Exploratory Analysis of Electrical Impedance Tomography Features Associated with Spirometric Airflow Limitation

Introduction: Electrical impedance tomography (EIT) is a noninvasive, radiation-free imaging modality that provides real-time information on regional lung ventilation from wearable sensors. Chronic obstructive pulmonary disease (COPD) is a highly prevalent respiratory disease causing persistent and progressive airway obstruction. The diagnosis is based primarily on spirometry, in which a Tiffeneau index (forced expiratory volume in 1 s/forced vital capacity ratio) below 0.7 is a key feature of the definitive diagnosis. Despite many advances in medicine, there is a lack of widely available methods for estimating airway obstruction using noninvasive bedside measurements, which could facilitate assessment in patients who are unable to perform standard spirometry, e.g., patients after laryngectomy and those with chronic tracheostomies. Moreover, the prevalence of COPD in this group of patients may be substantial. Methods: The study examined the relationship between EIT-derived signal features and the Tiffeneau index through a comprehensive statistical and machine learning analysis of patient data collected using the Dräger EIT system. Data from 15 adult patients who successfully completed conventional spirometry and EIT measurements were analyzed. Patients conducted the examination a couple of times; consequently, 30 measurements were collected. A total of 755 time-series features, complemented by physiological measurements, were extracted, analyzed, and evaluated using correlation metrics, P-value testing, categorical associations, and Shapley additive explanations explainability. Results: Frequency-domain features (fast Fourier transform angle coefficients), entropy measures, and autocorrelation-based descriptors have the strongest associations with the Tiffeneau index. Feature-selected machine learning models demonstrated that EIT-derived time-frequency features, combined with anthropometric variables (weight and body mass index), could approximate Tiffeneau index values in this small cohort (best model: $R^2=0.72$ on the testing set). Conclusion: These findings support the exploratory feasibility of EIT-based, noninvasive approaches for pulmonary function estimation in a general clinical cohort and motivate future validation studies in patients for whom spirometry cannot be performed.