A noise adaptive method for needle localization in 3D ultrasound images

Intra-operative ultrasound imaging is a widely used technique for navigation of many minimally invasive medical procedures. Using the ultrasound imaging has advantages such as safety, speed and low cost in expense of low quality of the image due to the speckle noise. In this paper, we describe two techniques for increasing the needle localization accuracy in high and low noise levels of the 3D ultrasound images. The level of the speckle noise in the input image is estimated using the patch based noise estimation method. In low noise levels we used the maximum likelihood criteria for model evaluation step of the fitting procedure. In high noise levels, the image is filtered before thresholding, using eigenanalysis of the Hessian matrix. The results show improvement in accuracy and speed with respect to the previous model fitting RANSAC based method.