Embedded Platform-Based Heart-Lung Sound Separation Using Variational Mode Decomposition

Cardiovascular diseases (CVD) are often identified by the audio characteristics of persistent heart and lung sounds of healthy and abnormal subjects. Quite often, these sounds recorded from various subjects are found mixed with each other which in turn leads to ambiguity in the diagnosis. Hence, accurate and precise separation of heart and lung sounds is the essential and foremost task in heart-lung sound analysis. Various signal processing techniques and algorithms have attempted to decompose the heart sound and lung sounds from mixed signals. However, many of them were implemented on bulky and standalone computerized systems and have limited exposure of low-power embedded implementation. In this work, we attempted to implement a novel framework for heart-lung sound separation on a low-power and resource-constrained embedded platform. As a part of the work, we have explored variational mode decomposition (VMD) algorithm to separate the heart and lung sounds from the mixed signals. The proposed framework is implemented on standalone Windows operating system and Raspberry PI-based embedded platform to emulate the offline and real-time scenarios, respectively. Further, the proposed framework is tested with most popular and publicly available heart and lung sound databases obtained from Michigan heart sound library and Littmann lung sound library. The obtained results on both stand alone and embedded platforms are compared with the state-of-the-art techniques. The obtained results on embedded platform are found promising as compared to that of standalone PC-based implementation. As the targeted embedded platform is a low-power, portable, and low-cost computational device, the precise implementation of this work can find fascinating applications in IoT-based heart sound analysis and handheld heart sound assessment devices.