.

Biomedical innovation often emerges from the interaction between clinical needs and technological expertise