Deep Learning-Enabled Robust Regional Lung Function Assessment in Mechanically Ventilated AECOPD Patients Using EIT Parameters

OBJECTIVE: Reliable quantification of regional lung function via electrical impedance tomography (EIT) in mechanically ventilated AECOPD patients requires identifying breaths suitable for steady-state regional parameter estimation and excluding breaths deviating from local steady state. However, automated screening remains difficult because existing methods primarily rely on inaccessible raw voltage data or fail to distinguish complex asynchrony patterns in reconstructed images. Approach. We developed an EIT robustness evaluator (EITRE), a spatio-temporal deep learning framework to identify breaths suitable for steady-state regional parameter estimation directly from reconstructed EIT image sequences. The architecture integrates an EfficientNet-b0 spatial encoder with an attention-augmented gated recurrent unit (GRU) and an XGBoost classifier to capture complex regional impedance dynamics. The framework's performance was evaluated in a multicenter validation protocol using operational reference labels derived from synchronized ventilator waveforms and a statistical deviation rule. The practical utility of EITRE was further demonstrated in a clinical positive end-expiratory pressure (PEEP) titration experiment to evaluate its effect on parameter-derived PEEP recommendations. Main results. Validated on a multi-center dataset comprising 58 AECOPD patients and 140,310 respiratory cycles, EITRE achieved robust cross-domain performance and generalizability (F1-score 0.933 on unseen data following single-subject fine-tuning). Quantitative analysis revealed diverse impacts of PVA on EIT parameters: parameters assessing temporal ventilation heterogeneity exhibited higher sensitivity and substantial deviations during PVA, whereas parameters assessing spatial ventilation heterogeneity remained relatively robust. In a PEEP titration experiment, EITRE-based screening produced indicator-and patient-dependent shifts in EIT-derived PEEP recommendations relative to a baseline statistical screening rule, supporting improved measurement consistency for steady-state parameter estimation rather than a uniform population-level correction. Significance. EITRE provides a specialized tool for supporting breath-by-breath measurement consistency for steady-state EIT parameter estimation, enabling more robust regional lung-function assessment in mechanically ventilated AECOPD patients.