Machine learning-based medical decision support system for diagnosing HFpEF and HFrEF using PPG

Abstract

Heart failure (HF) cases are increasing day by day. Rapid diagnosis, initiation of treatment, reduction of mortality and prolongation of lifespan are important. Left ventricular ejection fraction (LVEF) is an important determinant of HF, especially in HF with reduced ejection fraction (HFrEF, LVEF ≤ 40 %) and HF with preserved ejection fraction (HFpEF, LVEF ≥ 50 %). LVEF is a measure of the amount of blood pumped out of the ventricle in one heartbeat. HFrEF is easier to diagnose since LVEF ≤ 40%. However, the diagnosis of HFpEF is difficult even for some specialists, since LVEF is 50 % or higher also in a healthy person. LVEF is measured by echocardiography which is an expensive device and requires a specialist. There may be situations where access to the device is limited. Some invasive and laborious blood tests are also used for diagnosing HF. As an alternative diagnostic method, a new machine learning-based diagnostic algorithm was developed for diagnosing HF and its subtypes using photoplethysmography (PPG). PPGs from volunteers were cleaned with digital filters. Then HRVs were derived from PPG and features were extracted from both PPG and HRV. Features were reduced by statistical methods. The classification was done with three different machine learning algorithms. The evaluation was made with 10-fold cross validation and maximum performance parameters: accuracy %87.78, sensitivity 0.87 and specificity 0.94. With this triple classification, it is determined not only whether the individual has an HF, but also which HF is present.