Big data dimensionality reduction-based supervised machine learning algorithms for NASH diagnosis.

Area of Science:

• Hepatology
• Machine Learning
• Biomedical Data Science

Background:

• Non-Alcoholic Steatohepatitis (NASH) diagnosis remains challenging due to a lack of effective early detection methods.
• Significant amounts of redundant medical data are collected for NASH diagnosis.
• Early and accurate NASH identification is critical to prevent liver failure-related morbidity.

Purpose of the Study:

• To develop accurate Non-Alcoholic Steatohepatitis (NASH) prediction models.
• To identify the most informative blood test data for NASH diagnosis.
• To optimize NASH prediction using advanced machine learning algorithms.

Main Methods:

• Feature selection using Pearson correlation and Particle Swarm Optimization with Artificial Neural Networks (PSO-ANN).
• Optimization of NASH prediction models using Batch Least Squares (BLS) and Artificial Bee Colony (ABC) algorithms.
• Training and validation of machine learning models with selected blood test data.

Main Results:

• The BLS model achieved 100% accuracy for benign and 98% for malignant NASH cases.
• The ABC model achieved 90.5% accuracy for benign and 94.3% for malignant NASH cases.
• Both models demonstrated high performance in diagnosing NASH based on selected blood biomarkers.

Conclusions:

• Machine learning algorithms, particularly BLS, show high potential for accurate Non-Alcoholic Steatohepatitis (NASH) diagnosis.
• Optimized feature selection from big data enhances the predictive power of NASH diagnostic models.
• This approach offers a promising avenue for early and accurate NASH detection, aiding in the prevention of liver failure.