A Systematic Review and Evaluation of Sustainable AI Algorithms and Techniques in Healthcare

• Published in: IEEE access Vol. 13; pp. 139547 - 139582
• Main Authors: Ibrahim Alzoubi, Yehia, Topcu, Ahmet E., Elbasi, Ersin
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 2025; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Accuracy; Adaptive sampling; Algorithms; Artificial intelligence; Artificial neural networks; Back propagation networks; Complexity; Computational modeling; Edge computing; Energy consumption; Energy efficiency; Energy management; Federated learning; green AI; Hardware; Health care; healthcare; Machine learning; Medical services; Neural networks; Optimization; Sustainability; Sustainable AI; Sustainable development; Systematic review; Training
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2025.3596189

Concerns regarding energy use, environmental effects, and long-term sustainability have been highlighted in recent years by the expanding application of Artificial Intelligence (AI) in healthcare. This systematic review paper categorizes and classifies AI algorithms and tools in the healthcare sector to support more sustainable practices, focusing on reducing energy use while maintaining high standards in diagnostic accuracy and patient outcomes. AI algorithms and tools are categorized into three groups: explicit AI algorithms for sustainability for energy efficiency (e.g., Federated Learning, Hybrid Quantum-Classical Optimization, Modified Lempel-Ziv-Welch (mLZW)), traditional AI algorithms for sustainable healthcare (e.g., Bidirectional Long Short-Term Memory (Bi-LSTM), Backpropagation Neural Networks (BPNNs), Convolutional Neural Networks (CNNs)), and sustainable AI techniques (e.g., Adaptive Sampling, AutoML for Model Compression (AMC)) that support low-power computing (e.g., edge computing, neuromorphic hardware, adaptive sampling). A comprehensive performance analysis is presented across five dimensions: energy consumption, latency, accuracy, complexity, and cost. The review highlights mLZW as promising for energy efficiency, complexity, and cost, OFA for low-latency deployment, and Hybrid Quantum Classical Optimization for diagnostic accuracy. We propose an integration framework for deploying these methods in resource-constrained healthcare environments, identifying open research challenges and future directions. This work provides a foundational guide for researchers and sector practitioners to build energy-aware, high-performance AI systems in healthcare.