Description

Echocardiography (cardiac ultrasound) is the predominant technique for quantitative assessment of cardiac function. Tracing the borders of the cavities on the echocardiography images is essential for quantification of cardiac function.

This project describes a new approach for echocardiography video image segmentation. The proposed method uses the Fuzzy logic, Multilayer perceptron trained with back propagation algorithm and combination of these two approaches for automatic seed point selection as well as the selection of boundary pixels for completing the contour. Initially the key frames are selected based on marker symbol and the peak value of ECG wave available in the video image sequence. Preprocessing techniques namely image enhancement and presegmentation based on region growing technique followed by edge detection method is applied.

Three different approaches for segmentation namely Pure Fuzzy, hybrid A, hybrid B, were developed. In Pure Fuzzy (PF) approach the boundary pixels are selected by carefully applying the fuzzy rules. The fuzzy rules are framed based on mean and standard deviation. Gaussian membership function is used for each input and triangular membership function is used for output.

The hybrid A approach which combines BPN and fuzzy logic for single seed point selection by BPN with error rate 0.001 and completing the contour using Fuzzy logic. The hybrid B approach which combines BPN and fuzzy logic for multiple seed points selection by BPN with error rate 0.001 and completing the contour using Fuzzy logic.

Dataflow diagram for Video Segmentation

Matlab Gui for Video Segmentation