Accurate segmentation of ventilator waveforms is essential for detecting patient-ventilator asynchronies (PVAs), yet current heuristic methods can fail in noisy, real-world data. We developed and validated a deep learning model using a one-dimensional attention-gated U-Net architecture to identify inspiratory and expiratory onsets in mechanical ventilation waveforms. The model was trained and tested on 9719 breaths from 33 patients and outperformed published rule-based methods, achieving F1 scores of > 0.99 for both inspiratory and expiratory onset detection within a 0.1-s tolerance window. Performance remained robust in asynchronous breaths (F1 ≥ 0.98). When applied to quantify PVAs, the model reproduced reference standard asynchrony frequencies with no significant differences, whereas heuristic methods produced large deviations. Gradient-weighted class activation maps suggest that the model leveraged a diverse set of waveform features to inform segmentation. This computationally efficient model enables highly-accurate, clinically timely waveform analysis and provides a foundation for scalable, reproducible assessment of ventilator-patient interactions.

Respiratory-swallowing coordination is critical for safe swallowing. Respiratory phase patterns, indicating the direction of respiration bracketing swallowing, appear minimally altered when oral airflow is measured alongside nasal airflow. Measures from an oronasal facemask also demonstrate broader features of swallowing non-respiratory flow (SNRF) associated with the respiratory-swallowing pause than previously reported. Additionally, nasal and oronasal flow estimates show variability in within-individual respiratory phase patterning. In this study, an oronasal facemask and separate nasal-only mask were used to assess healthy individuals, exploring how instrumentation influences respiratory phase pattern estimates across swallowing tasks. Characteristics of SNRF and variability of respiratory phase patterns were also investigated. Adding oral flow to nasal flow appeared to reduce estimated exhale-swallow-exhale patterning by 13% and increased the estimated frequency of alternative patterns, although this may include some genuine variance between the datasets due to asynchronous data acquisition from each mask. SNRF occurred in 87-97% of swallows, with inward and outward flow occurring at the onset and offset of the respiratory-swallowing pause. Test-retest reliability of respiratory phase patterns within individuals ranged from κ = 0.4 - 0.54, with ≥2 patterns observed across four trials >40% of the time. Findings suggest that adding oral to nasal flow alters estimates of respiratory phase patterns, particularly during ingestive swallowing tasks. SNRF is observed more frequently when measured through a closed system, presenting as inwardly- or outwardly-directed airflow at the beginning or end of the respiratory-swallowing pause. Finally, there appears to be considerable variability in respiratory phase patterning across trials of the same swallowing condition.

Patient-ventilator asynchrony is highly prevalent during invasive mechanical ventilation, yet its detection at the bedside remains limited. Conventional waveform inspection is intermittent, operator-dependent, and insufficient to capture the complexity and temporal variability of patient-ventilator interaction. Automated systems based on advanced signal processing and artificial intelligence represent a paradigm shift, enabling continuous, objective, and scalable detection of asynchrony. Recent approaches highlight the value of entropy-based metrics to quantify the irregularity of ventilatory signals and capture subtle changes in respiratory variability. Similarly, the identification of asynchrony clusters, periods where multiple asynchronous events occur in succession, provides a clinically relevant framework to stratify severity and predict outcomes. These developments underscore the superiority of automated methods over human observation alone. From an implementation perspective, centralized monitoring architectures offer greater potential than stand-alone devices, as they allow multimodal integration, data aggregation, algorithm refinement, and interoperability across platforms. Looking forward, research must expand toward the fusion of ventilatory, hemodynamic, and neurological signals to provide a comprehensive picture of patient-ventilator interaction. Particular attention should be paid to the long-term consequences of poor synchrony, as evidence suggests that asynchronies may influence functional recovery and health-related quality of life well beyond intensive care unit discharge. Robust, standardized datasets will ultimately support the generation of synthetic data and patient-specific digital twins, paving the way for precision-guided, adaptive mechanical ventilation.

Objective.Quantification of ventilation inhomogeneity using electrical impedance tomography (EIT) typically relies on accurate identification of breathing cycles, which is often unreliable in spontaneously breathing patients. The objective of this study was to develop a robust, breath-independent metric for characterizing temporal ventilation heterogeneity.Approach.We propose a pixel asynchrony value (PAV), a window-based temporal correlation measure that quantifies the asynchrony between local pixel waveforms and a global ventilation reference without requiring breath segmentation. A global summary index, the global asynchronous index (GAI), is derived from spatial PAV distributions. The method was evaluated using EIT recordings from 21 high dependency unit patients acquired before and after airway clearance therapy.Main results.GAI demonstrated a consistent and significant reduction following treatment (p = 0.0011), indicating improved temporal synchrony of regional ventilation. In contrast, conventional inhomogeneity indices, including the global inhomogeneity index and the standard deviation of regional ventilation delay, showed weaker or inconsistent changes. Robustness analysis further showed that GAI remains stable across a range of window lengths and is insensitive to the absence of explicit breath-cycle detection.Significance.The proposed PAV-based GAI provides a physiologically interpretable and robust measure of temporal ventilation heterogeneity that can be applied without breath segmentation, making it particularly suitable for spontaneously breathing patients and routine clinical monitoring.