Preliminary Study of Chronic Liver Classification on Ultrasound Images Using an Ensemble Model

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Hepatology

Background:

• Chronic liver diseases are a major global health concern, necessitating early and accurate diagnosis.
• Ultrasound is a primary imaging tool for liver assessment, but differentiating conditions like cirrhosis and hepatocellular carcinoma (HCC) is challenging due to visual similarities.
• Timely diagnosis is crucial for effective treatment and improving survival rates in liver disease patients.

Purpose of the Study:

• To develop and evaluate a robust method for classifying four distinct liver stages: normal, chronic liver disease, cirrhosis, and HCC over cirrhosis.
• To address the diagnostic challenges posed by visually similar ultrasound liver images.
• To enhance the accuracy of automated liver condition assessment using medical imaging and machine learning.

Main Methods:

• Extraction of handcrafted texture features (Ranklet, GLDS, GLCM) from ultrasound liver images.
• Hierarchical feature fusion to create a comprehensive feature set characterizing echotexture and roughness.
• Development of an ensemble classifier combining k-NN, SVM, and Rotation Forest with a voting algorithm.
• Validation using a dataset of 754 segmented regions of interest from clinical ultrasound images.

Main Results:

• The proposed ensemble classifier achieved a high classification accuracy of 96.6%.
• Evaluation demonstrated the effectiveness of handcrafted texture features and the proposed ensemble strategy.
• The study confirmed the superior performance of the ensemble model compared to individual classifiers.

Conclusions:

• The developed method offers a highly accurate and reliable approach for differentiating between normal, chronic liver disease, cirrhosis, and HCC using ultrasound images.
• This automated classification system has the potential to significantly aid clinicians in early and precise diagnosis.
• The findings underscore the value of advanced feature engineering and ensemble learning in medical image analysis for liver disease detection.