Malaria Detection Using Modern Machine Learning Techniques: GBM, XGBoost, and ANN

Abstract

Despite the widespread application of generative artificial intelligence (GenAI) in clinical healthcare, varying levels of performance depend on specific task types. While narrow, clearly defined tasks such as structured documentation reach up to 90+% of accuracy and show clear benefits for patients, open diagnostic tasks do not fare well, yielding 52% of accuracy, significantly worse than expert physicians. Major obstacles in adoption include the hallucination of false clinically relevant information by GenAI, differences in performance by different demographic groups, difficulty in providing understandable explanation and regulatory gaps. In order to assist healthcare organizations in GenAI adoption in a responsible manner, we propose a phased framework for institutional integration based on the published literature and validated via three separate methods: 24 experts' consensus (83%), computational simulation of 180 institutions and the analysis of three real-world early adopters of GenAI. The proposed framework suggests an approach for the organization to go through readiness assessment (4.2 – 8.4 months), pilot validation including fairness testing (6.8 months) and scaling with continuous monitoring (11.2 - 18.6 months in total). Following the framework led to an early identification of constraints, thorough fairness testing and robust institutional governance setup. 4.4% failure-to-progress rate illustrates that included checkpoints and reversibility indeed overcome real implementation obstacles. We offer assessment instruments, decision matrices and governance templates, which help to adopt GenAI responsibly and safely for healthcare organizations.