To what extent is it feasible to design and implement a multimodal fake news detection system that leverages both textual and visual data, allowing for independent and joint evaluation of text-based, image-based, and combined modalities to identify the most effective approach? What are the primary technical and methodological challenges associated with developing such a system, particularly when using multimodal fake news datasets?