Abstract:
Chronic Obstructive Pulmonary Disease (COPD) is a persistent respiratory disease usually caused by toxic gases. The diagnosis is made by a specialist doctor on a report taken by a specialist technician using a spirometer. Diagnostic steps can only be carried out in hospital environment in the presence of a qualified technician. The diagnostic process is so troublesome that it leads to alternative system requirements. In this study, a portable software system based on photoplethysmography signal is proposed as an alternative method to reduce the burden of the diagnosis process of the disease. For this purpose, 26 features were extracted from the photoplethysmography signal in time domain. The extracted features were classified by machine learning based k - Nearest Neighbors algorithm and tried to diagnose the disease. The study included 8 patients with COPD and a control group of 6 patients. Parameters such as accuracy, sensitivity, specificity and f-metric were used to calculate the classification performance. According to some k values and distance algorithms, all data are correctly classified as % 100, with a sensitivity of 1, a specificity of 1 and a measurement of F of 1. Given the results of the study, it has come to the conclusion that machine learning-based COPD diagnosis can be done effectively and productively.
