Integrated neural network and evolutionary algorithm approach for liver fibrosis staging: Can artificial intelligence...

Area of Science:

• Medical Informatics
• Machine Learning
• Hepatology

Background:

• Accurate staging of liver fibrosis is crucial for patient management.
• Liver biopsy, the current gold standard, is invasive.
• Non-invasive methods for fibrosis prediction are highly sought after.

Purpose of the Study:

• To develop and evaluate an artificial neural network (ANN) model integrated with the Teaching Learning-Based Optimization (TLBO) algorithm.
• To predict the stage of liver fibrosis in blood donors and hepatitis C patients.
• To reduce the number of input features required for accurate fibrosis prediction.

Main Methods:

• Utilized machine learning classification methods: multilayer perceptron (MLP), Naive Bayesian (NB), decision tree, and deep learning.
• Applied Synthetic Minority Oversampling Technique (SMOTE) to handle dataset imbalance.
• Integrated MLP with the TLBO algorithm for optimized feature selection and prediction.

Main Results:

• The proposed MLP model with TLBO achieved a diagnostic accuracy of 0.891 using only seven features.
• The accuracy of standard MLP with 12 features was 0.903.
• SMOTE application improved the diagnostic accuracy for most methods, excluding the Bayesian network model.

Conclusions:

• The TLBO-optimized MLP model offers a straightforward approach with reduced feature requirements and comparable accuracy to more complex methods.
• Decision tree-based deep learning models showed the highest accuracy with 12 features.
• The proposed method demonstrates potential for accurate, less invasive liver fibrosis staging.