In one embodiment, a system for identifying a biomedical condition of a subject includes apparatus for collecting blood flow sounds from the subject and a computing device that stores computer-executable instructions that are configured to: receive the collected blood flow sounds, extract acoustic heart pulses from the collected blood flow sounds, segment the acoustic heart pulses to obtain acoustic heart pulse segments, compute a continuous time wavelet transform-based feature for each acoustic heart pulse segment, and perform clustering on the computed continuous time wavelet transform-based features to determine whether or not the subject is experiencing the biomedical condition.USF researchers have developed a machine learning framework to identify the patterns of cardiac sounds. The acoustic sounds are obtained from the body site and are preprocessed to identify the denoised heart sound using a segmentation and wavelet decomposition framework. To characterize the healthy and unhealthy heart sounds, multiscale energy feature was developed and obtained for the segmented denoised heart signals. These features are then used as input to the pattern recognition framework for further classification. The framework is quantitatively and qualitatively validated. The classification accuracy of the framework to identify the abnormalities is 94.37%.
Brochure