Phát hiện các mức suy dinh dưỡng ở bệnh nhi dựa vào trí tuệ nhân tạo

Malnutrition among pediatric patients remains a significant global health concern, frequently observed in clinical settings and impacting treatment outcomes, recovery, and overall quality of life. Traditional screening methods, while effective, often demand considerable time and specialized expertise, leading to potential delays or incomplete assessments in busy healthcare environments. The timely and accurate identification of malnutrition is therefore paramount for initiating early interventions and improving patient prognosis. In this context, artificial intelligence (AI) offers a promising avenue for developing advanced predictive tools to enhance the efficiency and precision of malnutrition detection. This study introduces an AI-driven model designed to classify malnutrition levels in children, aiming to address the limitations of conventional approaches and provide critical support for clinical decision-making.

Pediatricians, nutritionists, public health professionals, healthcare administrators, and AI/data science researchers in medical applications.

This study develops an interpretable artificial intelligence model to enhance the detection and classification of malnutrition levels in pediatric patients. Utilizing a descriptive cross-sectional design, data from 1,563 pediatric inpatients at Children's Hospital 2 were analyzed. A Logistic Regression model was employed to predict five distinct levels of malnutrition based on the hospital's existing screening tools. The model demonstrated a high overall accuracy of 96%, indicating its robust performance in identifying various malnutrition statuses. Significantly, the research highlighted that factors related to underlying diseases, such as metabolic disorders and malabsorption, alongside the patient's eating status, exerted the greatest influence on prediction accuracy, outperforming traditional simple anthropometric indices like BMI or Z-scores. While the model showed strong performance in specific disease groups like lower respiratory infections (95% accuracy), its effectiveness was comparatively lower for acute diarrhea (82% accuracy). A notable challenge encountered was the model's difficulty in accurately classifying "Very Severe" malnutrition due to a severe data imbalance in the dataset. The findings underscore the potential of AI models to provide effective and transparent support for malnutrition severity classification, emphasizing the critical role of underlying health conditions and dietary factors. For practical application, future research should focus on mitigating data imbalance, refining predictive models for specific patient subgroups, and exploring their utility in outpatient and community settings to maximize their applicability and impact on pediatric health outcomes.