<p>Supervised classification of image segments is a key component of several healthcare image segmentation techniques. When the training and test images are comparable, these techniques typically perform well. However, when the training and test images differ, which might happen because of differences in scanners, scanning techniques, or patient groups, issues may arise. Results can be improved in these situations by providing training images that resemble the test images. However, this method does not make dissimilar data more similar; it only works if some training images have previously been comparable to the test data. We investigate ways to enhance image weighting by minimizing the discrepancies between training and test images using a method known as kernel learning. Additionally, we present Dynamic Weighted Adversarial Network (DWAN), a novel image identification technique that reduces the disparity between training and test data. This DWAN technique enables simultaneous optimization of the image kernel and weight. Experiments on several brain sample image kinds demonstrate that dynamic weighting by itself greatly enhances performance on a range of data. The performance of the novel DWAN weighting technique is comparable to that of the current techniques. Whether optimized independently or together, image weighting and dynamic models can offer a slight performance improvement.</p>