Investigation of predictive factors for fatty liver in children and adolescents using artificial intelligence

Area of Science:

• Pediatric Hepatology
• Medical Informatics
• Machine Learning in Medicine

Background:

• Childhood obesity is a global health concern, increasing the incidence of non-alcoholic fatty liver disease (NAFLD), the most prevalent liver condition in children.
• Liver biopsy, the current standard for NAFLD diagnosis, can be invasive. Early detection through advanced methods is crucial for better patient prognosis.
• Machine learning (ML) offers a promising avenue for developing non-invasive, early diagnostic tools for pediatric NAFLD.

Purpose of the Study:

• To identify key predictive factors for NAFLD in pediatric populations using ML models.
• To evaluate the efficacy of various ML algorithms in diagnosing NAFLD based on liver biopsy outcomes.
• To assess predictive performance for specific NAFLD histological features like fibrosis, steatosis, and ballooning.

Main Methods:

• Analysis of data from 659 children with suspected NAFLD who underwent liver biopsy between 2011 and 2023.
• Data preprocessing involved one-hot encoding for categorical variables and standardization for numerical features.
• Training and evaluation of ML models including CatBoost, AdaBoost, Random Forest, and GradientBoosting using cross-validation and metrics like accuracy, precision, recall, F1 score, and ROC-AUC.

Main Results:

• The CatBoost Classifier achieved the highest predictive accuracy (91.8%) and ROC-AUC (92.3%) in cross-validation for NAFLD diagnosis.
• Adjusted models demonstrated improved performance, with CatBoost's F1 score increasing from 83% to 89% (AUC: 0.86-0.92).
• Other models like GradientBoosting and Bernoulli Naive Bayes also showed enhanced predictive capabilities after adjustments.

Conclusions:

• Machine learning models, especially CatBoost, exhibit significant potential for accurate and early diagnosis of NAFLD in children.
• These findings suggest ML can serve as a valuable tool to support clinical decision-making in pediatric NAFLD.
• Further development and validation of ML algorithms could lead to improved diagnostic strategies and patient management for childhood NAFLD.