Unsupervised vector quantization for robust lung state estimation of an EIT image sequence

Every year, several ten thousand patients die on mechanical ventilation. This happens because the lungs can currently not be monitored adequately in real-time, and thus suboptimal ventilator settings can cause severe lung tissue damage. Electrical Impedance Tomography (EIT) produces a real-time image sequence of the breathing lungs. So far, no automatic method has been available to detect the physiological regional lung states. We propose an algorithm that clusters the raw pixel-based data of the EIT image sequence into clinically relevant regions with similar physiological behavior. Our implementation is very robust regarding bad signal quality due to low signal to noise ratio (SNR). It is also highly efficient in terms of computational complexity by considering additional physiological knowledge. The functionality of the algorithm has been verified using EIT data of a human subject with acute lung failure at various Positive End-Expiratory Pressure (PEEP) levels. The results are in agreement with the study protocol. This method brings EIT treatment one step closer towards protective ventilation therapy.