De-noising cardiac ultrasound data in real-time can be accomplished using brushlet analysis (an offshoot of wavelet analysis) to greatly reduce noise and artifact while preserving crisp anatomical detail. The brushlet set has been empirically and mathematically developed for the specific characteristics of cardiac images, and relies on spatial and temporal data to selectively distinguish true signal from noise. Demonstrated benefits include reduced 'speckle' noise, improved definition of myocardial borders, improved accuracy of manual tracing, and reduced inter- and intra-observer variability in quantification of important clinical parameters. In addition, the algorithm has already been demonstrated to facilitate down-stream application of computer-aided analysis techniques.
Ultrasound has proven to be a useful technique for investigating pathology and physiology of the heart, and the use of real-time 3D echocardiography shows growing promise. However, the high level of noise present in these ultrasound images obscures subtle defects. Furthermore, this noise has greatly hindered the development of automated processes like computer-aided diagnosis or automatic segmentation, and thus requiring great time and expertise in interpretation. The techniques demonstrated using the brushlet signal processing technique smartly remove noise from the ultrasound reading by distinguishing between what is noise and what it an actual cardiac signal.
Patent Issued (US 7,542,622)
Available for licensing and sponsored research support
Tech Ventures Reference: IR M03-040