A Neuro-fuzzy Approach of Bubble Recognition in Cardiac Video Processing

2D echocardiography which is the golden standard in clinics becomes the new trend of analysis in diving via its high advantages in portability for diagnosis. By the way, the major weakness of this system is non-integrated analysis platform for bubble recognition. In this study, we developed a full automatic method to recognize bubbles in videos. Gabor Wavelet based neural networks are commonly used in face recognition and biometrics. We adopted a similar approach to overcome recognition problem by training our system through real bubble morphologies. Our method does not require a segmentation step which is almost crucial in several studies. Our correct detection rate varies between 82.7-94.3%. After the detection, we classified our findings on ventricles and atria using fuzzy k-means algorithm. Bubbles are clustered in three different subjects with 84.3-93.7% accuracy rates. We suggest that this routine would be useful in longitudinal analysis and subjects with congenital risk factors.