Artificial intelligence in modern healthcare: Applications, innovations, and ethical dimensions in personalized, predictive, and inclusive care

Artificial intelligence (AI) in healthcare is evolving from narrow single-task models to multimodal systems that integrate text, images, physiological signals, and structured clinical records. This transition is reflected in the recent World Health Organization guidance on large multimodal models for health. The evidence of clinical value is increasing when AI is embedded in responsive workflows. In a prospective multicenter study, timely confirmation of alerts from a real-time sepsis early warning system was associated with lower in-hospital mortality rates. In breast cancer screening, AI-supported single-reading mammography-maintained cancer detection while reducing the reading workload by approximately 44 %. Simultaneously, external evaluations of widely deployed tools have demonstrated context-dependent performance, miscalibration, and implementation gaps, highlighting the need for local validation and ongoing monitoring. Regulatory oversight is also maturing, including the United States Food and Drug Administration lifecycle framework and the European Union AI Act, which entered into force on August 1, 2024, with phased obligations through 2027 for high-risk uses. This review maps the healthcare data ecosystem, including electronic health records, imaging, waveforms, patient-generated and Internet of Things data, and multiomics. Model families from classical machine learning were compared to transformers and multimodal foundation models, including calibration, uncertainty estimation, robustness assessment, and life-cycle monitoring. Applications through personalized, predictive, and inclusive care were evaluated and ethical, legal, and implementation requirements were synthesized using reporting and evaluation frameworks. We conclude with a practical agenda for trustworthy, equitable, and operationally sustainable deployment in real-world health systems. • AI in healthcare is evolving from single-task models to multimodal systems integrating text, images, signals, and data. • Real-world impact is emerging, with AI-assisted sepsis alerts reducing in-hospital mortality. • AI-supported mammography maintains detection rates while cutting workload by ∼44%. • External evaluations reveal performance gaps and context dependence, highlighting the need for local validation.