At the recent years, major advances in artificial intelligence (AI) have contributed significantly in the analysis of medical images. Consequently, it has brought into focus new avenues for improving diagnostic accuracy, streamlining workflow and personalizing patient care. This chapter presents a comprehensive analysis of AI in medical imaging from an innovative perspective, drawing attention to multimodal and multitask learning, explainability and trustworthiness as well as emerging architectural paradigms. From a historical perspective, we examine how multimodal integration has developed. Multimodal integration refers to the use of analyzing radiological images together with genomic data, electronic health records (EHRs) and clinical notes, producing a more holistic diagnosis. We also address transformer-based architectures, specifically Multimodal Medical Transformers (MMT) and their applications in oncology and neurology where they perform better than unimodal approaches. Other challenges and prospects are discussed including explainability and trustworthiness issues in digital diagnostics. We describe post-hoc interpretability methods (e. g., Grad-CAM and attention mechanisms), and intrinsically interpretable models (e. g., Concept Bottleneck Models). A comparison and analysis of recent regulatory standards for transparency in healthcare provide insights into the future of AI in medicine. Lastly, we provide a critical look ahead with new directions of development, including dynamic multimodal fusion, causal reasoning frameworks and federated learning as approaches to privacy-preserving AI.