Breast cancer accounts for an estimated 2.22 million new cases and more than 684.000 deaths per year [1]. This emphasises the importance, necessity and promise of progress in deep-learning based computer-aided detection and diagnosis (CAD) systems for improved cancer detection at earlier stages. To train such deep-learning systems, vast amounts of patient imaging data is needed and ingested in the model, which may leak some of this private patient information after training [2, 3]. It, hence, can become necessary to actively protect patient information during model training, in particular, when the objective includes training deployable and shareable deep learning models for use in clinical practice. However, when training these models with a privacy guarantee based on differential privacy [4, 5], the performance and, hence, the utility of the resulting model is often substantially reduced, which is known as the privacy-utility trade-off [6]. To this end, we investigate the possibility of improving model utility by creating and adding synthetic cancer imaging data [2] into the deep learning model training workflow. We evaluate our framework on the task of breast tumour malignancy classification based on regions-of-interest extracted from mammograms.