Algorithmic models with artificial intelligence for disease diagnosis: A systematic literature review

Abstract

The integration of artificial intelligence (AI) in disease diagnosis is transforming the field of medicine, offering precise and efficient tools to detect critical health conditions. However, the diversity of algorithms and their applications raises questions about which of these are the most effective. Therefore, this systematic review aims to identify the most accurate AI models for detecting various diseases. A systematic review was conducted considering Scopus and Web of Science databases as main information sources, analyzing 416 studies that used AI algorithms within the medical field. Additionally, rigorous inclusion and exclusion criteria were applied, screening 26 articles that prioritize quantitative results relevant to clinical diagnosis. The most prominent model is Random Forest, with a frequency of use in 12 investigations and an average accuracy of 0.91. Likewise, the XGBoost, CNN, and SVM models were used in 9 investigations each and obtained accuracy results of 0.87, 0.88, and 0.96 respectively. These performances were particularly notable in applications related to dermatological, cardiological, and oncological diseases. The results position Random Forest as an efficient tool for medical diagnosis, although its practical implementation faces some technological and budgetary challenges. It is recommended to explore hybrid methodologies that combine advanced algorithms with more traditional approaches and conduct longitudinal studies to evaluate their impact in different clinical settings.