Artificial Intelligence (AI), particularly Deep Learning (DL), is rapidly transforming nutrition science by overcoming limitations of traditional self-reported dietary methods, which are prone to bias and inaccuracy. This review presents a comprehensive, integrative analysis of DL applications across five key domains: dietary assessment, food recognition and tracking, personalized nutrition, disease diagnosis and monitoring, and predictive modeling for disease risk. It highlights how established approaches, including CNNs and hybrid computer vision frameworks, have enhanced dietary assessment and food recognition, while emerging applications leverage genomic, behavioral, and clinical data to support precision and preventive health strategies. Beyond summarizing technical advancements, this review critically evaluates persistent gaps, including limited dataset diversity, lack of standardization, model interpretability, and ethical considerations. Finally, it proposes a structured roadmap that aligns technical innovation with clinical integration and ethical frameworks, offering actionable guidance to advance reliable, scalable, and equitable DL-driven nutrition solutions.