Abstract: The measurement of human respiratory signals is crucial in field of medicine. A disordered breathing pattern can be the first symptom of different mechanical or psycho-logical dysfunctions. Therefore, a real-time monitoring of the respiration pattern is a critical need in medical applications. In clinical settings, respiration measurement methods are used a large number of sensors to the patient’s body for recording vital signals, which might interfere with natural breathing of the patients and also cause discomfort if used for longer durations. This paper presents a novel algorithm for diagnosing lung status using photoplethysmographic (PPG) derived respiratory signals. The algorithm contains two steps. First one is the Respiratory cycle extraction process and second one is the Classification of respiratory activity. The pulse oximeter’s PPG signals can be well utilized for extracting respiratory activity, avoiding the usage of additional sensor for recording respiratory signal. Modified Multiscale Principal Component Analysis (MMSPCA) is used for extraction of respiratory activity embedded in the PPG signals. Functioning of the proposed algorithm is tested on the dataset consists of different PPG patterns i.e., normal, hypoventilations, hyperventilation’s and kussmaul recordings available with MIMIC database of Physio net archive. The second part is the evaluation of lung status according to the respiratory signal which is extracted from the PPG signal in the first part. Multiclass support vector machine (MSVM) classifier or Kernel Nearest-Neighbour (knn) classifier are used to classify the respiratory signals according to the features extracted from respiratory signal. This paper presents a comparative study of the classification of respiratory cycle using MSVM and k-nn classifier.

Keywords: Modified MSPCA (MMSPCA), principal component analysis (PCA), Pulse oximeter’s Photoplethysmo-Graphic (PPG) signal, respiratory signal, wavelets, Multiclass Support Vector Machine (MSVM), K-nn classifier.