Multi-sensor adaptive heart and lung sound extraction

Continuous monitoring of heart and lung sounds is of essential importance in medical diagnosis in patients with lung or heart diseases and detection of critical conditions in operating rooms. To obtain quantitative and reliable diagnosis and detection, it is critically important that cardiac and respiratory auscultation retains sounds of high clarity. Clinical acoustic environment imposes great challenges for heart and lung sound acquisition. Unlike acoustic labs in which noise levels can be artificially controlled and reduced, operating rooms are very noisy due to surgical devices, ventilation machines, conversations, alarms, etc. The unpredictable and broadband natures of such noises make operating rooms a very difficult acoustic environment. More technically, lung and heart sounds are weaker than environment noises, and have frequency bands which overlap significantly with noise frequencies. As a result, high fidelity microphones and traditional noise filtering or cancellation techniques cannot help much. Furthermore, due to large variations in patient physiological conditions, operating variables, surgical types, and operating room settings, sound transmission channels vary vastly from patient to patient and during the surgical process. Consequently, it becomes imperative to provide modeling capability for capturing individual characteristics of sound transmission channels. This paper presents a signal processing method that uses (1) an embedded signal for system excitation and identification, (2) an adaptive. algorithm for updating systems so that time variations can be compensated, (3) an signal separation algorithm to extract desired signals, and (4) an adjustable filter to reduce noise impact on the target signal components. This approach can potentially provide superior performance over algorithms that rely on individual fixed filters or statistical based sound separation.