Data augmentation represents a foundational methodological framework within deep learning, aimed at enhancing both the quantity and diversity of training datasets. This approach is particularly critical in the domain of medical imaging, where data availability is frequently limited due to privacy regulations, high acquisition costs, and inherent logistical challenges associated with data collection. This study explores the efficacy of advanced data augmentation techniques on the synthesis of contrast response in contrast-enhanced spectral mammography (CESM) images, employing state-of-the-art image-to-image translation models. The analysis includes conventional data augmentation techniques based on geometric transformations, as well as contrast variation achieved through randomized adjustments of the Window Center (WC) and Window Width (WW) parameters. Among the evaluated techniques, the random adjustment of WC and WW demonstrates superior performance in enhancing the qualitative synthesis of diagnostically relevant regions of interest (ROIs) exhibiting contrast agent responses, particularly in dense breast tissue. Nevertheless, the study highlights that traditional quantitative metrics may inaccurately capture clinical relevance, particularly in scenarios emphasizing the characterization of contrast agent dynamics. The findings underscore the promise of contrast-based augmentation strategies in advancing the generation of diagnostically significant images while simultaneously illuminating the intrinsic constraints of conventional quantitative metrics.