A technique for selecting portions of a multi-resolution medical image data set to be stored and the portions of the multi-resolution medical image data set to be discarded in order to reduce the overall amount of image data that is stored for each image data set. The selection is based on the clinical purpose for obtaining the medical image data. The clinical purpose for obtaining the medical image is used to define regions of interest in the medical image. At each resolution level of the multi-resolution medical image data set, the regions of interest are stored at the full resolution, while the remaining portions of the medical image are stored at a lesser resolution. A three-dimensional bit mask of the regions of interest is produced from a segmentation of the regions of interest. The segmentation list and the multi-resolution medical image data set are decomposed into multiple resolution levels. Each resolution level has a low frequency component and several high frequency components. The low frequency portions at each resolution level may be stored in their entirety. The segmentation list is used to select the regions in the high frequency portions of the multi-resolution image data that correspond to the regions of interest and those regions that do not. The regions in the high frequency portions of the multi-resolution image data that correspond to the region of interest are stored. Those regions in the high frequency portions of the multi-resolution image data that do not correspond to a region of interest are discarded.
