AI-Driven Model for Automated Detection of Chronic Heart Failure using Phonocardiography Data

Abstract

 Chronic heart failure (CHF) is a long-term condition characterized by the heart's inability to adequately supply perfusion to target tissues and organs, resulting in insufficient metabolic demands being met at physiological filling pressures. The prevalence of CHF in the population has reached alarming levels, with its incidence showing a steady annual increase of 2%. CHF has a significant impact on the population in developed countries, particularly among individuals over the age of 65. At present, the diagnosis and treatment of CHF accounts for about 2% of the annual healthcare budget. The United States spent around 35 billion USD solely on CHF treatment in 2018, with projections indicating that these costs will double within the next decade. Typically, a skilled doctor can identify the progression of HF by conducting a thorough examination of the patient and analyzing specific changes in the patient's heart failure biomarkers, which are obtained from blood samples. Regrettably, when a CHF patient experiences clinical worsening, it typically indicates that they are already facing a fully developed CHF episode that will likely necessitate a hospital admission. In certain patients, heart failure worsening may be accompanied by distinct changes in heart sounds that can be detected through phonocardiography. Thus, utilizing cutting-edge advancements in machine learning and deep learning models, this project aims to detect chronic heart failure from phonocardiography (PCG) data. This is achieved through an end-to-end average aggregate recording model that incorporates extracted features from both machine learning and deep learning techniques. The proposed ChronicNet model was also compared with individual ML and DL models.