Model-aware deep learning is a methodology that leverages knowledge of the underlying model architecture and its parameters to improve the performance of deep learning systems. This approach can lead to faster convergence, better generalization, and reduced overfitting compared to traditional deep learning methods. However, it requires a deeper understanding of the model architecture and the optimization process. This research project focuses on developing new model-aware deep learning approaches tailored to inverse imaging problems, including image reconstruction from partial and degraded data and inverse problems in image analysis, like the estimation of the similarity matrix for spectral clustering. This fundamental research project will establish a solid theoretical framework and provide useful insights for our related and more focused projects in the domain of medical image reconstruction and multimodal image analysis.