Automated Detection of Pulmonary Edema and Respiratory Failure Using Physiological Signals

In this paper we explore Computer-Aided Diagnosis (CAD) methods for the differentiation between pulmonary edema and respiratory failure. The CAD methods were constructed with feature extraction and classification algorithms. The feature extraction was based on Cross-Recurrence Quantification Analysis (CRQA) and feature ranking ensured that only discriminative features were used for classification. The classification tests were conducted with ten fold stratified cross-validation for Support Vector Machine (SVM) and Artificial Neural Network (ANN). With this experimental setup, we have conducted a competitive study of seven different physiological signals: Mean Arterial Blood Pressure (MABP), Systolic Blood Pressure (SBP), Diastolic Blood Pressure (DBP), Heart Rate (HR), Pulse Rate (Pulse), Respiration Rate (Resp), and Oxygen Saturation (SpO₂). We found that for both SVM and ANN the HR signals achieve the highest classification accuracy with 86.8% and 84.5%. This result is significant, because it establishes that HR signals can be used to discriminate between pulmonary edema and respiratory failure.