Look into pydicom to load your images. Then, you will need to either manually or - more difficult - automatically segment the breast(s), excluding the pectoralis muscles and adjacent air. The data in your images should be Hounsfield units, which are calibrated to quantitatively measure density (0 is water, -1000 is air at the scanner, everything positive is more dense than water). There are many published ranges you could use to start with to differentiate fat from non-fat tissue in your area of interest.
Scikit-Image will likely be most helpful in the segmentation phase. There are several thresholding algorithms which may be of use also, if a simple threshold isn't ideal.
Please note that differentiating fat from other soft tissue is relatively easy; differentiating glandular from 'other soft tissue' is not as they will share the same density. If this is a key part of the project, simple density-based thresholding will not be sufficient.