I would definitely suggest you try out MIPAR at http://MIPAR.us. We have a powerful adaptive threshold tool that selects pixels based on their being greater than either a certain percentage of, or a certain intensity away from the average local intensity. We also have a local threshold tool that thresholds within a set of masking features. However, the real power comes from the fact that tools like these become part of a larger non-destructive, easy-to-build, re-use and VISUALIZE recipe, with other steps for normalizing, cleaning, etc.
I would be more than happy to work on an example image of yours and send back results and a recipe file!
If your team is into open source software and you are willing to implement some programming, check out the Scikit-Image library for the python programming language. There are some easy examples of adaptive thresholding in their example gallery.
There is a simple and effective segmentation method for breast ultrasound proposed by Horsch et al. in the following paper: http://dx.doi.org/10.1118/1.1429239
First, a ROI should be cropped and some despeckling filter should be applied (e.g., median filter, anisotropic filter, etc.). Next a constraint Gaussian function is used to enhance the lesion region by attenuating distant pixels that do not belong to the lesion. Next an iterative thresholding (from the darkest to the brightest gray level of the enhanced image) is applied to create potential lesion margins. For each one, the average radial derivative (ARD) function is evaluated. At the final, the potential lesion margin that maximized the ARD is taken as the final lesion contour, which is related to an "optimum" threshold.
Later, I used this method to developed another segmentation method that uses watershed transformation aiming to depict fine details of the lesions contour. You can find this method in the following paper: http://dx.doi.org/10.1118/1.3265959