Improvement of the GMM-AR classification of multiframe contrast ultrasound images using Gaussian filter

Abstract

Despite the use of contrast agents that enhance the visualization of vascular zones, the backscattered signals from the contrast agent and tissue are still close which prevents the direct wide ultrasonic use in diagnosis. Thus, it was necessary to implement image-processing techniques that enhance the contrast echo and have the capability of classification. We have applied a new approach based on the autoregressive model where an image of prediction errors is calculated in the first phase. Then, a Gaussian filter is applied in order to model well afterward both agent and tissue behaviors by a Gaussian mixture model. The Agent to Tissue Ratio (ATR)factor and Fisher criterion are adopted to compare the performance of this method with existing techniques as the harmonic and B mode techniques. The experiments conducted have shown the advantages of our proposed approach where an increasing of ATR and Fisher are recorded. In fact, our ATR attains 19.20 dB which represents a good improvement in comparison with B mode (9.50 dB) and Harmonic technique (12.13 dB). Whereas Fisher, the parameter of classification feasibility, it reaches 2.01 which matches an excellent amelioration with respect the mentioning techniques with 0.97 and 1.00 respectively.