Deep Learning for Healthcare: Trends in Lightweight and Efficient Solutions

Deep learning has revolutionized modern healthcare, driving breakthroughs in medical imaging, disease detection, genomics, and personalized medicine. However, the substantial computational and memory demands of state-of-the-art models limit their adoption in real-world clinical environments, particularly in resource-constrained settings. This survey presents a comprehensive overview of lightweight and efficient deep learning approaches designed to address these challenges. We examine architectural innovations such as MobileNet, EfficientNet, ShuffleNet, GhostNet, and MobileViT, highlighting their design principles and the trade-offs between accuracy and computational complexity. We also review model compression techniques, including pruning, quantization, knowledge distillation, low-rank factorization, and neural architecture search, as complementary strategies for reducing computational costs while maintaining diagnostic performance. The paper explores the deployment of lightweight and efficient models across diverse healthcare domains, including disease detection and classification, remote monitoring, clinical decision support, reconstruction and restoration of medical images, genomics and precision medicine. Finally, we analyze emerging trends and identify open challenges related to model scalability, hardware-software co-optimization, and accuracy-efficiency balance. This work underscores the critical role of efficiency in advancing scalable, sustainable, and accessible AI-driven healthcare across global and clinical contexts.