Background & aim: The intensive care unit (ICU) is among the most critical hospital departments, requiring rapid and precise decision-making based on vast amounts of data. Artificial intelligence (AI), by providing advanced analytical tools, is playing an increasingly prominent role in enhancing diagnostic, therapeutic, and management processes in this setting. The primary aim of this review article is to present a systematic classification of different types of AI and their associated statistical methods, with a focus on their applications in critical care. This classification is intended to facilitate a clearer and more practical understanding of these technologies within the ICU.
Methods: This narrative review was designed and conducted in accordance with established standards for narrative review articles. The research process was structured into four main phases. First, a comprehensive literature search was performed across major scientific databases, including PubMed, Scopus, Web of Science, and IEEE Xplore. The search strategy employed a combination of keywords related to artificial intelligence, intensive care, machine learning, early diagnosis, and resource management in critical care, along with their English equivalents. The search was limited to articles published between 2010 and 2024. In the second phase, clear inclusion and exclusion criteria were established. Cohort studies, clinical trials, meta-analyses, and review articles focusing on the application of artificial intelligence in intensive care were included. Animal studies, non-systematic case reports, and articles without full-text availability were excluded. During the data extraction and analysis phase, key information from each study was collected, including the type of artificial intelligence approach (symbolic AI, machine learning, deep learning, fuzzy systems), the specific models used (neural networks, support vector machines, decision trees), the domain of application (diagnosis, risk prediction, treatment management), and measures of effectiveness (accuracy, sensitivity, improvement in clinical outcomes). The final phase involved the development of an analytical framework based on two main axes: first, the classification of AI approaches by capability, distinguishing between rule-based systems (symbolic AI) and data-driven systems (deep learning); and second, a performance-based classification encompassing diagnostic applications (such as radiological image analysis), predictive modeling (risk models), and decision-making (optimization of treatment protocols). Specialized models were further evaluated for their practical utility in intensive care settings, including the use of convolutional neural networks for physiological data analysis and fuzzy systems for resource management. Ultimately, the findings were organized in a comparative matrix structured by model type, domain of application, and level of supporting evidence. Interdisciplinary overlaps, such as the integration of reinforcement learning with clinical decision support systems, were also identified. This methodology was developed in accordance with best practices for narrative reviews, emphasizing transparency and rigor in reporting.
Results: The results of this review indicate that AI in critical care encompasses a wide range of approaches, including symbolic AI, machine learning (supervised, unsupervised, and reinforcement learning), deep learning, evolutionary AI, fuzzy systems, and swarm intelligence. Each of these approaches, utilizing specific statistical methods, offers unique capabilities in areas such as early disease detection, risk prediction, treatment optimization, and resource management.
Conclusion: While the proposed classification can enhance understanding of AI applications in critical care, it is important to note that these categories are not always mutually exclusive, and there is overlap between different approaches. The choice of the appropriate AI method and corresponding statistical technique depends on the specific characteristics of the problem and the available data. Nevertheless, a thorough understanding of AI types and their statistical foundations enables critical care professionals to take effective steps toward improving care quality and patient outcomes. The intersection of AI and critical care opens new horizons for advancing patient health and improving therapeutic results. AI, with its broad spectrum of approaches and techniques, holds transformative potential for critical care. Deep knowledge of these methods empowers specialists to select optimal strategies for improving patient health. The integration of AI and critical care is set to reshape the future of medicine. Responsible development requires attention to ethical challenges and ensuring equitable use of these technologies. |
