Classification of Nonalcoholic Fatty Liver Grades using Pre-Trained Convolutional Neural Networks and a Random Forest...

Area of Science:

• Medical Imaging Analysis
• Machine Learning in Healthcare
• Hepatology

Background:

• Nonalcoholic Fatty Liver Disease (NAFLD) is a prevalent condition with significant health implications.
• Early detection and accurate grading of NAFLD are crucial for effective management and treatment.
• Current diagnostic methods may have limitations in accessibility and expertise.

Purpose of the Study:

• To develop an advanced hybrid machine learning model for classifying NAFLD grades.
• To utilize ultrasound images for non-invasive NAFLD assessment.
• To enhance the accuracy and efficiency of NAFLD diagnosis.

Main Methods:

• Ultrasound images from 55 obese individuals were analyzed, with liver biopsy results serving as the ground truth.
• Features were extracted using pre-trained Convolutional Neural Network (CNN) models (e.g., EfficientNet-B7) and refined using the mRMR method.
• Classifiers including Random Forest (RF) were employed to categorize NAFLD grades (healthy, low, moderate, high fat).

Main Results:

• The hybrid model combining EfficientNet-B7 for feature extraction and RF for classification achieved the highest accuracy.
• The proposed automatic model demonstrated a remarkable classification accuracy of 96.83% for NAFLD grades.
• Specific classifier accuracies with EfficientNet-B7: LDA (88.48%), MLP (93.15%), SVM (95.47%), RF (96.83%).

Conclusions:

• An automated classification model using EfficientNet-B7 and Random Forest shows high efficacy in grading NAFLD.
• This approach can significantly improve NAFLD diagnosis, particularly in resource-limited settings.
• The study highlights the potential of advanced machine learning in liver disease management.