Artificial Intelligence in Health Care: Clinical Opportunities, Validation Trends, and Implementation Challenges (2020-2025).
Authors
Abstract
Artificial intelligence (AI) continues to reshape health care, supported by advances in computing power, affordable data storage, and widespread electronic health record adoption. This systematic review followed PRISMA 2020 guidelines, searching PubMed, IEEE Xplore Digital Library, and Web of Science for studies published between January 2020 and September 2025. Eligible articles included original research addressing AI applications in clinical contexts with prospective or external validation. Fifteen studies met inclusion criteria. Publication volume peaked in 2024 (n = 5, 33.3%). Deep learning was the most widely adopted method (60.0%), with convolutional neural networks frequently used in medical imaging. Radiology comprised 33.3% of applications, followed by oncology (20.0%) and cardiology (13.3%). The median diagnostic performance of imaging-based models was an area under the curve (AUC) of 0.91 AUC. Primary implementation challenges included regulatory compliance (53.3%), lack of algorithmic transparency (40.0%), data quality problems (33.3%), and clinical integration challenges (26.7%). Compared with earlier reviews (2015-2019), recent studies demonstrated increased external validation rates (rising from 23% to 46.7%) and greater focus on algorithmic fairness. Artificial intelligence demonstrates substantial potential to advance health care quality across multiple domains improving diagnostic accuracy and patient safety, enhancing care efficiency and timeliness, and supporting equitable care delivery when appropriately validated across diverse populations. However, realizing these quality improvements requires addressing persistent implementation barriers including regulatory uncertainty, algorithmic transparency, data quality concerns, and clinical workflow integration challenges.